A cell-based scrambling assay reveals the phospholipid headgroup preference of TMEM16F on the plasma membrane

Significance

Phospholipid scramblases on the mammalian plasma membrane are thought to act promiscuously without preference for headgroup. Thoroughly addressing this question, however, requires the development of new methodologies. We devised a cell-based phospholipid scrambling assay that utilizes the fluorescence polarization (FP) of nitrobenzoxadiazole (NBD)-labeled phospholipids, allowing the monitoring of their scrambling in a native environment. We found that the plasma membrane-residing calcium-activated phospholipid scramblase TMEM16F preferentially acts on phosphatidylserine and phosphatidylcholine over phosphatidylethanolamine.

Abstract

The asymmetric resting distribution of the three major phospholipid classes on the mammalian plasma membrane, with phosphatidylserine and phosphatidylethanolamine mostly on the inner leaflet and phosphatidylcholine mostly on the outer leaflet, is maintained by ATP-dependent flippases and floppases that exhibit headgroup selectivity. Upon signaling cues, this asymmetry can be dissipated by various phospholipid scramblases, allowing cells to respond to stimuli and adapt to different physiological contexts. The prevailing view in the field is that phospholipid scramblases on the plasma membrane act without headgroup preference. Here, we report contrary experimental evidence based on a phospholipid scrambling assay that quantifies the fluorescence polarization (FP) of nitrobenzoxadiazole (NBD)-labeled phospholipids for kinetic monitoring of phospholipid scrambling on the plasma membrane of living cells. Our experiments reveal that the plasma membrane-residing calcium-activated phospholipid scramblase TMEM16F preferentially acts on phosphatidylserine and phosphatidylcholine over phosphatidylethanolamine.

The asymmetric distribution of phospholipids (PLs) on the plasma membrane is maintained by several classes of ATP-dependent flippases and floppases that exhibit PL headgroup selectivity (1). The resting distribution of three major PL classes on the plasma membrane, with phosphatidylserine (PS) and phosphatidylethanolamine (PE) mostly on the inner leaflet and phosphatidylcholine (PC) mostly on the outer leaflet, is critical in sustaining mammalian cellular physiology (2–4). Of equal importance is the regulated reduction of this asymmetry by different classes of ATP-independent PL scramblases (4–9), including those in the apoptosis-activated Xkr family (10, 11), the calcium-responsive TMEM16 family (12, 13), as well as the mechanosensitive TMC (14, 15) and TMEM63 families (16). These PL scramblases mediate crucial physiological events such as exposing an “eat-me” signal for apoptotic clearance or fine-tuning plasma membrane fluidity, curvature, and tension to enable various plasma membrane processes (2–5, 7–9). Unlike flippases and floppases, which utilize ATP to maintain the asymmetric distribution of PLs in the lipid bilayer, scramblases harness the potential energy stored in the asymmetry itself, driving toward PL equalization on the two leaflets along the concentration gradient (2, 3, 8). Ubiquitously expressed and plasma membrane-localized Transmembrane Protein Family 16 Member F (TMEM16F), also known as anoctamin-6 (ANO6), is one of the best-characterized calcium-activated PL scramblases (4–6, 8, 9). Originally identified as a key determinant in the PS exposure and extracellular vesicle release from platelets for blood coagulation (13) and the genetic contributor to the rare hereditary bleeding disorder Scott Syndrome (12), TMEM16F with its dual functions of ion channel and PL scramblase also mediates membrane wound repair (17), cell signaling and immune responses (18, 19), as well as cell-cell (20, 21) and pathogen-cell fusion (17, 22–24).

Probing the movements of plasma membrane PLs is facilitated by exogenous application of their labeled analogs, which are readily taken up by living cells and incorporated into the plasma membrane with asymmetric distributions resembling those of their endogenous native counterparts (25–32). Fluorescent nitrobenzoxadiazole (NBD)-labeled PLs have been widely used for the determination of the lipid bilayer PL distributions (28). By applying this approach to red blood cells loaded with PLs labeled with NBD or nitroxide spin probe (for electron spin resonance measurements) and extraction of the labeled PLs from the outer leaflet, Williamson and colleagues reported comparable transmembrane movements for PC, PS, and PE stimulated by increases in cytosolic calcium concentration (30). While the approaches involving bovine serum albumin (BSA) back exchange and the related NBD-fluorescence quenching by BSA have been widely adopted for delineating the endpoint bilayer distribution of PLs, they are not ideal for monitoring PL scrambling because the time frame required for quantitatively extracting BSA-bound NBD-PLs is longer than that of the PL scrambling. In a 1991 landmark paper, McIntyre and Sleight (29) introduced the first fully kinetic scrambling assay using a reducing agent, dithionite, to chemically quench the fluorescence of the outer leaflet residing NBD-PLs. While dithionite quenching has been used widely in proteoliposome systems, its utility in living cells is significantly hampered by interference from plasma membrane transporters such as the Band 3 protein that actively imports dithionite from the extracellular milieu, thereby introducing a quenching signal unrelated to PL scrambling (33). Thus far, all reported NBD-PL-based assays utilize its fluorescence intensity and changes thereof to determine scrambling activities (8, 34). These approaches have collectively cemented the prevailing canon that PL scramblases do not display headgroup selectivity (3, 4, 8, 30, 31, 35–40). Given the limitations of the current repertoire of methods, we devised an assay involving quantitative assessment of fluorescence polarization (FP) to monitor PL scramblase activity in living cells.

