Introducing inducible fluorescent split cholesterol oxidase to mammalian cells

Cholesterol oxidase (COase) is a bacterial enzyme catalyzing the first step in the biodegradation of cholesterol. COase is an important biotechnological tool for clinical diagnostics and production of steroid drugs and insecticides. It is also used for tracking intracellular cholesterol; however, its utility is limited by the lack of an efficient temporal control of its activity. To overcome this we have developed a regulatable fragment complementation system for COase cloned from Chromobacterium sp. The enzyme was split into two moieties that were fused to FKBP (FK506-binding protein) and FRB (rapamycin-binding domain) pair and split GFP fragments. The addition of rapamycin reconstituted a fluorescent enzyme, termed split GFP-COase, the fluorescence level of which correlated with its oxidation activity. A rapid decrease of cellular cholesterol induced by intracellular expression of the split GFP-COase promoted the dissociation of a cholesterol biosensor D4H from the plasma membrane. The process was reversible as upon rapamycin removal, the split GFP-COase fluorescence was lost, and cellular cholesterol levels returned to normal. These data demonstrate that the split GFP-COase provides a novel tool to manipulate cholesterol in mammalian cells.

decreases membrane order because of cholesterol lowering and the disordering effect of cholestenone. Raft disruption leads to a spontaneous clustering of death factor Fas, formation of Fas-FADD complexes, activation of caspase-8, and apoptosis (16). An additional cytotoxic effect is caused by peroxide that increases the intracellular reactive oxygen species level (17). Despite that, a lysosome-targeted COase has been reported to attenuate the cytotoxicity of 7-ketocholesterol in human fibroblasts (18). Thus, COase activity influences multiple cellular processes.
Reconstitution of enzymatic activity through the noncovalent association of complementing fragments has been widely used to monitor dynamic protein-protein interactions (PPI) and to discover drugs modifying them (19). Development of split GFP and other fluorescent proteins enabled the visualization of PPI in a wide variety of organisms (20). Recently, a split horseradish peroxidase was developed to detect PPI in the extracellular space for studying communication between different cell types (21). Beyond that, complementing fragments are used in chemically induced dimerizers to restore enzymatic activity and trigger biological responses. In this case one fragment of an enzyme is fused with FRB (rapamycin-binding domain of kinase mammalian target of rapamycin) and the other with FKBP (FK506-binding protein), which form a pair upon the addition of rapamycin, reconstituting the functional enzyme (22,23). For example, split Cre and Cas9 nucleases are versatile tools for genome editing and transcriptional regulation (24)(25)(26). Split ubiquitin and tobacco etch virus protease are broadly used to regulate protein degradation and apoptosis (27,28). Additionally, splitting may increase spatiotemporal precision of enzymatic reactions and make them tissue-specific (24,29).
In this study we engineered a rapamycin-inducible fluorescent cholesterol oxidase from Chromobacterium sp. DS-1 (split GFP-COase) that belongs to class 2 of COase enzymes (13,30). Induction by rapamycin allows temporal control over COase activity and rapidly leads to an efficient reduction of intracellular cholesterol. After rapamycin removal, the cells loose the holoenzyme expression and recover their cholesterol levels. We propose this drug-inducible split GFP-COase as a new tool to manipulate the cellular cholesterol content and gain insights into cholesterol-dependent proteins and cholesterol-related cell pathologies.

