Measuring the hierarchy of molecular events during clathrin-mediated endocytosis

Abstract

A well-orchestrated hierarchy of molecular events is required for successful initiation and maturation of clathrin-coated pits (CCPs). Nevertheless, CCPs display a broad range of lifetimes. This dynamic heterogeneity could either reflect differences in the temporal hierarchy of molecular events, or similar CCP maturation processes with variable kinetics. To address this question, we have used multi-channel image acquisition and automated analysis of CCP dynamics in combination with a new method to quantify the time courses of recruitment of endocytic factors to CCPs of different lifetimes. Using this approach we have extracted the kinetics of recruitment and disassembly of fluorescently labelled clathrin and/or AP-2 throughout the entire lifetime of temporally-defined CCP cohorts. Based on these analyses, we can (i) directly correlate recruitment profiles of these two proteins, (ii) define 5 distinct CCP maturation phases, i.e. initiation, growth, maturation, separation and departure, (iii) distinguish events with absolute versus fractional timing and (iv) provide information on the spatial distribution of fluorophores during CCP maturation. Emerging from these analyses is a more clearly defined role for AP-2 in determining the temporal hierarchy for clathrin recruitment and CCP maturation. This method provides a new means to identify other such hierarchies during CCP maturation.

Introduction

The use of fluorescent fusion proteins combined with total internal reflection fluorescence microscopy (TIR-FM) has enabled detailed visualization of the assembly, maturation and budding of clathrin coated pits (CCPs) at the ventral cell membrane (1–5). The extent to which these analyses provide new insights into the mechanisms governing clathrin-mediated endocytosis (CME) depends, in part, on the ability to extract quantitative data from the resulting time-lapse movies. Important parameters that can inform mechanistic hypotheses include the lifetimes of CCPs, the concentration and stoichiometry of molecules associated with CCPs, and the timing of recruitment and localization of components of the endocytic machinery.

Using automated tracking and analysis of CCP dynamics (6) we have observed a remarkable heterogeneity in CCP lifetimes. This heterogeneity is in part due to (i) the co-existence of both short-lived, so-called ‘abortive’ CCPs (5;7) and longer-lived ‘productive’ CCPs, and (ii) the fact that even within the population of productive CCPs, lifetimes range from ~30s to several minutes. It is unclear whether this broad distribution is random or coupled to systematic variations between CCPs, particularly in terms of their molecular composition (8). The latter hypothesis implies that it will be possible to gain insights into rate-limiting steps during CCP maturation based on the correlation between the distribution of CCP lifetimes and the distribution of metrics describing CCP composition, e.g. the concentration, stoichiometry, and kinetics of recruitment of endocytic accessory factors.

Several studies have examined the temporal relationships that govern the recruitment of components of the endocytic machinery to CCPs (7;9–13). However, instead of exploiting the heterogeneity in CCP composition and lifetimes as a source of mechanistic information, heterogeneity was typically eliminated by averaging a subset of CCPs with similar lifetimes and stereotypic behaviors. Since averaging time courses of varying length is not straightforward, previous studies have aligned all observable events to either the appearance or disappearance of CCPs, thereby focusing on molecular events that occur either during the initial stages of CCP assembly, or during the late stages prior to CCP departure (i.e. CCV formation), respectively. By averaging trajectories with a wide range of lifetimes, information regarding events further away from the alignment point will be obscured. On the other hand, some averaging of intensity time courses is necessary, because intensity measurements of CCPs are typically very noisy. This is particularly true for TIR-FM measurements, where the intensity of fluorescent objects depends both on the number of fluorophores and on their distance from the plasma membrane, incorporating noise from both sources. Therefore, the methodical challenge in quantitatively analyzing CCP intensity time courses is to filter fluctuations that are related to noise without masking the meaningful differences in recruitment kinetics that determine the heterogeneous behavior of CCPs.

To overcome these problems and to extract mechanistic insight from the heterogeneous behavior of individual CCPs, we have developed a new approach that uses the entirety of CCP trajectories detected by TIR-FM to calculate averaged intensity time courses for temporally-defined CCP cohorts, spanning a wide range of CCP lifetimes. Using this approach, statistically significant differences in the averaged time courses would then indicate a critical kinetic determinant regulating CCP maturation. By combining this assay with multi-color TIR-FM, the time course of clathrin content can be directly correlated to the dynamics of other coat proteins or endocytic accessory factors across the entire lifetime of CCPs. This method is also amenable to analysis of combined TIR- with Epi-illumination, providing information on the spatial distribution of fluorophores at different stages of CCP maturation. Here we describe the development of these algorithms and their use in analyzing the relative recruitment kinetics and characteristics of clathrin and AP-2 complexes to CCPs.

Results and Discussion

To assess the dynamics of protein recruitment to CCPs, we used time-lapse TIR-FM to measure CCP intensities in epithelial BSC1 cells stably expressing either rat brain clathrin light chain a, fused to EGFP (LCa-EGFP) or σ2-EGFP, a fluorescently labeled subunit of the tetrameric adaptor protein complex AP-2 (see Methods). These cells are well-suited for this type of analysis due to (i) their morphology and adherence to the substratum, (ii) their homogeneity in CCP appearance, and (iii) their well characterized CCP dynamics (5;7;8;14–16). Despite the homogenous appearance, the distribution of CCP lifetimes in these cells is remarkably broad. Therefore, BSC-1 cells provide an excellent model to begin to identify the molecular events that determine this heterogeneity.

