STED-FLCS: An Advanced Tool to Reveal Spatiotemporal Heterogeneity of Molecular Membrane Dynamics.

Heterogeneous diffusion dynamics of molecules play an important role in many cellular signaling events, such as of lipids in plasma membrane bioactivity. However, these dynamics can often only be visualized by single-molecule and super-resolution optical microscopy techniques. Using fluorescence lifetime correlation spectroscopy (FLCS, an extension of fluorescence correlation spectroscopy, FCS) on a super-resolution stimulated emission depletion (STED) microscope, we here extend previous observations of nanoscale lipid dynamics in the plasma membrane of living mammalian cells. STED-FLCS allows an improved determination of spatiotemporal heterogeneity in molecular diffusion and interaction dynamics via a novel gated detection scheme, as demonstrated by a comparison between STED-FLCS and previous conventional STED-FCS recordings on fluorescent phosphoglycerolipid and sphingolipid analogues in the plasma membrane of live mammalian cells. The STED-FLCS data indicate that biophysical and biochemical parameters such as the affinity for molecular complexes strongly change over space and time within a few seconds. Drug treatment for cholesterol depletion or actin cytoskeleton depolymerization not only results in the already previously observed decreased affinity for molecular interactions but also in a slight reduction of the spatiotemporal heterogeneity. STED-FLCS specifically demonstrates a significant improvement over previous gated STED-FCS experiments and with its improved spatial and temporal resolution is a novel tool for investigating how heterogeneities of the cellular plasma membrane may regulate biofunctionality.

The role of the cellular plasma membrane is central in many biological processes. It is well acknowledged that the plasma membrane is not just a simple fluidic system, but it is rather a highly heterogeneous environment constituting a plethora of different lipids, proteins, and sugars, with links to the underlying actin cytoskeleton and the extracellular matrix. Important cellular functions are often triggered by specific interactions between these entities.[1−3] Such interactions usually result in highly heterogeneous diffusion patterns of the involved molecules[4,5] (Figure a). For example, molecules will not diffuse freely but are transiently trapped when interacting with immobilized or slow moving entities. Further, compartmentalization of the membrane by the underlying actin cytoskeleton can result in a hindered, e.g., hop-like, diffusion.

Figure 1

Diffusion modes and impact on STED-FCS measurements. (a) Sketch of representative molecular tracks for different diffusion modes, such as free diffusion (red, upper), trapping diffusion or transient partitioning into domains of higher molecular order (blue, upper, with dots marking the points of transient stops in diffusion), and hop diffusion (black, lower) due to compartmentalization of the membrane by the underlying actin meshwork (brown). (b) Schematic dependency of the apparent diffusion coefficient D on the diameter d of the observation spot, as expected for the different diffusion modes and as for example determined by STED-FCS. Insets: Exemplary observation spots of decreasing size (arrows) as formed by scanning over single emitters for increasing STED intensities. (c) Molecular structures of the fluorescent lipid analogues PE (phosphoethanolamine, acyl chain length C15, saturated, label at headgroup) and SM (sphingomyelin, C13, saturated, labeling via acyl-chain replacement) labeled with the organic dye Atto647N (red) as used in this study. (d) Dependency D(d) measured by single-point STED-FCS for PE in a supported lipid bilayer (SLB, DOPC MICA-supported lipid bilayer, left) and for PE and SM in the plasma membrane of live PtK2 cells (right, PE, black; SM, red), exemplifying free diffusion of PE in both cases, trapping diffusion for SM in cells, and an increased heterogeneity for diffusion in cells (error bars as s.d.m. from 20 measurements).

A widespread tool for investigating molecular diffusion dynamics is fluorescence correlation spectroscopy (FCS), which determines average molecular diffusion coefficients D from thousands of molecular transits through the microscope’s observation spot.[6,7] With measurement times of only a few seconds, and by placing or scanning the spot over distinct points in space, FCS may deliver information on heterogeneity in diffusion over space and time.[8,9] Revealing heterogeneous diffusion modes requires probing diffusion coefficients D at different spatial scales, i.e., for different observation spot diameters d(5,10) (Figure b). Unfortunately, due to the limited spatial resolution of conventional far-field microscopy, nanoscopic details of such heterogeneity are usually only indirectly inferable in such spot-variation FCS experiments, by extrapolating to relevant sub-100 nm spatial scales.