Engineering of split GFP-COase enzyme
COase from Chromobacterium sp. strain DS-1 is composed of 21 ␣-helices and 20 ␤-strands that form four structural domains (13). We used two approaches to find successful sites for the split of the COase. Assuming that protein domains fold independently, we decided on three split sites that would separate entire domains and may complement in restoring enzymatic activity (Fig. 1A). Our second approach was to split COase in the loops with maximum flexibility. We chose several loop regions with maximal temperature B-factor that reflects their maximal mobility (Fig. 1A) (13). Presumably, splitting at these sites should preserve ␣-helices and ␤-sheets as well as bind to FAD and cholesterol.
We first tested the activity of split COase in bacteria, co-expressing both fragments from a single plasmid. To induce complementation of the separated N-and C-terminal fragments, they were fused to coils with oppositely charged amino acid residues (named E-coil and K-coil). These coils electrostatically interact with each other with high affinity (31) and bring COase fragments together, restoring their enzymatic activity. COase activity in bacterial extracts was detected for the constructs split at positions 75, 95, 415, and 485 ( Fig. 2A). Notably, all these splits located in close proximity to N-terminal or C-terminal ends of the protein. The most active split at the position 485 had a 20-fold lower enzymatic activity than the full-length enzyme. The remaining active constructs (splits at positions 75, 95, and 415) were even less active. This result was expected because splitting may decrease solubility and destabilize protein structure. Importantly, no enzymatic activity was detected when only one COase part was expressed in the cells (data not shown).
We further tested four splits with maximum activity (75, 95, 415, 485) in mammalian cells. For this purpose we fused the N-terminal COase part with FRB and the respective C-terminal part with membrane-targeted FKBP constructs (Fig. 2B). COase activity was only detected in splits 75 and 95, whereas splits 415 and 485 were inactive in HeLa cell lysates (Fig. 2C). COase activity in splits 75 and 95, however, was detected even in the absence of rapamycin (Fig. 2C). Detection of COase activity in living mammalian cells requires incubation of cells with dichlorofluorescin diacetate, which does not react directly with peroxide but with hydroxyl radicals generated by the Fenton reaction (32). Therefore, dichlorofluorescin diacetate-dependent fluorescence reflects the release of redox-active metals from lysosomes or cytochrome c from mitochondria rather than accumulation of peroxide. A better detection method would involve the co-transfection of cells with HyPer peroxidase biosensor (33), but this biosensor is very pH-sensitive, posing another drawback.
To avoid these obstacles, we decided to monitor COase activity by generating chimera of the COase fragments fused to splithalves of a fluorescent protein (FP). Consequently, fusion of the COase fragments would induce fluorescence and allow the detection of cells expressing the active COase by microscopy and cell sorting. Thus, we fused the COase parts with rapamycin-inducible split EGFP constructs (Fig. 2D). The addition of EGFP parts only slightly decreased the enzymatic activity of both 75 and 95 splits. More importantly, the fusion of the EGFP parts to split 95 abolished background activity in the absence of rapamycin (Fig. 2E). This effect might be explained by the lower amount of expressed protein and increased solubility of the chimeric constructs. In conclusion, the split 95 COase fragments (split-EGFP COase) generated the most promising combination to allow efficient control of COase enzymatic activity by rapamycin induction.
To investigate the kinetics of split-EGFP COase reconstitution in living HeLa cells, we induced cells continuously with rapamycin and analyzed them by flow cytometry. Unexpectedly, the EGFP signal was barely detected by flow cytometry and was too dim to be detected by fluorescence microscopy (Fig. 3, A and B). This is explained by the very low amount of reconstituted split-EGFP COase as shown in Fig. 3C.
To increase construct brightness, we replaced parts of EGFP with the corresponding fragments of super-folder GFP (sfGFP) (Fig. 3A), which has been shown to fold very efficiently with a high chromophore formation rate (34). However, a split sfGFP construct gave a strong fluorescent signal even in the absence of rapamycin due to spontaneous self-association of the complementing fragments (Fig. 3B). This self-association was eliminated by using a pair that contained the N-terminal fragment of sfGFP and C-terminal fragment of EGFP (Fig. 3, A and B). This construct, termed split GFP-COase, possessed the same brightness as sfGFP after 4 h of continuous induction but was practically undetectable in the absence of rapamycin (0 h) (Fig.  3, B and C). Both the protein amount of the complemented split GFP-COase chimera and its COase activity increased linearly after induction, reflecting a constant rate of protein synthesis (Fig. 4, A-C). Finally, we compared the enzymatic activity of split GFP-COase with full-length EGFP-COase. The specific activity (ratio of resorufin to GFP fluorescence) of split GFP-COase was about half that of the full-length enzyme (Fig. 4D).