FP is a detection modality that is especially suitable for probing rotational mobility of biomolecules and has been widely applied for ligand binding and kinase assays (41). In FP experiments (Fig. 1A), polarized photons are preferentially absorbed by, and excite, fluorophores whose excitation dipole aligns with the polarization of the photons, leading to a population of excited fluorophores with high orientational coherence. If the fluorophores rotate slowly and emit light before significant randomization of their orientations takes place, the emitted photons retain the polarization of the excitation light beam (vertical as shown in Fig. 1A). In contrast, fluorophores with high rotational motility will have their orientations randomized at the time of emission, leading to reduced polarization (depolarization) of the emission. The degree of polarization of the emitted photons can be calculated from the relative intensities (Materials and Methods) measured using two independent and perpendicularly oriented polarization filters. The rotational correlation times of NBD-PLs are comparable to their mean fluorescence lifetimes (both in the nanosecond regime), rendering FP values a sensitive probe for changes in rotational mobility (42–44). Thus, FP measurements provide a clear advantage compared to fluorescence intensity assays in many experimental setups designed to elucidate molecular interactions due to their diagnostic dependence on the rotational mobility, as well as their independence from the fluorescence intensity, which may change for a variety of reasons, including self-quenching (45).

Fig. 1.

A fluorescence polarization (FP) assay for phospholipid movement across bilayers. (A) Simple schematic describing the FP principles. In the initial unpolarized light beam (blue), all directions of the individual photon polarizations (black two-headed arrows) are represented. An excitation polarization filter that preferentially absorbs photons with horizontal polarization lets through a vertically polarized excitation beam. Excitation photons passing the sample preferentially excite the (possibly mobile) fluorophores that happen to have their excitation dipole vertically oriented, leading to a population of excited fluorophores that have a high orientational coherence. Fluorophores with high rotational mobility (Top) have their orientations randomized before emission, and the emitted photons (green) are highly depolarized. Fluorophores with low rotational mobility (Bottom), e.g., due to Alb binding, retain their orientations, and the emitted photons retain the high vertical polarization. The emitted photons are passed through independent vertical and horizontal emission polarization filters that preferentially absorb photons with horizontal and vertical polarizations, respectively, and the intensities are measured. (B) Interpretation of FP data in a hypothetical experiment using nonapoptotic and apoptotic cells as a model system, where the bilayer distributions, and changes thereof, of PS and PE are well known. In nonapoptotic conditions, PS and PE are nearly completely on the inner leaflet of the plasma membrane, inaccessible to Alb on the extracellular side. The rotational mobility of the NBD-PLs is high, and the measured FP value is low. In apoptotic conditions, some NBD-PS and NBD-PE are exposed on the outer leaflet, hallmarks of apoptosis, where they are accessible for Alb binding, and where their rotational mobility is hindered, resulting in a higher net FP value. (C) A proof-of-principle FP experiment using nonapoptotic and apoptotic cells. Normalized Alb-FP values measured immediately after Alb addition (normalized to the initial FP at time = 0 s after adding raptinal or vehicle control) from nonapoptotic and apoptotic HeLa cells preloaded with either NBD-PS or NBD-PE. Datapoints from independent replicate experiments (N = 3, light circles) as well as their averages (dark circles) ± 1 SD (shaded box) are indicated. P-values from +Raptinal (apoptotic) vs. -Raptinal (nonapoptotic control) comparison (two-tailed Welch’s test): NBD-PS: 0.001, NBD-PE: 0.019. The normalization is based on the first collected datapoint of a kinetic recording. The increased normalized FP signal in apoptotic compared to nonapoptotic cells supports the postulates described in A and B, indicating that a functional scrambling assay could be designed based on FP. * statistically significant with P-value ≤ 0.05, ** statistically significant with P-value ≤ 0.01.

Here, we report a cell-based microplate PL scrambling assay whereby FP, rather than fluorescence intensity of NBD-PLs, is monitored, enabling investigations into the headgroup preference of PL scramblases on living cells. By applying this assay on cells with or without endogenous TMEM16F, we deciphered its headgroup preference on the plasma membrane. Coarse-grained molecular dynamics (CGMD) simulations support these findings.

Results

A FP Assay to Distinguish the Phospholipid Bilayer Distribution On the Plasma Membrane.

We hypothesized that the physical binding of albumin (Alb) to NBD-PLs on the outer leaflet of the plasma membrane (46), a prerequisite preceding the extraction and quenching utilized in existing fluorescence intensity-based assays, would retard their rotational mobility and hence, according to well-established physical principles, increase the FP signal (Fig. 1A). The degree of increase in the FP signal upon Alb binding would signify the abundance of the NBD on the outer leaflet. If true, this scenario with the immediate binding of Alb to outer leaflet residing NBD-PLs, termed Alb-associated FP or Alb-FP, would be an enticing portal for delineating the dynamics of PLs on biological membranes (Fig. 1B).

As a litmus test for our hypothesis, we compared the Alb-FP values of NBD-PS and NBD-PE in control and apoptotic HeLa cells, as increases in the extracellularly exposed PS (47) and PE (48) are well-established apoptotic signatures. To elicit the extracellular exposure of PS and PE that are normally located on the inner leaflet of the plasma membrane, we treated cells preloaded with NBD-PS or NBD-PE either with an apoptosis-inducer raptinal (49) or DMSO vehicle for 30 min (the minimum time for raptinal-treated cells to undergo apoptosis). After adding Alb, we immediately detected significantly higher normalized FP values (normalized to a FP reading taken before the inducement of apoptosis) in the apoptotic cells compared to the nonapoptotic cells (Fig. 1C, red and black dots, respectively). Thus, we confirmed that Alb-FP readings can be used to quantitatively uncouple the NBD-PL signals in the two leaflets.

Establishing Kinetic FP Recording for a Calcium-Stimulated Scrambling Activity.