Lifetime binning and aligned averaging retains fast dynamics of the maturation process

We applied a fully automated detection and tracking assay (5) to extract hundreds of complete CCP trajectories of variable lifetimes in each cell movie. CCP intensities were read out directly at the tracked positions in a small circular area corresponding to the size of an Airy disc, i.e. the diffraction-limited spot image of a point source. Intensities were also read out 15 frames before the start and 15 frames after the end of each detected trajectory, using the first or last known position (P_first and P_last, respectively) of the tracked object.

Because these intensity time courses (particularly in TIRF) are inherently noisy, the definition of a representative intensity time course generally requires averaging of a large number of individual CCP trajectories. In most previous studies of CCP dynamics, the total number of time courses was relatively low, i.e. a few tens to hundreds. Thus, all available data had to be pooled for averaging. To accommodate the varying lengths of the trajectories, the intensity time courses are usually first aligned to a reference point and then averaged (e.g. Fig. 1A, B appearance- and disappearance-aligned, respectively). This strategy represents the ‘typical’ behavior of CCP intensities only in the immediate vicinity of the reference point. At time points farther away, the variability of the trajectory length obscures prominent features in the averaged intensity time course. Furthermore, this approach has the drawback of disregarding any potential systematic heterogeneity in maturation behavior between CCPs of different lifetimes.

Figure 1. Discrete stages of CCP maturation defined by lifetime binning and aligned average.

(A) Example of raw intensity time courses (grey) aligned to the point of appearance, and their average (green). (B) Intensity time courses as in A aligned to the point of disappearance, and their average (red). (C) Average intensity time courses derived from CCPs within a single lifetime cohort. The final curve (purple) is the weighted average of the appearance-aligned (green) and disappearance-aligned (red) curves; the relative weights of the two curves follows a sigmoid distribution (inset). (D) Temporally-defined landmarks during CCP maturation defined in an average intensity time course of LCa-EGFP (illustrated for a 60s cohort). Reference points and phase segmentation as described in the main text.

The availability of lifetime information and intensity data from tens of thousands of CCPs has enabled us to address the averaging problem of length variability, and to further analyze their behavioral heterogeneity. To this end, we subdivided the pool of all CCP intensity time courses into groups of similar lifetime termed ‘cohorts’. We chose the mean lifetimes (lft_mean) of the cohorts to cover the entire spectrum of CCP lifetimes ranging from a few seconds to minutes; the range of lifetimes within an individual cohort (lft_mean±Δlft) was chosen such that Δlft did not exceed 10% of lft_mean (e.g. the 60s-lifetime cohort comprises all CCP trajectories with lifetimes ranging from 54–66s). The individual intensity time courses within different cohorts were then averaged independently, creating one distinct average intensity time course per cohort. Due to the high number of available trajectories, each lifetime cohort still contained some hundreds or thousands of trajectories. Thus, this strategy has the desired effect of sufficiently ‘averaging out’ intensity noise within the cohort, while at the same time preserving systematic heterogeneities between CCPs with different lifetimes. While the lifetimes of CCPs within a given cohort fell within a sufficiently narrow range to ensure similar maturation kinetics, the individual trajectory lengths still varied; for example, at a frame rate of 2s, the lengths of the trajectories in the 60s-cohort ranged from 27 to 33 frames. Dynamic events like the internalization of CCPs can be as short as 2–3 frames. Therefore, any alignment of time courses within a cohort to a fixed reference point (e.g. the time point of disappearance) may still introduce significant distortion in the averaged time course, as shown by the difference between the ‘appearance-aligned’ and the ‘disappearance-aligned’ average time courses (Fig. 1C). Thus, pre-sorting of trajectories alone is insufficient for an integral analysis of recruitment kinetics.

Alternative approaches are possible to obtain higher resolution information away from the time point of alignment. For example, time courses within a particular cohort could be interpolated to the mean lifetime of the cohort, resulting in a linear contraction of longer trajectories and a linear expansion of shorter trajectories, respectively. This procedure would retain features that scale linearly with CCP lifetime (e.g. a maturation event that typically takes place 20% into the CCP’s total lifetime), but would strongly distort non-linear features (e.g. an event that typically takes place 10s after initiation, regardless of the total CCP lifetime). To best preserve both of these temporally-distinct features while averaging, we have implemented a sliding sigmoidal weighting between the ‘appearance-aligned’ and ‘disappearance-aligned’ time courses (Fig. 1C, inset). As indicated by the purple band (Fig 1C), the final time course specifically retains the fast intensity dynamics at the beginning and end of CCP trajectories, while it preserves potential non-linear features during the maturation process. In addition, the grouping into lifetime cohorts allows us to measure heterogeneity in maturation behavior between CCPs of different lifetimes.