Intracellular distribution and activity of split GFP-COase
We next studied the intracellular distribution of split COase in transfected HeLa cells after 2 and 4 h of rapamycin induction. Split GFP-COase was rather uniformly distributed over the cytoplasm and was not detected in the nucleus, in contrast to split EGFP control that was found in both cytoplasm and nucleus (Fig. 5, A and B). The absence of cytoplasmic aggregation suggested correct folding of the protein. Exclusion of split GFP-COase from the nucleus indicated that the chimeric protein exceeds the diffusion limit of nuclear pores.
To determine the amount of oxidized cholesterol, we stained rapamycin-induced cells with filipin. This fluorescent antibiotic detects both plasma membrane and endomembrane pools of cholesterol and allows their quantification in fixed cells (35). Upon the induction of split GFP-COase, filipin largely disappeared from the plasma membrane region within 2 h, whereas intracellular accumulation still remained intense (Fig. 5A). This remaining signal corresponded to ϳ40% of filipin intensity compared with control cells expressing split EGFP (Fig. 5C). Within 4 h of induction, filipin almost totally disappeared from cells with the highest level of split GFP-COase expression (Fig.  5B), with the average remaining signal being ϳ20% (Fig. 5C). Because the binding of filipin to residual cholesterol can be blocked by some membrane components (36), we confirmed this finding by biochemical cholesterol determination. We sorted cells expressing split GFP-COase using FACS and extracted lipids and analyzed them by thin layer chromatogra-phy (TLC). This verified that the cholesterol amount of the COase-expressing cells was about half that in split-EGFP transfected or non-transfected cells after 2 h of rapamycin induction ( Fig. 5D; supplemental Fig. S1). To assess if intra-endosomal/lysosomal cholesterol is protected from oxidation, we incubated split GFP-COase-expressing cells with U18666A, an inhibitor of the late endosomal cholesterol transporter NPC1 (37). This resulted in a filipin-staining pattern typical of storage lysosomes ( Fig. 5E) and elevated filipin intensity compared with control split GFP-COase-expressing cells (Fig. 5F). This suggests that the enzyme did not access intra-endosomal/lysosomal cholesterol.

Split GFP-COase promotes apoptosis
The accumulation of cholestenone is highly toxic to cells and organisms (38). In most cells lipids are distributed asymmetrically in the plasma membrane, with phosphatidylserine found in the inner leaflet of the plasma membrane (39). Upon induction of apoptosis, phosphatidylserine becomes translocated to the outer surface of the membrane and exposed to annexin V (40). To assess the effect of cholesterol oxidation on phosphatidylserine exposure, we stained split GFP-COase-expressing CHO cells with annexin V and propidium iodide (41). Cells expressing split GFP-COase showed a time-dependent increase in the fraction of annexin V-and propidium iodide-positive cells compared to control split EGFP cells (Fig. 6). These results suggest that COase expression changes membrane asymmetry

Split cholesterol oxidase
and enhances apoptosis over time. The precise mechanism(s) involved remains to be elucidated, but the rapid flip-flop and membrane-disordering effects of cholestenone (15) may contribute to phosphatidylserine translocation.