Encouraged by these results, we proceeded with a kinetic FP scrambling assay composed of three stages (Fig. 2A). After a baseline FP reading (blue-shaded area, typically 1 min), calcium-dependent scrambling was triggered by treating cells with a calcium ionophore, ionomycin, a known pharmacological inducer for TMEM16F scrambling activity (brown-shaded area). For each NBD labeled PL, we compare the Alb-FP signals of cells with endogenous TMEM16F with cells of the same cell line with a genetic knockout of TMEM16F (16FKO). If TMEM16F displays PL headgroup selectivity, we expect to find significant differences in the Alb-FP signal for one type of PL with NBD label but not for another type of PL with NBD label, between cells with TMEM16F and cells without TMEM16F.

Fig. 2.

A kinetic scrambling assay. (A) Cartoon depiction of a three-stage time-course FP experiment using cells preloaded with NBD-PL: 1) baseline stage (blue shading), where baseline FP reading is established, 2) scrambling stage (yellow shading), where, if scrambling is induced, the asymmetry of PL on the two leaflets is diminished, and 3) binding stage (green shading), where the binder (Alb) interacts specifically with the extracellularly accessible NBD. In nonscrambling conditions, NBD-PS and NBD-PE are nearly completely located on the inner leaflet of the plasma membrane with high rotational mobility and the measured FP value is low. Subsequently, adding Alb to the extracellular milieu has little or no effect to the net FP value. In scrambling conditions, there is a net relocation of NBD-PS and NBD-PE to the outer leaflet, where they are accessible to interaction with Alb. The Alb-bound NBD-PLs have lower rotational mobility, leading to higher net FP values. In nonscrambling conditions, NBD-PC is mostly located on the outer leaflet, leading to high accessibility for Alb binding and eventually high polarization values. In scrambling conditions, some of the NBD-PC is relocated to the inner leaflet, leading to lower accessibility for Alb binding, and lower net FP values. (B) Time-course kinetic traces for FP scrambling assay from HeLa cells preloaded with NBD-PS (N = 5), NBD-PE (N = 5), or NBD-PC (N = 5). The indicative of ionomycin-induced PL scrambling can be inferred by their corresponding Alb-FP traces (green shading, with gray outline). NBD-PS and NBD-PE have lower FP values in nonscrambling (control, black traces) conditions compared to the scrambling conditions (red traces), indicating net transfer of these NBD-PLs to the outer leaflet. Conversely, NBD-PC has a lower FP value in scrambling conditions compared to the nonscrambling (control) conditions; because in resting conditions there is more NBD-PC on the outer leaflet and upon scrambling, there will be a net transfer to the inner leaflet. The time-course dynamics of the FP have contributions not only from the continuous NBD-PL scrambling and the Alb binding but also some cellular processes including flipping, the latter of which are also present in the control experiment. After a brief period of increasing FP value, a plateau is reached, indicating saturated binding by Alb. The kinetic recording was paused for manual addition of ionomycin as indicated with the white gap between the baseline (blue shading) and scrambling (yellow shading) stages.

To maximize the difference in the Alb-FP signal from the scrambling cells and control cells loaded with NBD-PS, we allowed ionomycin to act for 2 min before Alb addition. Given the tendency of NBD-PE to be metabolized (50), which is suggested to be triggered by the addition of ionomycin (51), we shortened the NBD-PE scrambling duration to 30 s before adding Alb (SI Appendix, Fig. S1). Since PC predominantly resides on the outer leaflet in resting cells, adding Alb too early would sequester too much of the outer leaflet-residing NBD-PC and reduce the difference in the FP signal between the scrambling and nonscrambling cells. Thus, we extended the scrambling duration of NBD-PC to 4-9 min to allow for a greater accrual from inward scrambling before we added Alb. Next, Alb is added (green-shaded area) to selectively reduce the rotational mobility of outer leaflet residing NBD-PLs and prevent them from subsequently scrambling or flipping to the inner leaflet. The FP time-course readings from control and scrambling conditions were normalized to the mean value of their baseline period. Since FP values in the absence of Alb do not provide the necessary information to decipher the bilayer distribution of NBD-PL, the interpretation of scrambling events focuses on the Alb-FP readings (Fig. 2B, green-shaded area), plotted as a set to infer the direction of the NBD-PL transbilayer movement.

When we compared the kinetic Alb-FP measurements in the ionomycin-induced scrambling condition and the vehicle-treated control condition in HeLa cells preloaded with NBD-PS, NBD-PE, or NBD-PC, they displayed a change in Alb-FP that matched the predicted models based on their known plasma membrane bilayer distribution (Fig. 2B). NBD-PS and NBD-PE-preloaded cells displayed higher Alb-FP values in scrambling conditions as compared to the nonscrambling control, reflecting a net transfer of NBD-PS and NBD-PE from the inner to the outer leaflet. Conversely, NBD-PC-preloaded cells displayed a decrease of Alb-FP values in scrambling conditions as compared to the nonscrambling control, owing to a net transfer of NBD-PC from the outer to the inner leaflet. We confirmed the ability of FP measurements to discern the directionality of PL movement using two additional cell models, U2OS and A549, where ionomycin also triggered a net outward movement of NBD-PS and a net inward movement of NBD-PC (SI Appendix, Fig. S2). These data demonstrated that a kinetic FP assay can be used to infer the directionality of NBD-PLs in scrambling conditions. Notably, we detected a decrease in the FP reading in the scrambling state (before Alb addition) in the NBD-PC-loaded and ionomycin-treated cells (Fig. 2B), reflecting an increase in membrane fluidity during plasma membrane PL scrambling, consistent with a previous report (52).

The Headgroup Preference of TMEM16F.