A remedy to this limitation is the use of FCS on a super-resolution stimulated emission depletion (STED) microscope.[11] STED microscopy delivers spatial resolution of <50 nm in living cells and allows a straightforward tuning of the observation spot by the intensity of the added STED laser beam, i.e., a straightforward determination of the D(d) dependency down to the relevant scale.[12,13] Using STED-FCS, we have previously shown that this D(d) dependency varies for the diffusion characteristics of fluorescent lipid analogues of phosphoethanolamine (PE) and sphingomyelin (SM) (Figure c) in the plasma membrane of live mammalian PtK2 cells, revealing transient trapping for the SM analogue (once again experimentally verified in Figure d). Close inspection of the FCS data at small d and for different lipid analogues revealed that, at least for the cell types investigated:[12−14] (i) trapping was due to lipid-specific transient interactions with other membrane entities such as proteins on time-scales of 1–10 ms; (ii) that the lipids hardly moved during trapping (i.e., the binding partner is relatively slow-moving or even immobilized); (iii) that trapping was locally confined in spots of <80 nm in size that were transient in the time scale of 1–10 s; and (iv) that these interactions were independent of the lipid’s preference for liquid-ordered or -disordered environments in model membranes (often referred to as “rafts”). Furthermore, with previous control experiments using different dye labels and label positions as well as different experimental conditions,[12−15] we ensured a nondetectable (at least by STED-FCS) influence by the dye label on the lipid dynamics, by photobleaching, heating, or other (nonlinear) laser-driven effects, and the accurate integration of the lipid analogues into the membrane. Moreover, as expected[16] diffusion of both the Atto647N-labeled PE and SM was free in a fluid model membrane such as (100% DOPC) supported lipid bilayers (SLBs) on plasma-cleaned glass, as here experimentally verified in Figures d and S1.

In the previous STED-FCS experiments, the recording of the D(d) dependency took at least minutes since it required the recording of at least 5–15 s-long FCS data for various intensities of the STED laser, with an unavoidable displacement of the observation (i.e., laser) spot in-between measurements. As a consequence, the STED-FCS experiments averaged over temporal and spatial heterogeneity. Notably, the standard deviations of the averaged values determined for D in live-cell membranes were much larger than those recorded for the fluorescent lipid analogues diffusing in the fluid SLB (Figure d), revealing a hidden heterogeneity in the plasma membrane data. To better highlight heterogeneity in diffusion modes, we urged for a measurement mode that can access the D(d) dependency within one FCS recording, i.e., within 5–15 s only.

Most STED-FCS experiments so far employed pulsed laser light for both excitation and stimulated emission action. In this configuration, the only way to tune the observation spot is by varying the intensity of the STED laser, which is unfortunately not straightforwardly realizable within a single FCS recording. However, we have previously shown that in an experimental arrangement that uses pulsed excitation, continuous-wave (CW) STED laser beams and time-gated detection, the observation spot size d can also be tuned by the position of the time gate (gated STED, gSTED).[17,18] This is because in this arrangement, the (cumulative) probability for stimulated emission increases over time after the excitation laser pulse, and thus, fluorescence photons detected after different time-delays, i.e., at different time-gates starting from time Tg with respect to the excitation laser pulse, originate from increasingly decreased central areas of the observation spot[18,19] (Figure S2). As a consequence, FCS data can be generated for different observation spot sizes d and thus the D(d) dependencies constituted out of a single measurement (not requiring recordings at multiple STED intensities as before). Specifically, by setting different time gates Tg and calculating data from photons arriving only at times > Tg, different correlation data are acquired out of a single photon stream recorded by time-correlated single-photon counting (TCSPC) for one STED laser beam intensity (Figure a).[18] However, the gSTED-FCS analysis in its current state is still limited. For small time delays Tg, the gSTED-FCS approach averages over a large time period of STED laser beam action, resulting already in rather reduced spot sizes d. As a consequence, the range of diameters accessed by gated STED-FCS from a measurement at a single STED laser beam intensity is rather low, e.g., 50 nm < d < 110 nm, as determined both theoretically (Figure S3d) and experimentally (Figure b, gated STED-FCS analysis of free diffusion of PE in the fluid SLB). Consequently, a full D(d) dependency (from confocal >150 to 50 nm diameters) would require the recording at additional, lower STED intensities, again limiting the time resolution.