Recovery of cellular cholesterol after short induction of split GFP-COase
Analysis of split GFP-COase showed that it rapidly and efficiently reduces cellular cholesterol in a few hours after induction. The longer the COase is active, the more cells become apoptotic. To evaluate the potential recovery of cholesterol levels after a short oxidation by split GFP-COase, we pulsed transfected HeLa cells with rapamycin for 30 min, washed it out, and then continued cultivation in the absence of the drug. The intensity of the green signal of split GFP-COase increased rapidly and linearly reached its maximum at 2 h, followed by an exponential decrease. The calculated half-life time of split GFP-COase complex was 6 h (Fig. 7A). For control cells expressing split EGFP the mean cell brightness was ϳ5-fold higher than for split GFP-COase-expressing cells (Fig. 7B). In addition, the intensity of the green signal of split EGFP increased and decreased more slowly than of the split GFP-COase. The half-life time of split EGFP was 12 h, i.e. double that of split GFP-COase (Fig. 7B).
These findings suggested that the GFP-COase complex is less stable than the EGFP complex, and the formation of the GFP-COase complex might destabilize the enzyme. Consequently, this property might allow cells to restore their cholesterol level once split GFP-COase dissociates and/or is degraded in the cells. To investigate this hypothesis, we followed the cholesterol levels of HeLa cells after short rapamycin induction of the split GFP-COase by filipin staining. We observed a rapid decrease of filipin staining already at 2 h after the rapamycin pulse that correlated with the formation of the GFP-COase complex and the increase in green fluorescence (Fig. 7C). Filipin intensity was minimal between 2 and 6 h. After 6 h, the filipin signal started to gradually recover, reaching half of the initial level at 24 h and a full recovery at 48 h of chase. This corresponded to almost total disappearance of green fluorescence (Fig. 7, A-D). These data indicate that a short pulse of rapamycin causes rapid cholesterol oxidation within few hours, whereas replenishment of cholesterol levels is a slower process. Notably, the transfected cells remained viable until the end of the experiment and restored their cholesterol content (Fig. 7D).
To study the mechanisms by which the cells regain cholesterol, we performed the chase after short split GFP-COase induction in DMEM with 5% lipoprotein-deficient serum (LPDS). Under these conditions, cells rely on endogenous synthesis for their cholesterol supply. We found that in LPDS medium, cells replenished ϳ75% of the cholesterol based on filipin intensity (Fig. 7C). This implies that cells rely both on de novo cholesterol synthesis and lipoprotein cholesterol uptake in regaining cholesterol. We also noted that upon adding statin during the recovery from split GFP-COase expression, most of the cells actually died, implicating the necessity of cholesterol synthesis during recovery. Accordingly, the active cleaved form of the major transcriptional regulator sterol regulatory element-binding protein 2 (SREBP-2) was increased upon split GFP-COase expression (supplemental Figs. S2, A and B). The kinetics of SREBP-2 cleavage indicated that the maximal amount of the active form of SREBP-2 accumulated between 2 and 4 h of split GFP-COase induction (supplemental Fig. S2C).
To study whether the cleaved SREBP-2 activates transcription of genes involved in cholesterol metabolism, we analyzed the kinetics of mRNA expression of two representative SREBP-2 gene targets, i.e. low density lipoprotein receptor and HMG CoA reductase. We found that the transcripts were up-regulated upon induction of split GFP-COase and SREBP-2 activation (supplemental Fig. S2D), with low density lipoprotein receptor mRNA showing a more substantial effect. The modest up-regulation of HMG CoA reductase mRNA may be accompanied by post transcriptional regulation of the enzyme, contributing to increasing cholesterol synthesis.

Split GFP-COase induces translocation of a cholesterol biosensor
Because filipin staining cannot distinguish between cholesterol residing in the cytosolic versus exofacial/luminal leaflets of the plasma membrane, we analyzed the localization of mCherry-D4H cholesterol biosensor. This probe has been reported to recognize cholesterol in the cytosolic leaflet of the plasma membrane and to dissociate upon cholesterol extraction and perturbation of its trafficking (42). Moreover, D4H did not have affinity for cholestenone (supplemental Fig. S3). Consistent with Maekawa et al. (42), mCherry-D4H mainly localized at the plasma membrane of CHO cells, whereas in HeLa cells there were also some cytoplasmic mCherry-D4H aggregates (Fig. 8A). When split GFP-COase was induced, we observed translocation of mCherry-D4H from the plasma membrane to intracellular aggregates within 8 h in most cells (Fig. 8, A-C). After COase induction, the cells gradually changed their morphology and became more rounded. Dissociation of mCherry-D4H suggested that prolonged COase treatment effectively oxidizes cholesterol in the cytoplasmic leaflet of the plasma membrane. This in turn may result in the perturbation of actin cytoskeleton and altered cell morphology.

Discussion
Engineering of split GFP-COase included selection of active split enzyme variants in bacteria followed by their testing and further optimization in mammalian cells to make them inducible and easily detectable. To date, only Chromobacterium sp. DS-1 COase demonstrated ultimate stability in organic solvents, detergents, and at high temperatures (30). It has an isoelectric point of 8.6 and possesses high positive electrostatic surface potential that increases its affinity toward negatively charged membranes. Additionally, its activity in mammalian cells is reported (18), providing an additional justification for choosing this enzyme. The best splitting strategy generates separate fragments containing isolated domains that are correctly folded and functional by themselves, as in the classical two-hybrid system. COase consists of four structural domains (Fig.  1A); however, splitting between them failed to produce active fragments, probably because they were misfolded. Splitting in loops with maximal mobility resulted in several fragments that were active in bacterial extracts ( Fig. 2A). This strategy was also successful in generating split ubiquitin, GFP, and DHFR (43)(44)(45).
COase contains a large central water filled cavity and a channel that connects it with the surface (13). This channel serves as an entrance for substrate and molecular oxygen to the active site (14). The cavity consists of two oppositely charged areas bound to a cofactor and a substrate. Negatively charged phosphate groups of FAD bind to a positively charged part of the cavity, whereas a cholesterol molecule interacts with a polar part near the channel entry (Fig. 9). The FAD is deeply buried inside the protein structure, and movement in positions 75 and 95 allows cofactor entry into the cavity. Splitting at position 75 (or 95) removes the N-terminal part that covers the cavity together with His-63, which makes a covalent bond with the cofactor. In the remaining C-terminal fragment, FAD is totally exposed to water and cannot reside in COase, whereas a cholesterol-binding pocket remains intact (Fig. 9). Splitting at position 485 also opens a substrate-binding cavity, leaving the FAD