CCP growth and maturation is a multi-step process

Based on this approach, the averaged intensity time courses of the LCa-EGFP signal could be segmented into distinct consecutive stages (representative 60s-cohort average time course is shown in Fig. 1D; see also Fig. 2A). Reference points P (characterized by a time point t and a corresponding intensity I) that marked the boundaries between phases were computationally identified. These reference points represent position or intensity features, or locations where significant changes in curvature occurred (Fig. 1D). Specifically, P_first and P_last are the first and last detected points of the trajectory. Averaging of these intensity time courses enabled us to detect residual signal intensities for a few seconds before CCP appearance, P₁, and after, P₆, CCP disappearance, respectively. This indicates that while the intensities in these frames were below the detection threshold in the individual noisy CCP trajectories, the camera still recorded a slightly elevated accumulation of fluorescent label, now visible in the averaged time course. P_max is the location where maximum intensity is reached. P₀ and P₀₀ are the points 15 frames before P_first and after P_last, respectively, thereby defining the baseline intensity, which did not differ significantly from cell surface regions between CCPs. Points P₁-P₆ were determined computationally as the positions where the highest change in slope occurred on a local stretch of the curve (see Suppl. Fig. 1A) and mark the boundaries between distinct intensity phases of CCP maturation.

Figure 2. Differential behaviors of AP-2 and clathrin during CCP maturation.

(A) Single-channel LCa-EGFP time courses. (B) Single-channel σ2-EGFP time courses. (C) Example of time point-vs-lifetime plots for events of known timing. (D) Time points of segmented references in LCa-EGFP intensity time courses. (E) Time points of segmented reference points in σ2-EGFP intensity time courses. (F) Zoom-in of the initiation phase in LCa-EGFP; intensity time courses are normalized to the intensity at t_first. (G) Zoom-in of the initiation phase in σ2-EGFP; intensity time courses are normalized to the intensity at t_first. (H) Measured slope of LCa- and σ2-EGFP time courses during initiation phase. (I) Zoom-in of the departure phase in LCa-EGFP; intensity time courses are normalized to the intensity at t_last. (J) Zoom-in of the departure phase in σ2-EGFP; intensity time courses are normalized to the intensity at t_last. (K) Measured slope of LCa- and σ2-EGFP time courses during departure phase.

The parameters we could extract from this segmentation and then compare across lifetime cohorts were the (i) absolute time points t, (ii) intensities I of the reference points, (iii) length of the phases between these reference points, and (iv) slope of the maturation time course at points of interest. Based on these temporal landmarks, we designated (i) a previously undefined (8) initiation phase (P₁-P₂) corresponding to the first burst of clathrin recruitment, followed by (ii) a slower growth phase (P₂-P₃), (iii) a plateau phase (P₃-P₄) during which the fluorescence intensity leveled off and reached its maximum, (iv) a separation phase (P₄-P₅) corresponding to a slow decrease in intensity, and (v) the final departure phase (P₅-P₆) characterized by a very rapid decay of the fluorescence intensity (Fig. 1D).

To determine the contribution of these different phases in CCPs of different lifetimes, and to quantify differences between clathrin and AP-2, we applied this cohort averaging approach to analyze CCP trajectories in TIR-FM movies acquired with a frame rate of 0.4s, using either LCa-EGFP or σ2-EGFP as the CCP marker. We found that intensity profiles for clathrin and AP-2 had distinctly different shapes (Fig. 2A,B). Clathrin time courses were characterized by a longer growth phase, which for longer-lived CCPs turned into a plateau phase of increasing duration, followed by a fast intensity decay. As a result, LCa-EGFP intensity profiles were markedly skewed to the right (Fig. 2A). In contrast, AP-2 exhibited a slower intensity decay phase, making the time courses relatively symmetric (Fig. 2B). Reference points in LCa-EGFP and σ2-EGFP intensity time courses were then identified computationally as described above (see Fig. 1D). Next, we examined the time courses in more detail by plotting the time points of the phase boundaries as a function of the total lifetime (Fig. 2C–E). We defined ‘absolute’ events that occur at constant time intervals relative the start or end of each trajectory and ‘fractional’ events that occur at a specific fraction of the total CCP lifetime. For example, a hypothetical absolute event that occurs with a fixed lag time relative to the start of the CCP trajectory (e.g. t_a=1s after CCP appearance, Fig. 2C) will give rise to a straight line parallel to the x-axis, with an offset corresponding to the time lag. An event that occurs at a fixed time prior to the end of the CCP trajectory (e.g. t_b=2s before CCP disappearance, Fig. 2C) creates a straight line parallel to the line of unity, again with the offset corresponding to the timing of its occurrence before CCP disappearance. In contrast, events that occur at a specific fraction of the total CCP lifetime create a straight line of slope 0<m<1, where the slope indicates the relative timing of the event (e.g. t_c=80% of total lifetime, slope m=0.8, Fig. 2C). The timing of some events may, of course, be neither entirely absolute nor fractional; an event could have a saturating behavior (e.g. Fig. 2C, t_d and t_e), increasing close to linearly for short lifetimes but reaching plateau levels at longer lifetimes. Even so, the length of the phase between non-linear events (in this example, the difference t_f=t_e-t_d) can still be linear. Thus, we can establish a quantitative hierarchy of the timing of events during CCP maturation by plotting the time points at which specific events occur (or the phase length between specific events) versus the lifetimes of CCPs. A linear fit of time point-vs-lifetime plots (or phase length-vs-lifetime plots) allows us to extract the slope m, which designates the relative timing during CCP maturation, and the linear offset b, which designates an absolute (non-lifetime dependent) time shift relative to the start or end of the trajectory.