Figure 2

Principle of STED-FLCS. (a) Time t courses (left panels) of the excitation laser (green, upper, sketched), the STED laser (brown, below, sketched), decay of fluorescence emission (middle, experimental data from single emitter) at the focal center (green) and periphery (hollow green), and detection windows (lower panels) for cwSTED-FCS (light blue), gSTED-FCS (dark blue with time gate Tg), and STED-FLCS (violet) with detection window Δ and for increasing time gate Tg (red arrow). The right panels show experimental images of the focal intensity distribution of the excitation (upper, green) and STED (second upper, brown) lasers, as well as the fluorescence images of a single emitter as obtained for the different gating conditions, as labeled or indicated in the left panels. Scale bars 200 nm. (b) Experimentally determined dependency of the observation spot diameter d (full width at half-maximum, fwhm) on the time gate Tg for the different STED-FCS modes as indicated in the legend, and for different STED laser powers P in the case of STED-FLCS, as indicated in brackets in mW (P(gSTED-FCS) = 300 mW). The dynamic range in d is highest for STED-FLCS. The diameter d was determined from FCS analysis of free diffusion of the PE analogue in the SLB. (c) Dependency of anomaly α on d for the gSTED-FCS and STED-FLCS data of free diffusion of PE in the SLB (P = 300 mW). An apparent anomaly appears in the case of gSTED-FCS, especially for small Tg, i.e, . larger d; an artifact that is avoided in the STED-FLCS recordings. (d) Representative dependencies of the apparent diffusion coefficient D on d for single 10 s STED-FLCS recordings of PE diffusion in SLBs, allowing the determination of D and D for large and small diameters, respectively (gray bars). Inset: Scatter of value pairs (D, Δ = D/D) for 20 STED-FLCS recordings of different measurement times (green, 5 s; red, 10 s; black, 25 s), indicating low heterogeneity of diffusion with a free Brownian characteristics (center at Δ = 1 and D0 = 3.8 μm2/s).

The basics of gSTED-FCS are similar to that of fluorescence-lifetime-correlation spectroscopy (FLCS), where also different FCS data is generated from the same TCSPC photon stream. In FLCS, the photons are weighted differently as to separate FCS data between labels of different fluorescence lifetime.[20,21] Starting from that idea, we here introduce STED-FLCS as an improved version of gSTED-FCS. In STED-FLCS, correlation data is separated for different observation spot sizes d by choosing photons from time gating intervals ΔT centered at different Tg (instead of from the whole time span > Tg as for gSTED-FCS) (Figure a). By choosing intervals ΔT instead of whole time spans, we both theoretically (Figures S2c and S3a) and experimentally (Figure b) showed a continuous reduction of the diameter d of the observation spot from close to diffraction-limited 240 nm down to in this case approximately 50 nm. In addition, the spatial profile of the fluorescence emission in the observation spot of the STED-FLCS data was well described by a Gaussian, while this was less the case for gSTED-FCS (Figure S3b). Since the gSTED-FCS data, especially for small Tg, results from averaging over a large time period of STED laser beam action, the profile of the fluorescence emission in the observation spot is more Gaussian–Lorentzian.[22] gSTED-FCS data thus revealed a more stretched decay (compared to conventional FCS data resulting from free diffusion through a Gaussian-shaped observation spot), and in our case we had to introduce an anomaly α < 1 to accurately fit the experimental gSTED-FCS data (PE diffusing freely in the fluid SLB) using conventional FCS data analysis (Figure c). Unfortunately, values of α < 1 usually indicate anomalous or hindered diffusion. Thus, the accurate fitting of gated STED-FCS data would have required more complex fitting routines.[23,24] Yet, we received α ≈ 1 from the conventional FCS analysis of the STED-FLCS data instituted from the same data set (Figure c).