Split cholesterol oxidase
binding site intact; however, this construct was inactive in mammalian cells, likely due to its low stability.
The addition of rapamycin brings the COase fragments close together, resulting in the formation of the roof of the FADbinding cavity. In the reconstituted split COase, the FAD must be covalently bound because full-length H63A COase mutant is inactive (Fig. 2A). Typically, the activity of split enzymes is on the order of magnitude lower than full-length proteins. The activity of split GFP-COase is half of the full-length enzyme, which also indicates formation of stabilizing covalent bond with cofactor (Fig. 4D). The maximum of fluorescence signal for split GFP-COase was obtained at 2 h, that corresponded to the drop of filipin intensity. After that, the fluorescence of split GFP-Coase exponentially decreased with a half-time of 6 h, whereas the signal for split EGFP exhibited prolonged growth and twice slower decay (Fig. 7, A and B). This indicates lower stability of split GFP-COase compared with split EGFP. Similar induction and degradation kinetics were observed for split luciferase in live mice after single administration of rapamycin (46).
Perhaps even faster degradation of split GFP-COase could be achieved by fusing them with a destabilizing tag, named FRB* (47). In the FRB*-FKBP system, a half-life of the fusion protein was 2 h, which is 3-fold shorter than for split GFP-COase. The FRB*-tagged protein was continuously degraded, which resulted in a loss-of-function, whereas stabilization that occurred only upon the addition of C20-methallylrapamycin (C20-MaRap), an improved analogue of rapamycin. However, the complete stabilization of tagged protein was obtained after 24 h, the time much longer than observed for split GFP-COase activity.
Our system employs two separate vectors encoding the complementing parts. The enzymatic activity of a single fusion protein containing the FRB and FKBP pairs can be significantly higher than that of two separately expressed fragments (48). However, single fusion-based enzymes often demonstrate a high level of background activity in the absence of rapamycin. In our system the fragments of the final split GFP-COase  construct expressed separately show no enzymatic activity, indicating that the affinity of the fragments in the absence of rapamycin is negligible. Therefore, the complemented split GFP-COase provides an easily measurable fluorescent readout that correlates with COase activity.
Recombinant COase has been used to probe the cellular distribution of cholesterol (8). When added to cultured cells extracellularly, purified COase oxidizes only about half of the cholesterol in intact cells because it does not penetrate the cell membrane (8,17). We found a similar fraction of cholesterol oxidized in HeLa cells expressing split GFP-COase, suggesting that only cholesterol in the inner leaflet of the plasma membrane was attacked by intracellular COase. The split GFP-COase changes membrane organization and produces multiple effects on cell metabolism and architecture. Previous findings have revealed that both cholesterol depletion and peroxide generation are implicated in COase action (17). Cholesterol depletion results in down-regulation of the Akt/PIK3 kinase pathway, whereas elevation of reactive oxygen species up-regulates BAX and increases the amount of released cytochrome c (16,17,49). Notably, the oxidation product, cholestenone, produces long-lasting defects in actin dynamics and cell mobility (15). Moreover, cholesterol oxidation may indirectly affect cell shape, e.g. via proteolysis of cytoskeletal elements induced by apoptotic signaling (50).
Interestingly, expression of the lysosome-targeted COase alleviated oxysterol cytotoxicity (18). Oxysterols accumulate in lysosomes and cause membrane permeabilization and provoke cell death in a concentration-dependent manner. Overexpression of lysosome-targeted COase was found to decrease oxysterol-induced apoptosis, although the exact mechanism of protection is unknown (18). Our observations show that cytoplasmically expressed COase results in cellular cholesterol reduction and plasma membrane cholesterol biosensor translocation.
In summary, we have engineered the cytoplasmic rapamycin-inducible split GFP-COase and characterized it biochemically in vitro and functionally in mammalian cells. Short expression of split GFP-COase is reversible and well tolerated by mammalian cells. This new tool allows easy visualization of cells expressing the active chimeric COase using microscopy and flow cytometry and estimation of its enzymatic activity by quantification of the fluorescence signal. In comparison to other experimental methods of cholesterol depletion, split GFP-COase acts more acutely than statins. Moreover, the split GFP-COase acts exclusively from inside the cell, targeting the cytoplasmic leaflet, as compared with cyclodextrins that also act rapidly but target the exoplasmic leaflet. This is also the major difference between split GFP-COase and extracellularly applied COase. Overall, our findings suggest that intracellular COase is a potent cellular regulator and that the inducible split GFP-COase with improved temporal control of enzyme activity may provide new insights into cholesterol-dependent functions.