The time point plot of the reference points in LCa-EGFP and σ2-EGFP (Fig. 2D,E) extracted from the intensity time courses in 2A,B showed that both ‘absolute’ and ‘fractional’ behaviors were present during CCP maturation. The plots of the initiation phase start and end time points were parallel to the x-axis, and the plots of the departure start and end time points were parallel to the line of unity, indicating that for both clathrin and AP2, the initiation and departure phases were of constant absolute length (i.e. 2s) across the entire range of examined lifetimes.

For both LCa-EGFP and σ2-EGFP labeled CCPs, the start of the plateau phase, the time point of the maximum, and the start of the separation phase increased with lifetime, and could be well approximated with a linear function. Notably, the relative lengths of the growth and plateau phases were significantly longer for LCa-EGFP compared to σ2-EGFP, at the expense of the separation phase. This is consistent with the hypothesis that AP-2 moves out of the evanescent field prior to clathrin, either through disassembly or through redistribution on the CCP surface (4;8;13;17).

CCPs accumulate clathrin and AP-2 with different kinetics

To directly compare the kinetics of clathrin and AP-2 accumulation in CCPs, we extracted the characteristic slopes of the intensity profiles during the initiation and departure phases (i.e. the slope at points P_first and P_last in the schematic in Fig. 1D) for different CCP lifetimes. During the initiation phase (Fig. 2F,G; intensity time courses normalized at P_first), the positive slopes of LCa-EGFP intensity time courses were constant across a wide range of CCP lifetimes (Fig. 2H, blue trace), indicating that under control conditions, the rate of clathrin incorporation into CCPs during initiation is independent of the CCP lifetime. A slight reduction in the rate of clathrin recruitment was observed only in extremely short-lived CCPs (<5s lifetime), which may correspond to the ‘early’ abortive population (5). In contrast, the slopes of σ2-EGFP intensity time courses during initiation (Fig. 2H, red trace) were constant for CCPs with longer lifetimes, but significantly and continuously decreased for CCPs of lifetimes below ~30s. These observations indicate that short-lived CCPs recruit AP-2 at a reduced rate compared to longer-lived species. Previously, it has been shown that this short-lived population represents primarily abortive CCPs that fail to pass a putative maturation checkpoint, and that AP-2 plays a critical role in stabilizing nascent CCPs during this transition (5;14). Thus, the observed bi-phasic behavior further supports the hypotheses that (i) abortive CCPs are characterized by a reduced rate of AP-2 recruitment, consistent with a stabilizing role of AP-2, (ii) a threshold rate of AP-2 recruitment shortly after CCP initiation might be necessary or beneficial for successful completion of CCP maturation, and (iii) the rate of AP-2 recruitment may be predictive of a CCP’s fate (i.e. whether it becomes abortive or productive).

A bi-phasic behavior for both AP-2 and clathrin intensities was observed during the departure phase (Fig. 2I,J; intensity time courses normalized at P_last). The negative slopes of LCa-EGFP and σ2- EGFP intensity profiles (Fig. 2K, blue and red traces, respectively) increased similarly for CCP lifetimes below 30s and became constant for CCP lifetimes above 30s. Again, this behavior suggests that a transition takes place at this lifetime threshold, reflecting mechanistic differences between abortive and productive CCPs. Specifically, we propose that for cohorts with lifetimes <30s (i.e. predominantly abortive structures), the slopes indicate the kinetics of coat disassembly, whereas cohorts with lifetimes >30s (i.e. predominantly productive structures), the slope is dictated by the (much faster) speed of movement of an invaginated CCP out of the evanescent field. The similar behavior of LCa-EGFP and σ2-EGFP intensity profiles also suggests that clathrin and AP-2 depart together. In the case of abortive CCPs, this would imply that both coat components disassemble with similar kinetics (at the level of our time resolution). For productive CCPs, it would imply that in the final stages of internalization at P_last, the two proteins move out of the evanescent field together as a coherent unit.

Defining temporal relationships of molecular events at CCPs by dual-channel TIRF

We have identified differences in the intensity time courses of clathrin and AP-2 that reflect mechanisms of assembly, turnover and maturation. However, a direct comparison of the clathrin and AP-2 dynamics shown in Fig. 2D,E is hampered by the fact that these intensity time courses were measured by imaging the two fluorescent markers (LCa-EGFP and σ2-EGFP) independently in single-channel time lapse movies. Thus, there is no direct one-to-one correspondence between the measured CCP lifetimes. As our goal is to analyze multiple markers simultaneously and to establish a hierarchy of their relative timing, a direct correspondence between the signals of different proteins must be established. Therefore, we performed dual-color TIR-FM experiments using LCa-mCherry and σ2-EGFP.