After establishing the conditions for the FP scrambling assay, we set forth to address our central question: Does TMEM16F possess headgroup preference in vivo? We reasoned that it is possible to deduce whether TMEM16F possesses headgroup preference by examining the scrambling of different PLs in cells with or without endogenous TMEM16F expression. The degree of reduction in the scrambling for a given PL in 16FKO cells compared to WT cells indicates the importance of TMEM16F in the calcium-activated scrambling of that PL. If TMEM16F does not exhibit any headgroup preference for PL scrambling, we would see a universal reduction in transbilayer relocation of all NBD-PLs in 16FKO cells. We started by comparing PS scrambling in NBD-PS-preloaded WT and 16FKO HeLa cells. Ionomycin-triggered TMEM16F-dependent PS scrambling is a well-characterized phenomenon detected by annexin conjugates that are canonical PS-specific probes (12, 13, 53). Unlike the significant increase in Alb-FP values upon ionomycin stimulation of NBD-PS-preloaded WT cells, we detected no significant difference in the Alb-FP values of NBD-PS-preloaded 16FKO cells with or without ionomycin treatment, indicating minimal NBD-PS outward movement in the absence of TMEM16F expression (Fig. 3A and SI Appendix, Fig. S3). Next, we tested the scrambling of NBD-PC and NBD-PE using a similar experimental setup, while adjusting the duration of the scrambling period to account for the differing bilayer distributions and stabilities of these lipids. The Alb-FP values from NBD-PC-loaded cells in scrambling and nonscrambling conditions were very different in WT cells compared to 16FKO cells (Fig. 3B), indicating a dominant role for TMEM16F in calcium-activated plasma membrane PC scrambling. The Alb-FP values from NBD-PE-loaded cells in scrambling and nonscrambling conditions in WT cells were similar to those in 16FKO cells (Fig. 3C), indicating that removal of TMEM16F does not significantly affect calcium-activated PE scrambling in HeLa cells.

Fig. 3.

Scrambling index for evaluating TMEM16F scrambling specificity. Alb-FP traces for WT and TMEM16F HeLa cells were recorded using cells preloaded with (A) NBD-PS (N = 7), (B) NBD-PC (N = 7), and (C) NBD-PE (N = 2) using in both scrambling (red trace) and nonscrambling (black trace) conditions. Scrambling index (Right panels) is calculated for a given combination of NBD-PL and cellular genotype by subtracting the Alb-FP trace from nonscrambling condition from that of the scrambling condition. The first 60 s of the scrambling index trace was fitted to a linear function (blow-up), the slope of which yields the scrambling efficiency for the PL. (D) NBD-PS, WT: 0.0515, NBD-PS, 16FKO: 0.0145, 3.54-fold difference, two-tailed Welch’s test P-value 0.0044; (E) NBD-PC, WT: 0.0145, NBD-PC, 16FKO: -0.0289, 5.10-fold difference, P-value 1.64e−5; (F) NBD-PE, WT: 0.0487, NBD-PE, 16FKO: 0.0321, 1.51-fold difference, P-value 0.384. The kinetic recording was paused for manual addition of Alb in (C) as indicated with the white gap before the binding stage (green shading).

We calculated a “scrambling index” for each PL to obtain a quasi-quantitative view of scrambling effectiveness for comparison between WT and 16FKO cells. The scrambling index for a PL in a given cell type is calculated by subtracting the Alb-FP measurement in nonscrambling conditions from the Alb-FP measurement in scrambling conditions, and fitting the first 60 s, the time frame before significant extraction of NBD-PLs from the outer leaflet of the plasma membrane has taken place (34, 54), to a linear model (Materials and Methods). The ratio in the slopes of the fitted lines indicates the scrambling efficiency for the PL (Fig. 3 D–F and SI Appendix, Fig. S3). For both PS and PC, the scrambling indices of 16FKO cells were significantly different from those of WT cells (Fig. 3D, 3.54-fold difference for NBD-PS, P-value 0.0044; Fig. 3E, 5.10-fold difference for NBD-PC, P-value 1.64e−05), indicating a drastic reduction in the calcium-activated PS and PC scrambling at the plasma membranes of cells devoid of TMEM16F expression. In contrast, the scrambling index for PE of WT and 16FKO cells was comparable (Fig. 3F, 1.52-fold difference for NBD-PE, P-value 0.384), indicating that significant calcium-activated PE scrambling activity at the plasma membrane is retained in the absence of TMEM16F. These data further affirm our findings that TMEM16F displays headgroup preference for PS and PC over PE in vivo.

To investigate whether the trends observed in our FP assay are also present at the molecular length-scale on the multimicrosecond time frame, we employed CGMD simulations, which have been successfully utilized for evaluating PL scrambling by TMEM16 proteins (55–57). We embedded an open conformation of calcium-bound TMEM16F derived from extensive all-atom simulations (57–59) in a symmetric bilayer composed of equimolar DOPS, DOPC, and DOPE. We hypothesized that if TMEM16F exhibits headgroup selectivity, this would be manifest in a scenario where all three PLs compete for TMEM16F grooves, even in the absence of PL gradients. We carried out three independent 10 µs simulations and measured scrambling by tracking the orientation of individual lipids with respect to the membrane normal to determine when a lipid moved from one leaflet to the other. In agreement with our FP assays, TMEM16F exhibits a significant preference for PS and PC over PE with average rates of about 9, 7, and 3 events per 9 µs, respectively (Fig. 4A). The averaged density maps (Fig. 4B) and radial distribution profiles (Fig. 4C) reveal that DOPE is depleted in the vicinity of the protein, while DOPS is the enriched and DOPC adopts an intermediate density between the other two. Interestingly, while DOPE scrambled 2-3 times less than the other lipids, the duration of each scrambling event, defined as the time spent with lipid tilt angle between 35° and 145°, was much faster on average with a value of 602 ± 65 ns, compared to 929 ± 126 ns and 1,038 ± 161 ns for DOPC and DOPS, respectively.