Design of plasmids
pC4-RHE and pC4M-F2E were obtained from Ariad Pharmaceuticals. Split EGFP plasmids were obtained from S. Gambhir (48). sfGFP fragments were amplified from sfGFP plasmids provided by E. Snapp and by G. Waldo (54). Desired DNA sequences were amplified with primers (Biomers), digested with endonucleases (New England BioLabs) and ligated. Constructs were verified by sequencing. The designed plasmids are summarized in Table 1.

Cholesterol oxidase assay in bacterial cells
Both N-and C-terminal parts of cholesterol oxidase were co-expressed in Escherichia coli host BL21(DE3) upon overnight induction with 0.5 mM isopropyl 1-thio-␤-D-galactopyranoside. Cells from 20 ml of culture were collected and disrupted in 1 ml of 1ϫ PBS buffer, and 50 l of clarified supernatant was added to 1 ml of reaction buffer (50 mM sodium potassium phosphate, 64 mM sodium cholate, 0.34% Triton X-100, 1.4 mM 4-aminoantipyrine, 21 mM phenol, 0.89 mM cholesterol, and 5 units/ml horseradish peroxidase). Development of red color in the assay mixture was tracked by monitoring the absorbance at 500 nm at 24°C for 30 min.

Cholesterol oxidase assay in HeLa cell lysate
HeLa cells grown in DMEM complete medium supplemented with 1 units/ml of penicillin, 100 g/ml streptomycin, and 10% FBS were transfected with split constructs by the calcium phosphate method (51) for 24 h and induced with 5 nM rapamycin for the indicated times. Cells were washed 3 times with PBS, cholesterol oxidase activity was assayed by the addition of reaction buffer (1ϫ PBS, 0.5% Triton X-100, 0.1 mM cholesterol, 20 M Amplex Red (fresh batch), 20 units/ml catalase, 1 units/ml horseradish peroxidase), and cells were incubated for 30 min at 37°C. The addition of catalase to the reaction buffer significantly reduced background signal. Fluorescence of oxidation product resorufin (excitation/emission maxima are at 570/585 nm, respectively) was measured on Victor X3 multilabel plate reader (PerkinElmer Life Sciences). To compare the activity of full-length EGFP-COase with split GFP-COase, resorufin fluorescence was divided by the total amount of green signal.

Filipin staining
HeLa cells grown in complete medium were seeded on glass coverslips and transfected with split GFP-COase or split EGFP constructs by calcium phosphate for 24 h (51) and induced with 5 nM rapamycin for the indicated times. The cells were fixed for 15 min with 4% paraformaldehyde in PBS, quenched with 50 mM NH 4 Cl in PBS for 10 min at room temperature, and stained with filipin (500 g/ml, Sigma) in PBS supplemented with 10% FBS for 30 min at 37°C. After washing with PBS, the cells were mounted with Mowiol 4 -88/Dabco (100 mg/ml), both from Sigma.

Flow cytometry
Flow cytometry analysis of rapamycin-induced split protein complementation was performed using Accuri C6 flow cytometer (BD Biosciences) equipped with a 488-nm laser and a 533/ 30-nm emission filter. Green cells were gated using a negative control (0.9% of live cells in the gate). Mean fluorescence intensity multiplied by a number of green cells gave an integrated signal in the gate. Each time point was made in triplicate, and minimum of 10 4 cells were analyzed per sample. The data were analyzed using a CFlow v. 1.0.264.15 and OriginPro v. 8.6 software.