In multi-channel experiments, one channel must be chosen as the reference channel (i.e. master channel), in which objects were detected and tracked, yielding x,y-coordinates and lifetimes of all CCP trajectories; the intensity time courses in all other channels (i.e. slave channels) are then read out at the time points and locations defined by the master channel, regardless of when (or whether) a detectable intensity signal occurred. Given their different maturation time courses, the measured lifetime of an individual CCP may be different depending on whether its clathrin or its AP-2 signal is detected and tracked. However, because (i) detectable AP-2 exclusively marks CCPs at the plasma membrane and (ii) some of the structures visible in the clathrin channel may be non-endocytic, intracellular structures (that contain clathrin but no AP-2), the σ2-EGFP TIR-FM channel was chosen as the endocytic master channel in all following multi-channel experiments. As a result, clathrin time courses measured in these multi-channel experiments represent only true endocytic structures (Fig. 3A) that could be directly compared to the corresponding AP-2 intensity time courses (Fig. 3B,C). Importantly, the choice of the master channel can have an effect on the shape of intensity curves, due to different recruitment kinetics of the proteins, differences in the imaging of the fluorophores, and some inherent filtering effects of the alignment of time courses in the master and slave channels (see Suppl. Fig. 2).

Figure 3. AP-2 is a determinant of clathrin behavior during CCP maturation.

(A) LCa-mCherry time courses, TIRF illumination (slave channel). (B) σ2-EGFP time courses, TIRF illumination (master channel). (C) Time points of segmented reference points and phases in LCa-mCherry TIRF intensity time courses. (D) Time points of segmented reference points and phases of σ2-EGFP TIRF intensity time courses. (E) Example of superimposed LCa-TIRF and σ2- TIRF time courses; time courses are segmented independently to establish temporal hierarchy of reference points and phases in both channels.

As for single-channel experiments, clathrin and AP-2 displayed distinctively shaped intensity profiles (skewed vs. symmetrical; compare Fig. 3A,B to Fig. 2A,B). These data establish that differences in clathrin dynamics relative to AP-2 are not due to contamination of the pool of CCSs by non-endocytic structures. The slower frame rate necessitated by dual-channel acquisition (2s/frame) cannot resolve the very fast dynamics of the initiation and departure phase determined in Fig. 1; therefore, we focused on the length and timing of the growth, plateau, and separation phases. The time points measured in dual-color experiments (Fig. 3C,D) were similar to the single-channel results; the phases showed ‘fractional’ behavior, i.e. their length increased linearly with lifetime. Compared to AP-2, clathrin had a longer growth and plateau phase at the expense of the separation phase. Considering the relative timing of the clathrin and AP-2 maturation process, we observed that plateau, maximum, and separation phases in AP-2 all significantly preceded the corresponding phases in clathrin (Fig. 3C–E). Thus, AP-2 assembly terminates before clathrin assembly. Interestingly, for all lifetimes, the start of the separation phase in AP-2 (green trace in Fig. 3D) preceded the maximum intensity in clathrin by seconds (grey trace in Fig. 3D, also see Fig 3E). This observation suggests that continued clathrin incorporation into CCPs depends on the availability of AP-2, and/or that AP-2 separation is a regulatory cue that slows the further growth of the clathrin lattice.

Clathrin and AP-2 redistribute differentially prior to internalization

The above data suggests that AP-2 recruitment directed the temporal hierarchy of events during CCP maturation and that AP-2 separation significantly precedes clathrin separation. However, it does not answer conclusively how lifetime and maturation relate to final CCP size, or what the observed shifts in separation events between AP-2 and clathrin mean for the termination of CCPs. While AP-2 is a major component of CCPs, the question of its continuous association with CCPs throughout maturation has been controversial; it has been proposed that (i) the kinetics of AP-2 removal from CCPs are similar to those of clathrin (7); (ii) AP-2 completely dissociates from CCPs prior to internalization (2;13); or (iii) AP-2 spatially redistributes on the surface of CCPs without dissociation (4). To resolve differences between these models, we expanded the dual-color TIRF analysis to include information extracted from image acquisition by Epi-illumination.

Using single-color LCa-EGFP labeling with dual acquisition in TIRF and Epi-illumination (with TIRF as the master), we observed that the Epi-illumination intensity time courses differ from their TIRF counterparts (Fig. 4A,B) most significantly during the separation phase, i.e. just before CCP disappearance. For better comparison, representative Epi- and TIRF-intensity time courses were overlaid for one lifetime cohort (Fig. 4C, top panel). The EPI-illumination time courses indicated that clathrin continued to accumulate until quite late in the CCP’s lifetime. When we systematically compared the time points of the intensity maximum in TIRF to the intensity maximum in Epi-illumination across CCP lifetimes (Fig. 4D), we observed that the time point of the TIRF maximum was fractional, while the time point of the Epi-illumination maximum was absolute. In addition, the TIRF and Epi-illumination maxima coincided for CCP lifetimes <30s, and start to diverge for lifetimes >30s (Fig. 4D, blue and red traces, respectively). This result is consistent with the idea that the short-lived CCPs are likely to be abortive, and thus disappear through disassembly (as opposed to internalization), leading to a simultaneous disappearance from the Epi- and TIRF illumination. In contrast, longer-lived CCPs, which are presumably productive and internalized, display a TIRF signal decaying before the Epi-illumination signal, indicative of their inward movement in axial direction with respect to the evanescent field. This suggests that the intensity decrease observed in TIRF during the separation phase represents primarily an axial redistribution of the fluorophore, as opposed to a dissociation of the pit.