Fig. 4.

Coarse-grained molecular dynamics (CGMD) simulation of TMEM16F embedded in an equimolar and symmetric bilayer of DOPS, DOPC, and DOPE phospholipids. (A) The total number of scrambling events for DOPS, DOPC, and DOPE in the last 9 μs of the simulation. Datapoints from independent replicate experiments (N = 3, light circles) as well as their averages (dark circles) ± 1 SD (shaded box) are indicated. P-values from two-tailed Welch’s tests: DOPS vs. DOPC: 0.321 (n.s.), DOPS vs. DOPE: 0.038 (*), DOPC vs. DOPE: 0.044 (*). n.s., not significant; * statistically significant with P-value ≤ 0.05. (B) Lipid density maps for DOPS, DOPC, and DOPE in the vicinity of TMEM16F (N = 3, 1 to 10 μs). (C) Radial lipid concentrations as a function of distance from the TM4-TM6 groove (residue I591, N = 3, 1 to 10 μs). Average enrichment for each PL (distance ≤ 45 Å): DOPS = 1.21 ± 0.02; DOPC = 1.01 ± 0.00; DOPE = 0.78 ± 0.02. The gray line indicates 33.3%.

Discussion

The question concerning headgroup preference of PL scramblases is relevant to the burgeoning field of lipid scrambling and membrane biology in general. While there is a consensus in the published literature that phospholipid scramblases do not exhibit headgroup preference (3, 4, 8, 30, 31, 35–40), we wanted to critically revisit this question given the limitations in currently available detection methods. Thus, we designed an approach based on FP that is applicable to living cells. In contrast to the prevailing views, we found that TMEM16F displays headgroup preference for PS and PC over PE, a finding supported by CGMD simulations.

Our scrambling assay employs the same tail-derivatized NBD-PLs that have been utilized in prior PL scrambling assays, which rely on the selective elimination of the NBD-PL fluorescence located on the outer layer of the plasma membrane, by either dithionite reduction of NBD to its nonfluorescent derivative (29), or physical removal of NBD-PL by BSA (34). The dithionite quenching has the advantage of enabling real-time recording, yet it suffers from a high degree of false-positive signal from dithionite internalization, where it can act on the inner leaflet-residing NBP-PLs as well (29, 60). While the BSA backexchange (also called quenching) is specific for the NBD-PLs on the extracellular leaflet, the kinetics of NBD-PL removal are much slower than the scrambling and is often coupled with a physical isolation step for accurate measurements at a single time point (34), rendering this approach more suitable for probing steady-state PL bilayer distributions.

We opted for detection of FP to capture the rotational mobility of the NBD-PLs rather than probing for their fluorescence intensity, and we applied Alb to the extracellular media to specifically reduce the mobility of the outer leaflet-residing NBD-PL. Hence, with FP measurement, we attain temporal resolution comparable to the dithionite-based method, while also achieving the spatial specificity that can be afforded by BSA backexchange/quenching. Using a kinetic recording mode on a plate reader, we could capture the change in Alb-FP signals correlated with directional movements of NBD-PLs as a readout of PL scrambling. We adjusted the durations of the scrambling period in our measurements for different PLs to accommodate their different resting bilayer distributions and stabilities on the plasma membrane. Aiming to maximize the difference in the FP-Alb signals from the nonscrambling and scrambling conditions, we tuned the scrambling duration for each PL to allow a significant amount of the inner leaflet-residing NBD-PS and NBD-PE to accumulate on the outer leaflet and outer leaflet-residing NBD-PC to accumulate on the inner leaflet during this period. Due to the faster internalization and metabolism of NBD-PE (50) compared to NBD-PS, we shortened its scrambling duration. The FP value for NBD-PC-loaded cells exhibited a gradual decrease in the scrambling compared to the nonscrambling condition even before the addition of Alb, a phenomenon that is not seen for NBD-PS and NBD-PE scrambling. Reduction in lipid-bound FP values of some environmental sensing fluorophores is known to correlate with an increase in membrane fluidity (61, 62). In a recent study, Wang and colleagues demonstrated that TMEM16F-mediated scrambling influences plasma membrane lipid packing with a concomitant increase in the plasma membrane fluidity (52). In the absence of scrambling, NBD-PC is mostly located in the outer leaflet of the plasma membrane, which is more densely packed (63) than the inner leaflet where NBD-PS and NBD-PE are enriched. It is possible that the dense packing makes outer leaflet-residing PLs more sensitive probes for membrane fluidity compared to those on the inner leaflet during ionomycin-induced scrambling.

In view of the intrinsic physical complexities behind the FP signal, especially in a cell-based assay, our assay is currently not fully quantitative—a limitation that we discussed next. Nonetheless, comparing the scrambling indices between WT and 16FKO cells enabled us to decipher the ionomycin-mediated bilayer movement of PS, PE, and PC and extrapolate the headgroup preference of TMEM16F. In addition to confirming that TMEM16F is the dominant calcium-activated PS and PC scramblase on the plasma membrane that responds to ionomycin treatment, we detected significant retention of NBD-PE scrambling activity in 16FKO cells, indicating TMEM16F is not the major PE scramblase in HeLa cells.