Fluorescence microscopy and quantification of fluorescence intensity
Filipin and GFP were imaged using a Nikon Eclipse Ti-E inverted microscope equipped with a Nikon Intensilight metal arc light source and a 40 ϫ 0.75 NA air objective lens (Plan-Fluor, Nikon). mCherry-D4H imaging was performed using an Olympus IX83 inverted epifluorescence microscope equipped with a 200-watt metal halide arc lamp (Lumen 220PRO, Prior) and a 100 ϫ 1.35 NA oil immersion objective lens (UPlanSApo, Olympus). Filipin and GFP intensities were quantified using FiJi ImageJ software. Regions of interest were drawn manually, and mean fluorescence intensities were measured. Average background mean intensities from two regions outside the cells were subtracted from the values. Samples from at least two independent experiments with 60 -120 transfected cells per time point were quantified. All the filipin intensities were normalized to the value of the non-induced sample (defined as 100%).

Liposome preparation
Liposomes with various lipid compositions were prepared by mixing the required amount of lipid stocks and drying them under nitrogen gas and hydrating in the assay buffer (25 mM Hepes-KOH, 100 mM NaCl, 0.5 mM EDTA, pH 7.3) followed by ultrasound sonication.

Quantitative RT-PCR
Total RNA from HEK293 cells was extracted with the Nucleospin RNA extraction kit (Macherey-Nagel), and cDNA was synthesized from total RNA using ProtoScript II first strand cDNA synthesis kit (New England BioLabs). PCR reactions were performed using cDNA, LightCycler 480 SYBR Green I master mix (Roche Applied Science), and the corresponding primers (supplemental Table 1) in a LightCycler 480 system (Roche Applied Science). Samples were heated for 5 min at 95°C and amplified in 45 cycles of 10 s at 95°C, 15 s at 64°C, and 20 s at 72°C. Analyses were conducted using LightCycler 480 software, and the threshold cycle (CT) was extracted from the PCR amplification plot. The ⌬CT values were calculated to evaluate the difference between the CT of a target gene and the CT of the housekeeping gene SP-1, as follows: ⌬CT ϭ ⌬CT (target gene) Ϫ ⌬CT (SP-1). mRNA expression level for each data point was normalized to the corresponding level in split EGFP expressing cells. All samples were analyzed in quadruplicate.

Liposome sedimentation assay
GST-D4H protein (0.2 mg/ml) was incubated with 1 mM multilamellar vesicles for 20 min at 24°C, and the mixture was centrifuged at 100,000 ϫ g for 30 min at 20°C. Supernatants

mCherry-D4H distribution
HeLa and CHO cells were seeded on glass coverslips and incubated with calcium phosphate-DNA complexes. The mass ratio of split EGFP or split GFP-COase constructs per mCherry-D4H construct with was 10:1. After 4 h of incubation, CHO cells were briefly treated with 1ϫ PBS supplemented with 20% glycerol, media from both HeLa and CHO were replaced, and cells were allowed to grow for 16 h. Cells were induced with 5 nM rapamycin continuously, fixed with 4% paraformaldehyde, and mounted with UltraCruz mounting medium (Santa Cruz Biotechnology).

Biochemical lipid analysis
Lipids were extracted according to Bligh and Dyer (53) and separated by high performance thin layer chromatography using hexane/diethyl ether/acetic acid (80:20:10) as the running solvent. Lipids were visualized by dipping the TLC plate in 3% CuSO 4 , 8% H 2 PO 4 and charring. Lipids were identified based on standards, and the signal intensity was quantified using ImageJ software.

Annexin V analysis
CHO cells were transfected with Turbofect (Thermo Scientific) for 24 h on 12-well plates. Expression of chimeric con-struct was induced with 5 nM rapamycin. Cells were collected, resuspended in 200 l of binding buffer (10 mM Hepes-KOH, 150 mM NaCl, 1 mM MgCl 2 , 1.8 mM CaCl 2 ,10 g/ml propidium iodide, pH 7.3), and incubated with 1 l of annexin V-Cy5 (Abcam) for 5 min. A minimum of 10 4 cells were analyzed by flow cytometry in red (excitation 488 nm, emission 585/40 nm) and far red (excitation 640 nm, emission 675/25 nm) channels. All experiments were performed in triplicate.
Author contributions-K. G. C., M. N., and I. B. acquired the data. K. G. C., E. I., and V. V. V. analyzed and interpreted the data. K. G. C., E. I., and V. V. V. wrote the manuscript. All authors reviewed the manuscript.