To highlight the advantages of the STED-FLCS analysis, we anticipated a direct comparison to the previous STED-FCS studies, thus using again the example of the membrane diffusion dynamics of the PE and SM analogues. Figure d shows representative D(d) dependencies from the analysis of subsequent 5–25 s long STED-FLCS measurements of the diffusion of the PE lipid in the fluid SLB. As expected, the D(d) dependencies were constant, indicating free diffusion. Also, the FCS data revealed a sufficient quality (Figure S4), and consequently the variation in parameters was very low, even for measurement times as low as 5 s. Each STED-FLCS recording let us determine two important parameters at once, the apparent diffusion coefficient D of the confocal recordings, roughly outlining the overall macroscopic mobility, and the ratio Δ = D/D of the apparent diffusion coefficients determined for the smallest observation spots (D) and the largest observation spot (D). This ratio is Δ = 1 for free diffusion, <1 for heterogeneous diffusion due to, for example, transient trapping, and >1 for hop diffusion[13,14,25] (compare Figure b,d). For both parameters and for measurement times of 5–25 s, we observed a low scatter in values around average values of D = 3.8 μm2/s and Δ = 1, as well as expressing the negligible change in D(d) and thus diffusion mode in the time regime of seconds (inset Figure d, for SM see Figure S1b). The observation of variation of diffusion modes on the second time scale was impossible before since D and D had to be accessed from separate measurements, i.e., with decreased time resolution.

The range of diameters d accessed for different time delays Tg within a single STED-FLCS measurement varied with the STED intensity as well as with the width ΔT of the detection windows. As shown experimentally and theoretically in Figures b and S5a, respectively, the smallest observation spots for determining D were created with larger STED intensities, but this also came along with a reduction of the observation spot determining D, i.e., the whole range of diameters accessed (max to min) was shifted to smaller values. While the choice of ΔT did not change the size of the smallest observation spots reached for large Tg, we more closely approached the diffraction-limited observation spot for small Tg with narrower time windows, i.e., smaller ΔT (Figure S5c). As pointed out before (Figure b), this results from the fact that, especially for small Tg, large ΔT entail averaging over a larger time periods of STED laser beam action. Consequently, the largest range of diameters d (max to min) within a single STED-FLCS measurement is in principle realized with small ΔT. However, smaller ΔT impose decreased number of detected photon and thus decreased signal-to-noise ratios; the signal-to-noise ratio in any case decreases with increasing Tg(18,19) (Figure S5c). Therefore, one has to balance the range of accessible d-values against signal-to-noise ratios. In our case, when using Atto647N-labeled lipids and powers P = 300 mW of the STED light at 770 nm we achieved sufficient signal-to-noise in the FCS data for ΔT = 1.5 ns and 10 s recordings (Figure S4).