Figure 4. Stages of CCP internalization revealed by dual channel TIRF/Epi illumination.

(A) LCa-EGFP time courses, TIRF illumination. (B) LCa-EGFP time courses, Epi-illumination. (C) Representative superimposed TIRF/Epi clathrin time courses. The lower panels show the corresponding relative distances calculated from the TIRF/Epi intensity ratio; the time point of elongation onset P_e is determined as described in the text. (D) Time points of intensity maxima and elongation onset in LCa-EGFP intensity time courses. (E) LCa-EGFP distance time courses calculated from TIRF/Epi intensity ratio aligned to either t_first (brown traces) or t_last (purple traces).

To conclusively answer whether the earlier onset of intensity decay in AP-2 relative to clathrin represents differential fluorophore redistribution or disassembly, the same type of TIRF/Epi analysis was applied to AP-2, and directly correlated to the clathrin signal. We collected 4-channel movies that combined two fluorophores (σ2-EGFP and LCa-mCherry) with image acquisition in both TIRF and Epi-illumination (Suppl. Fig. 3A,B); the σ2-EGFP TIRF channel was used as the master channel. Although the raw data was of lower quality and resolution due to technical issues involved in switching between the four different channels, the results confirmed that for the productive range of lifetimes (i) the intensity maxima in TIRF, for both AP-2 and clathrin, significantly preceded the Epi-illumination maxima (Suppl. Fig. 3C,D), and (ii) in TIRF, the intensity maxima in AP-2 preceded the intensity maxima in clathrin (Suppl. Fig. 3C). These results indicate that the onset of signal decay in the TIRF channel is primarily a result of axial redistribution for both clathrin and AP-2 (4) and that there is a differential redistribution between AP-2 and clathrin at this stage. Interestingly, the intensity maxima in AP-2 also systematically preceded the intensity maxima in clathrin in Epi-illumination (Suppl. Fig. 3D,E). If we take the Epi-illumination intensity maximum to mark the onset of disassembly, then this result indicates that late in the CCP lifetime, AP-2 does in fact start to uncoat before clathrin, which may contribute to the faster decrease of the AP-2 intensity.

Axial redistribution of proteins during CCP internalization

The TIRF/Epi intensity ratios also provide a direct measure of the relative distance of the center of mass of the fluorophore assembly from the TIRF interface (1;4;18). To calculate time courses of the relative axial positions - i.e. the centers of mass of the fluorophore assemblies - throughout the CCP lifetime in the single-color LCa-EGFP TIRF/Epi data (representative distance time course in Fig. 4C, lower panel), we assumed an exponential decay of the evanescent field with penetration depth d_p=100nm in the EGFP channel (see Methods). As the distance measurement involves a ratio of two signals, it is highly susceptible to noise. Thus, we restricted the distance time courses to the ‘detectable’ stretch of the intensity trajectories between t_first and after t_last where the signal-to-noise ratio is sufficient. The distance time courses of the different lifetime cohorts show a small early shift towards the interface just after the appearance of the CCP (Fig. 4E). This is followed by an extended phase of constant distance, where restructuring of the CCP may occur, but where little net axial movement of the center-of-mass of the fluorophore distribution is observed. These behaviors are detected regardless of whether the data are aligned at t_first or t_last (Fig. 4E). Finally, the fluorophore distribution undergoes a strong and sustained shift away from the interface shortly before disappearance, which is consistent with the elongation of the CCV neck prior to internalization, moving the entire CCP into the cytosol. We termed this phase the elongation phase starting at the time point P_e (Fig. 4C, lower panel; the position of P_e is determined computationally in identical manner as points P₁ and P₆). In the range of productive lifetimes, the plot of the time point P_e (i.e. t_e) vs. lifetime is almost parallel to the unity line with a significant offset (Fig. 4D, green line). Assuming a primarily ‘absolute’ type behavior with m=1, the onset of the elongation phase occurs 23.8±2.5s before the disappearance of the CCP. In the corresponding 4-channel data (with LCa-EGFP TIRF master channel), the measured length of the elongation phase was 25.0±3.9s for clathrin and 20.0±2.1s for AP-2 (data not shown). As shown in Figure 4D, the elongation phase started on the order of 10s before the maximum signal in Epi was reached, indicating that CCP growth is finalized while the structure is moving away from the plasma membrane.