Despite performing parallel control experiments, we acknowledge that the undefined composition and complex PL dynamics in living cells utilized in our experiments pose limitations. Many kinetic processes of NBD-PLs occur during the binding stage, all of which can contribute to the FP value. These include the ongoing scrambling, the rotational movement of the NBD-PL around the long axis of the PL (normal to the membrane plane), various molecular vibrations, and possible ATP-dependent flipping and flopping of PLs. The addition of Alb extracellularly will retard all types of NBD-PL movements on the outer leaflet, but not the inner leaflet. Albumin itself is fluorescent, with its own binding kinetics to the outer leaflet, so its addition can also contribute to the FP value. Extraction of NBD-PLs can occur as well, but it occurs on a longer timescale [after 1 min, given that the half-time for NBD-PL removal via albumin extraction is 0.51 min (54)]. Additionally, it is known that albumin promiscuously extracts NBP-PLs and native lipids from the outer leaflet of the plasma membrane (64, 65), although we mitigated these effects by only considering the Alb-FP data for the first 60 s after Alb addition. Future development of an NBD-specific binder that does not cross-react with other chemical moieties found on the membrane milieu could improve the assay performance.

Importantly, our findings on the headgroup preference of TMEM16F on the plasma membrane are supported by CGMD simulations of TMEM16F in a bilayer with equal proportions of DOPS, DOPC, and DOPE. In our simulations, we observed 2.6 and 3.5 times as many scrambling events for DOPC and DOPS than for DOPE, respectively (Fig. 4A). This qualitatively corroborates the selectivity for PC and PS over PE we found in our cell-based FP scrambling assay (Fig. 3 D–F). The scrambling selectivity in CGMD is explained by a lateral concentration gradient around the protein, with DOPE being depleted and DOPC/DOPS being enriched near the TMEM16F grooves (Fig. 4 B and C). These data suggest that TMEM16F creates an immediate microenvironment that favors DOPS and DOPC over DOPE leading to differing effective concentrations of these PLs in the substrate pool that compete for TMEM16F’s active site. Interestingly, while DOPE scrambling was rarer, it spent less time in transit compared to DOPS and DOPC. The reason for this counterintuitive result is not clear, but it is likely related to the energy landscape experienced by each PL during scrambling. For instance, PC and PS may be enriched in or near the groove compared to PE (as we see in Fig. 4 B, C), and this may be due to more or stronger interaction sites, which slow their diffusion compared to PE, resulting in slower transit times. Additional analysis will be required to reveal the true underlying reason.

While our in vivo experimental measurements and simulation results demonstrate the headgroup preference of TMEM16F, we need to consider the reasons that may contribute to the difference between our work and the previously reported in vitro study (38). Watanabe and colleagues reconstituted purified murine Tmem16f onto a synthetic lipid bilayer array and measured the recovery of fluorescence signals after photobleaching TopFluor-TMR-conjugated PLs that are spiked into the bilayer (38). Using this elegantly designed platform, they recorded comparable scrambling rates when TopFluor-TMR-conjugated PS, PE, or PC were introduced into the bilayer. One major difference from our studies is the relatively simple and skewed chemical composition of PLs on the reconstituted lipid array, where the spiked-in TopFluor-TMR-conjugated PLs are embedded in three-hundred-fold excess of the POPC matrix. Given the tendency of PLs to undergo self-sorting according to their headgroup chemical and physical properties [e.g., the formation of lipid microdomains on the cellular plasma membrane (66)], the scrambling capacity of Tmem16f on the reconstituted lipid array may not reflect its intrinsic nature under physiological conditions. While the lipid composition in our CGMD is also artificial, the presence of three major PLs in equal amounts imposes spatial competition among the PL headgroups for interaction and scrambling by TMEM16F that is absent in the reconstituted lipid array experiment reported by Watanabe and colleagues.

With the finding of the scrambling preference of TMEM16F for PC and PS on the plasma membrane, we believe this study will open many avenues for future research that could be best carried out using FP measurements either on native membranes of living cells or in reconstituted systems. These questions include the following: What are the structural determinants conferring PL headgroup preference to TMEM16F? Which scramblase(s) preferentially act on PE and possibly other PLs? Do other scramblase families, such as the Xkr, TMC, and TMEM63 families, also display PL headgroup preference? Which proteins, lipids, and small molecules modulate membrane scrambling in physiological conditions? We anticipate that this FP scrambling assay can complement the existing assays on proteoliposomes and other reconstituted systems that provide useful model systems. This experimental approach may also pave the way for uncovering the underlying mechanisms of organellar membrane scrambling.

Materials and Methods

Reagents.

We purchased 18:1-06:0 phospholipids NBD-PC (cat# 810132), NBD-PS (cat# 819194), and NBD-PE (cat# 810155) from Avanti Polar Lipids, dissolved them in anhydrous DMSO (Invitrogen, D12345) to 0.2 mM (100×stock), and stored at −20 °C as single-use aliquots. We purchased ionomycin from Cayman Chemical (cat# 10004974) supplied at 14.1 mM in ethanol, diluted it to 5 mM with 100% ethanol, and stored it at −20 °C as single-use aliquots. We purchased Raptinal (cat# 9626-10) from BioVision, prepared it in anhydrous DMSO to 5 mM (500 × stock), and stored at −20 °C as single-use aliquots. Gelatin solution (cat# ES-006-B) used for cell culture plate coating was purchased from EMD Millipore Sigma. We used fatty acid-free BSA from Gemini Bio-Product (cat# 700-107p) in the initial experiments and later switched to recombinant human albumin (10% w/v ALBIX) from Albumedix.

Cell Culture.