We next moved to the investigation of the diffusion characteristics of the fluorescent lipid analogues in the plasma membrane of living cells.[12−14] Using STED-FLCS with parameters as defined in the previous paragraph, various D(d) dependencies were determined from 10 s recordings of the Atto647N-labeled PE and SM lipids diffusing in the plasma membrane of living Ptk2 cells at different times and places (Figures a and S6a). Compared to the model membrane data (Figure d), the D(d) dependency varied strongly in time and space, with changing values in both D and Δ (compare also Figure S7, depicting specifically changes in D and Δ over time). However, averaging over a multiple of these data again revealed a D(d) characteristic as in Figure d, with the same large standard deviations. The heterogeneity in space and time of plasma membrane lipid diffusion is further highlighted in Figure b, where the scatter in value pairs (D, Δ) is depicted for the STED-FLCS recordings of PE and SM. Both fluorescent lipid analogues not only showed vast changes in diffusion modes, mainly between free (Δ ≈ 1) to trapping (Δ ≪ 1) diffusion, but also strong changes in overall macroscopic mobility D over space and time. While the values scattered around D = 0.4 μm2/s and Δ = 1 for PE, confirming close to free diffusion, in the case of SM these parameters ranged from values similar to PE down to values as low as D < 0.1 μm2/s and Δ < 0.14, indicating trapping diffusion, i.e., interaction with less mobile membrane entities at these points in space and time. For SM, values of D and Δ correlated in a positive way (compare also Figure S6). This is to be expected since trapping also leads to a macroscopic slow-down.[12] Especially incidences of more pronounced free diffusion (Δ around 1) showed large variations in overall macroscopic mobility D, indicating, for example, variations in membrane viscosity, fluidity or molecular crowding. Therefore, distinct interaction sites and spots of different mobility changed strongly over space and time. Since values of D and Δ were distinct even for the STED-FLCS recordings of 10 s, the different conditions (trapping, free diffusion, or certain overall mobility) had to be stable for at least a few seconds; otherwise, they would not have dominated the respective 10 s long single recordings. These findings are in accordance with previous STED-FCS recordings using fast beam-scanning, revealing for SM several distinct interactions sites that were stable for several seconds up to 10 s.[14] However, those scanning STED-FCS recordings could not determine D and Δ out of one recording (but rather either only D or D). Therefore, while scanning STED-FCS highlighted changes in space at a distinct time-point in either D or D, our current STED-FLCS recordings directly revealed changes in D, D and most importantly Δ over time (Figure S7). In contrast, heterogeneity in space was only approached by comparing value pairs (D, Δ) taken at different spots and different cells at different time points (Figure ).

Figure 3

STED-FLCS measurements of lipid diffusion in the plasma membrane of live PtK2 cells. (a) Dependencies D(d) for representative 10 s measurements of SM with the parameters D and D marked by gray bars. A large heterogeneity in diffusion modes becomes obvious (purple, free; blue, trapping; magenta, hopping). (b–d) Scatter of value pairs (D long-range apparent diffusion constant, Δ = D/D diffusion mode) for STED-FLCS recordings of (b) PE (red, N = 28) and SM (green, N = 84, the dotted vertical line indicates the arbitrary threshold for events of strong trapping D < 0.14), (c) SM following LatrB treatment for actin depolymerization (N = 109), and (d) SM following COase treatment for cholesterol depletion (N = 56). Diffusion of PE is mainly free (Δ ≈ 1), while most of the SM measurements indicate trapping (Δ ≪ 1), which is much less upon treatment with COase or LatrB. In all cases, diffusion is very heterogeneous as indicated by the strong scatter in values. (e,f) Boxplots of values of D (e) and Δ (f) determined from the respective data of b–d as labeled, indicating average values and standard deviations. For SM we differentiated between strong (D < 0.14) and weaker (D > 0.14) trapping events. Heterogeneity in membrane diffusion (especially in Δ) is slightly reduced after COase treatment.

Our STED-FLCS data also observed some incidences of hop diffusion (Δ ≫ 1). In this mode, diffusion of molecules is hindered by compartments.[4,10,25,26] While molecules diffuse freely within these compartments, diffusion from one compartment to the next is hindered (Figure a). One reason for such compartmentalization is the cortical actin cytoskeleton, which usually forms a meshwork and might act as a barrier for diffusion. As a consequence of this hop or compartmentalized diffusion, apparent diffusion coefficients measured for large observation spots (D) are rather low since a molecular transits involve crossing of several of these barriers, while those measured for small observation spots (D) are large since they probe the free diffusion inside the compartments only.[13,25] This characteristic was reflected in our data, but present for a few points only. A recent study indicated that hop diffusion was not a dominant feature in STED-FCS recordings on PtK2 cells due to an average meshwork size of 25 nm, i.e., below the spatial resolution of those as well as our current measurements.[25] The few incidences of hop diffusion might indicate larger meshwork sizes at that point in space and time.