Interestingly, the substantial and continuous intensity increase (about two-fold) in maximum EGFP-CLC intensity from shorter to longer CCP lifetimes that we observe by TIRF measurements is less pronounced in the Epi-measurements (compare Fig. 4A, B). In contrast to TIRF, we observe a more modest increase (about 25%) from very short-lived to longer-lived CCPs in Epi, and no strong increase over the range of productive lifetimes. The same trend is observed in 4-channel data for both σ2-EGFP and LCa-mCherry (Suppl. Fig. 3A,B). While this result is consistent with abortive CCPs being smaller than productive CCPs, it also suggests that within the population of productive CCPs, the longer-lived ones are on average not much larger than the shorter-lived ones. This does not preclude the existence of systematic CCP size or lifetime variations with e.g. the concentration of accessory factors or cargo load (8;15), but it indicates that the variations of maximum intensities with lifetime observed in TIRF reflect not only variations in size, but at least partially variations in evanescent illumination intensity that might be due to local variations of relative distance, topology, and/or refractive index.

Summary and Conclusions

Building on our previous studies using automated image and data analysis of CCP dynamics (5;6), we have developed new algorithms to produce weighted-average intensity time courses for CCPs belonging to different lifetime cohorts. These methods extract representative average kinetics of protein recruitment and disassembly that preserves both the fast rates of pit initiation and disappearance, and the heterogeneity between different lifetime cohorts. Our analyses have enabled us to: i) trace the relationship between two CCP components throughout the variable CCP lifetimes, ii) distinguish between fractional and absolute timing in protein recruitment, iii) identify temporal and mechanistic differences for events during CCP maturation depending on CCP lifetime. Together these studies have revealed a stereotypical series of distinct maturation stages of productive CCPs. Using multi-color TIRF and Epi-illumination imaging, we were able to convert these stages into a temporal hierarchy for the recruitment of clathrin and AP-2 to the growing CCP (Fig. 5). Lifetime-independent (‘absolute’) events occur at constant time intervals relative to the start or end of trajectories (characterized by m close to 0 or 1), indicating that variations in length of these phases are random and have little systematic effect on total lifetime. In contrast, ‘fractional’ events or phases reflect the fact that variations in their duration correlates with the lifetime of the CCP. We find that the duration of CCP growth, plateau and separation phases are fractional. In contrast, the onset of clathrin disassembly, the onset of neck elongation, and the fast departure phases are independent of lifetime. This indicates that the internalization machinery functions at a speed that is not significantly correlated to (and thus not a limiting factor for) CCP lifetime. Thus, total CCP lifetime is likely directed by early factors that also direct invagination and the speed of maturation. Future work will build on this classification of fractional and absolute events to systematically identify the molecular factors underlying the broad heterogeneity in CCP lifetime.

Figure 5. Schematic of the temporal hierarchy of molecular dynamics of AP-2 and clathrin.

Our analysis of the relative behaviors of AP-2 and clathrin defines both ‘fractional’ events, i.e. they occur at a fixed fraction of total CCP lifetime, and ‘absolute’ events, i.e. they occur at a fixed time relative to initiation or departure, regardless of CCP lifetime.

Materials and Methods

Cells and image acquisition

Epithelial BSC1 cells (African green monkey kidney cells) stably expressing rat brain clathrin light chain a-EGFP (LCa-EGFP) or the AP-2 rat brain σ2-adaptin fused to EGFP were kindly provided by Dr. T. Kirchhausen. Cells were grown under 5% CO₂ at 37°C in DMEM supplemented with 20 mM HEPES, 10 μg/ml streptomycin, 66 μg/ml penicillin, 10% (v/v) fetal calf serum (FCS, HyClone) and 0.5 mg/ml geneticin (Invitrogen).

For dual color TIR-FM and quadruple channel acquisition (Epi/TIR-F), BSC1 cells stably expressing σ2-EGFP were transfected with LCa-mCherry using Lipofectamine²⁰⁰⁰ (Invitrogen) according to manufacturer’s recommendation 48h before observation.

For TIR-FM, cells were prepared as previously described (5). In brief: the day before imaging, 80,000 BSC1 cells were seeded per well in 6-well plates containing a 22×22 mm glass coverslip. The next day, cells were washed 3 times with PBS, incubated with imaging medium (DMEM without phenol red, supplemented with FCS 2.5%) and mounted on coverslides. For imaging, cells were transferred to a prewarmed microscope stage (37°C; controlled with a custom-modified stage incubator) and 4–7 cells were imaged per coverslip. Single channel TIR-FM movies were acquired in fast (frame rate 400ms, exposure time 90ms) and slow (frame rate 2s, exposure time 150ms) acquisition mode during 10 min, using a 100× 1.45 NA objective (Nikon) mounted on a Nikon TE2000U inverted microscope (Nikon) and a 14-bit mode operated Hamamatsu Orca II-ERG camera. Dual color TIR-FM movies were recorded with a frame rate of 2.0 s and 150 ms exposure time for each channel, whereas EPI-TIR-FM movies were recorded at a frame rate of 2.5 s and 150 ms exposure time for each channel.