We purchased STR profile-certified HeLa and U2OS cells from the UCSF Cell and Genome Engineering Core facility. A549 cells were a gift from Dr. Yu-Ting Chou (Trever Bivona laboratory, UCSF). Cells were cultured in 37 °C incubators supplied with 5% CO₂. Clonal TMEM16F CRISPR-Cas9 knock-out HeLa cells were generated, screened, and validated as described in detail for another human cell line (67). We cultured the HeLa and U2OS cells in DMEM media (Gibco, 10569044) supplemented with 10% FBS (Cultra Pure US origin, Axenia BioLogix), and A549 cells in advanced DMEM media (Gibco, 12491-015) supplemented with 2% fetal bovine serum (Gibco, 26140-079), 1 × GlutaMAX (Gibco, 35050-061), and 100 μg/mL Normocin (InvivoGen, ant-nr-05). Cells were dislodged for passaging with TrypLE (Gibco, A1217701) after washing twice with ambient temperature DPBS (Gibco, 14190136). We froze these cells in Recovery Cell Culture Freezing Medium (Gibco, 12648010) in liquid nitrogen. We used MycoStrip (InvivoGen, rep-mys) to test for Mycoplasma contamination and Cell Culture Contamination Detection Kit (Invitrogen, C7028) to test for fungi and mold contamination.

FP Scrambling Assay.

We seeded cells on gelatin-coated 96-well flat-bottom black plates (Greiner, 655090) at a sufficient cell density to produce 80 to 90% confluency for the next day’s experiment. Due to size and growth rate differences, the following seeding densities (10⁴ cells per well) were used: HeLa: 0.8, U2OS: 1.2, A549: 1.6. Based on our experience, cells that are either too dense or too sparse would affect the FP reading and thus compromise experimental accuracy. Additionally, viable, healthy, and contaminant-free cells (Mycoplasma and fungi/mold) were found to be essential for these experiments.

Experiments were carried out with FP-equipped BMG Labtech PHERAstar FSX and Agilent Synergy H4 microplate readers. Considering the data collection rate of these instruments in kinetic FP mode, we only plated cells to 12 (PHERAstar FSX experiments) or 4 (Synergy H4 experiments) adjacent wells of a 96-well plate for a single experiment so that the FP was recorded for every 7 to 10 s for a given well. In each experiment, the plate layout was as follows: Cells in all but two blank wells received NBD-PL. Both blank and NBD-PL wells were split into two groups with an equal number of wells for apoptosis and nonapoptosis or scrambling and nonscrambling control conditions. First, the cells were washed three times (150 μL each) with Buffer W (HBSS (Gibco, 14065056) supplemented with 10 mM HEPES (Fisher Scientific, BP299-100)) and incubated with 100 μL of 0.2 μM NBD-PL in Leibovitz’s L-15 medium (Gibco, 1141504) for 20 min at ambient temperature covered from light. After two washes (150 μL each) with Buffer W, we added 100 μL of Buffer H (1x HBSS, 20 mM HEPES, 20 mM glucose (Fisher Scientific, BP350-1), 1 mM sodium pyruvate (Sigma-Aldrich, P2256), 1.8 mM CaCl₂, (Fisher Scientific, C79-500)) to each well. For apoptosis experiments, 10 μM raptinal or DMSO are included in the Buffer H. We then placed the plate in the plate reader and launched the preset method with parameters listed below. For scrambling experiments, we manually added 5 μL of 42 μM ionomycin (scrambling conditions) or ethanol vehicle (nonscrambling conditions) diluted in Buffer H during the pause cycle/step as specified below. Ionomycin and control working solutions were prepared freshly for each plate. We employed the built-in injector to add 10 μg ALBIX to each well in experiments performed on the PHERAstar FSX, or manually pipetted 10 μg of BSA to each well in experiments performed on the Synergy H4. Since Alb is intrinsically fluorescent, we used 10 μg Alb in each well to avoid detector saturation. Once added, both calcium and ionomycin were present throughout the remainder of the assay. The final concentration of albumin is 0.0689 μg/μL. The albumins were prepared in buffer H and filtered through a 0.45 μm surfactant-free cellulose acetate Nalgene syringe filter prior to use (ThermoFisher Scientific, 723-2545). We found no significant difference in data quality between recombinant albumin and BSA used in the assays. We chose recombinant albumin in the later experiments. Data acquisition parameters are listed below.

Method	PS_2min	PE_30s	PC_4min	PC_9min	Raptinal
Total cycle #	61	50	76	90	window 1: 60; window 2: 38
Cycle time (s)	8	8	8	10	window 1: 30; window 2: 8
Pause cycle #	8	8	8	7	N/A
Injection cycle #	18	12	38	61	74

PHERAstar FSX from BMG Labtech operated with the PHERAstar software (v5.70 R5) was used to collect data for Fig. 1C, 2B, and 3 A and B, and SI Appendix, Fig. S2 and S3 with the following details. Optical settings: FP module: 488_Ex, 520_EmA 520_EmB, kinetic plate mode, focal adjustment enabled, gain adjustment mP = 180. General settings: bottom optic, settling time 0.1 s, target temperature 37 °C, number of flashes = 100. Injection settings: 40 μL volume (equivalent to 10 μg of ALBIX), 100 μL/s injection speed, standard injection spoon (type A1), smart dispensing option enabled. Shake settings: after the pause cycle #, at 100 rpm for 2 s with the double orbital option.

Synergy H4 from Agilent with Gen5 (v2.09) was used to collect data in Fig. 3C with the following settings. Detection method: FP. Excitation: 485/20 nm; Emission: 528/20 nm. Optic position: Top, 510 nm, Gain: 80, Read speed: Normal, Read height: 4 mm. Light source: Xenon flash. Read type: Kinetic plate mode with three windows. Kinetic window 1: 1 min (7 s interval, 9 reads); plate out (manually added 5 μL of 42 μM ionomycin); shake for 2 s at medium speed; kinetic window 2: 30 s (7 s interval, 5 reads); plate out (manually added 100 μL of 100 μg BSA); kinetic window 3: 5 min (7 s interval, 43 reads).