Our previous STED-FCS recordings demonstrated a dependency of the transient trapping of SM on levels of cholesterol and on the actin cytoskeleton.[13]Figure c,d depicts the scatter in value pairs (D, Δ) as determined from STED-FLCS recordings of SM diffusion in the plasma membrane of live Ptk2 cells treated with Latrunculin B (LatrB, for actin cytoskeleton depolymerization) or Cholesterol Oxidase (COase, for cholesterol depletion). While the overall variation in values of D and Δ was still strong, the extent of trapping decreased, as reflected by an overall increase in values of Δ (and more strongly for COase than for LatrB, compare also Figure S6). The STED-FLCS data confirmed our previous observations that the transient interactions of SM were cholesterol assisted and the mobility of the binding partners was hindered by the actin cytoskeleton.[13] Yet, our STED-FLCS data now allowed us to take a closer look on whether heterogeneity in membrane mobility had changed upon COase or LatrB treatments. Figure e,f depicts the boxplots (average and standard deviations) of the values D and Δ for all cases, diffusion of PE, of SM, and of SM following COase or LatrB treatments. For SM we have differentiated between events of strong trapping (D< 0.14) and weak trapping or free diffusion (D > 0.14). Clearly, as noted before, average values of both D and Δ were highest for the generally free diffusing PE lipid analogue, and those of SM whether for the case of D0 > 0.14 or after LatrB or COase treatment all revealed weak trapping only. However, more notably, the standard deviations indicate that the variation in both overall macroscopic mobility (D) and more strongly in differences in diffusion modes (Δ) was reduced following COase treatment. Consequently, heterogeneity in lipid diffusion dynamics was reduced upon cholesterol depletion.

For the drug treatments, one cannot fully exclude incidences where a cell started apoptosis. However, we applied concentrations and incubation times for these drugs (Supporting Information) as used multiple times before in membrane diffusion studies, allowing a safe comparison.[5,12,13] In addition, we checked for a correlation between trapping Δ and fluorescent lipid analogue concentration. We estimated the concentration from the average particle number N = 1/G(0) in the observation spot (calculated from the amplitude G(0) of the correlation data generated for large observation spots) and determined a random distribution in value pairs (Δ, N) and no notable change in scatter of values when changing the concentration in fluorescent lipid analogues by an order of magnitude (Figure S8).

In conclusion, we have introduced STED-FLCS as an advanced tool to investigate molecular diffusion modes and especially their spatiotemporal dynamics. We highlighted changes in diffusion modes with a time resolution of down to a few seconds. Extending our previous studies, we now demonstrated large spatiotemporal heterogeneity in the diffusion dynamics of fluorescent lipid analogues in the plasma membrane of living cells. Specifically, a sphingolipid analogue SM showed distinct temporal and spatial incidences of interaction or trapping sites, where, as pointed out before,[12,13,27] individual lipids resided for approximately 10 ms. The interaction sites were transient for at least a few seconds, and confirmed previous scanning FCS data, but now directly highlighting that the diffusion mode in these trapping sites was indeed transient trapping. As expected, treatments for cholesterol depletion and actin depolymerization reduced the occurrence of these interaction sites. However, STED-FLCS also for the first time demonstrated a slight reduction in the spatiotemporal heterogeneity of lipid diffusion upon treatment for cholesterol depletion, specifically with regards to changes in diffusion mode. We have to note that due to the need to record data for at least 10 s (to reach an accurate enough signal-to-noise ratio) the STED-FLCS recordings averaged over any heterogeneity on faster time scales than around a second. The only way to increase the time resolution will be the development of fluorescent labels with further increased brightness. An option in STED-FLCS analysis is the adaptation of the width ΔT of the gating window to the position of the time gate Tg, specifically reducing the width for smaller Tg and thus optimizing the range of accessible observation spot diameters from a single recording (Figure S5d), as discussed above. In addition, STED-FLCS can readily be applied on a commercial STED setup (Figure S9). The ultimate STED-FCS measurement would be scanning STED-FLCS, i.e., the combination of fast beam-scanning and STED-FLCS, allowing the simultaneous disclosure of diffusion modes (D(d) dependencies) at different spatial points (e.g., along a line or circle). This would with great precision allow highlighting changes in heterogeneity of lipid diffusion dynamics over different parts of a single cell (e.g., cell body versus lamellipodia). STED-FLCS represents a new technology to highlight important molecular processes in the cellular plasma membrane, but should be extendable to other (intracellular) membranes or cytosolic processes.