Fluorescent background signals from the wells without NBD-PL scrambling and nonscrambling controls were taken into account via preselecting them as Blanks in the method acquisition file. For data acquired on the PHERAstar FSX, the calculated mP values from the PHERAstar analysis MARS software (version 4.01 R2) were exported to the Apache OpenOffice Spreadsheet.

For data acquired on the Synergy H4, raw reads of fluorescent signals were exported to Microsoft Excel, and FP values were manually calculated using the following equation:

$$FP=\frac{\left[\left(I_{\mathit{parallel}}^{\mathit{NBD}}\right)-\left(I_{\mathit{parallel}}^{\mathit{Blank}}\right)\right]-\left[\left(I_{\mathit{perpendicular}}^{\mathit{NBD}}\right)-\left(I_{\mathit{perpendicular}}^{\mathit{Blank}}\right)\right]}{\left[\left(I_{\mathit{parallel}}^{\mathit{NBD}}\right)-\left(I_{\mathit{parallel}}^{\mathit{Blank}}\right)\right]+[\left(I_{\mathit{perpendicular}}^{\mathit{NBD}}\right)-\left(I_{\mathit{perpendicular}}^{\mathit{Blank}}\right)]},$$

where $I$ is fluorescence intensity, superscripts $\mathit{NBD}$ and $\mathit{Blank}$ refer to the cell preloading, and subscripts $\mathit{parallel}$ and $\mathit{perpendicular}$ refer to the emission filter orientation.

Before statistical analyses and plotting, the FP time-course readings were normalized (by shifting on the FP axis) by setting the average of the baseline period FP values to 100. Additionally, for plotting the scrambling index graphs, the Alb-FP values were normalized (by shifting on the FP axis) by setting the last FP value of the scrambling period (immediately before Alb addition) the same within replicates.

Data Fitting for Scrambling Index.

Scrambling index was calculated by subtracting FP-Alb trace from nonscrambling condition from that of the scrambling condition for each experiment. The first 60 s of the resultant trace are fitted to a line using the lm function in R, and the slope was extracted as a measure of scrambling efficiency for that PL. The statistical significance between the slopes of scrambling and nonscrambling conditions was compared with Welch’s test.

CGMD Simulation.

A symmetric model of the TMEM16F dimer with dilated TM4-TM6 grooves in both subunits was created by copying and superposing the chain with an “open groove” conformation that resulted from extensive all-atom simulations [cluster 10 reported by Khelashvili and colleagues (59)]. Coarse-grained systems were prepared as described previously (57). Briefly, martinize2 (68) and insane (69) scripts were employed to embed the CG protein into a symmetric and equimolar DOPC:DOPS:DOPE bilayer. This process was repeated independently for three replicates to ensure randomized placement of the lipid molecules in each system.

CGMD simulations were performed using the Martini 3.0.0 force field (70) and Gromacs 2023.3 (71) using a 20 fs time step. Nonbonded interactions were described by reaction-field electrostatics and Van der Waals potentials, using an 11 Å cut-off. Constant temperature (T = 310 K, τ_T = 1 ps) and pressure (P = 1 bar, τ_T = 12 ps) were maintained using the velocity rescaling thermostat (72), and semi-isotropic Parrinello–Rahman barostat (73), respectively. After energy minimization and a short NPT equilibration, all systems were simulated for 10 μs. The first 1 μs was considered a system equilibration phase and omitted from analyses.

Lipid scrambling was analyzed from CGMD simulations as described previously (56, 57). Briefly, for every lipid in every simulation frame, the angle ϑ between the director vector (pointing from the lipid head—NC3 for DOPC, CNO for DOPS, NH3 for DOPE—to the center-of-mass of the two final tail beads C4A and C4B) and the membrane normal was measured. After smoothing the data using a 100 ns running average, a scrambling event is recorded when ϑ transitions from the outer leaflet value (between 145° and 180°) to the inner leaflet value (between 0° and 35°) or vice versa.

Lipid density maps were generated from protein-centered and aligned MD trajectories. For each lipid type $x$, we counted the PO4 beads in 2 × 2 ${\mathrm{\AA }}^2$ bins, collapsing both leaflets onto the XY-plane. Then, for each bin, we divided the total count $n_x$ by the total count for all lipids ${\displaystyle {\sum }_x{n}_x}$, and normalized it by the bulk concentration of 33.33 mol% to obtain the concentration enrichment.

Protein outlines are represented by a concave hull drawn around the protein coordinates (plus Van der Waals radii) from 50 timepoints taken at 180 ns intervals from the last 9 μs of each aligned trajectory. The upper and lower leaflet portions (solid/dashed lines) only include atoms in upper membrane leaflet ($z_{\mathit{mid}}\le z\le z_{\mathit{mid}}+10\mathrm{\AA }$), and the lower membrane leaflet ($z_{\mathit{mid}}\ge z\ge z_{\mathit{mid}}-10\mathrm{\AA }$), respectively, with $z_{\mathit{mid}}$ being the mean z-coordinate of all PO4 beads in the system.

Radial lipid concentrations around the TM4-TM6 groove (represented by the backbone bead of residue I591) were calculated using Gromacs’ gmx rdf module with a 1 $\mathrm{\AA }$ bin width. For each lipid type $x$, the concentration in bin $i$ was calculated as ${[x]}_i=\frac{n_{x,i}}{{\displaystyle {\sum }_x{n}_{x,i}}}\times 100\%$. Means and SE were calculated from the aggregated data from the two grooves and three replicates, excluding the first microsecond for equilibration.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Simulation results and normalized data have been deposited in Figshare (https://doi.org/10.6084/m9.figshare.30197017.v1) (74). All study data are included in the article and/or SI Appendix.