Data analysis

CCPs were detected and tracked as described previously (5;6) in LCa-EGFP, LCa-mCherry, and/or σ2-EGFP expressing cells. Matlab software for detection and tracking can be downloaded from http://lccb.hms.harvard.edu/software.html. In dual-channel (or multiple-channel) image acquisition, one channel has to be chosen as the ‘master’ channel in which detection and tracking of CCPs takes place (also see Suppl. Fig. 2), and in which the lifetimes of all individual CCPs are thus determined. In single-channel experiments, the one available channel is automatically used as the master channel for lifetime determination; in LCa/σ2 dual-color experiments, the master channel was σ2 (unless stated otherwise), and in Epi/TIRF dual-channel experiments, the master channel was TIRF. To extract individual CCP intensity time courses, the intensities are then read out in all available channels in the master positions - i.e. at the locations specified by the detected and tracked trajectories in the master channel, using a mask of 2 pixel radius to cover the area of an Airy disc. Intensities were read out starting 15 frames before the first and up to 15 frames after the last detected frame of the master trajectory, using the first (or last) available position in the master channel. All individual intensity time courses were corrected for the local intensity background, which is determined by averaging over image pixels in a ring around each individual CCP for the same time window. As described in the text, after background correction the individual (noisy) CCP intensity time courses were binned into lifetime cohorts, and each cohort underwent weighted averaging to produce a single cohort intensity time course. The local background correction was sufficient to strongly reduce of the spatial variation in intensity between individual CCPs; since the resulting intensities were similar between cells in our experiments, it was sufficient in our case to average the single cohort intensity time course between cells. Matlab software for automatic readout of intensity time courses, grouping into lifetime cohorts, and weighted averaging can be downloaded from http://lccb.hms.harvard.edu/software.html.

These final time courses were then automatically segmented into different phases by finding ‘landmark’ points, which are positions where strong changes in curvature occur; these points are the positions that maximize the area of a triangle spanned by this point and adjacent reference points. For example, P₁ is the point on the curve that maximizes the area of the triangle between P₀P₁P_first (see Suppl. Fig. 1). The timing of the landmark points within the segmented cohort intensity time courses, and the duration of the characteristic phases between reference points, were then compiled in time point-vs-lifetime plots, as illustrated in Fig. 2c, to distinguish absolute from fractional events.

TIRF/Epi intensity ratios were converted into relative distances assuming an exponential decay of the evanescent field (e.g. (1)):

$$\text{d}[\text{in}\text{units}\text{of}\text{the}\text{penetration}\text{depth}{\text{d}}_{\text{p}}]=-ln({\text{I}}_{\text{Tirf}}/{\text{I}}_{\text{Epi}}).$$

where I_Tirf and I_Epi were the fluorescence intensities above the local background. The exact in vivo penetration depth was unknown, but we estimated d_p(green) at 100nm and d_p(red) = d_p(green)*(λ_red/λ_green) = 116nm with λ_red=568nm and λ_green =488nm. The landmark point P_e in the distance time course (see Fig. 4c), which marks the onset of the elongation phase, is determined computationally in identical fashion as the landmark points in the intensity time courses (see Fig. 1 and Suppl. Fig. 1) as the position of highest curvature change; specifically, P_e marks the position on the time course that maximizes the area of the triangle between points P_first, P_e and P_last.

Supplementary Material

Supp Fig S1. Supplementary Figure 1: Definition of temporal landmarks in average intensity time courses.

(A) Cohort intensity time courses are segmented into phases based on positions where strong changes in curvature occur. The positions are determined computationally as the positions that maximize a mathematical value related to the area of a polygon between neighboring reference points and the search point. For a given average intensity time course, the first and last detection time points (P_first and P_last), the points designating the intensity baseline (P₀ and P₀₀) and the maximum intensity time point (P_max) are fixed (marked in red). P₁ and P₆ (marked in orange) are calculated as the points on the curve that maximize the value area(P₀P₁P_first) and area(P_lastP₆P₀₀), respectively (triangles marked in orange). P₂ and P₅ (marked in green) are calculated as the points on the curve that maximize the values area(P₀P₁P₂)/dist(P₂-P₁) and area(P₀₀P₆P₅)/dist(P₅-P₆), respectively (triangles marked in green). Finally, P₃ and P₄ (marked in light blue) are determined as the points on the curve that maximize area(P₂P₃P₄P₅) (i.e. the area of the trapeze marked in cyan).

Supp Fig S2. Supplementary Figure 2: Effects of assigning ‘emaster’ and ‘eslave’ channels in dual-channel data acquisition.

In dual-channel (or multi-channel) acquisition, intensities are read out in all channels at the positions and time points determined by the ‘master’ channel, in which detection, tracking and lifetime assignment takes place. In dual-color σ2-EGFP/LCa-mCherry movies, the choice of master channel has a significant effect on the shapes of the clathrin and AP-2 intensity time courses. (A) LCa-mCherry time courses, TIRF illumination (master). (B) σ2-EGFP time courses, TIRF illumination (slave). Compare to Fig. 3A,B where master/slave roles are reversed. The differences in shape are partly due to the different properties of the fluorescent proteins themselves, but they are also influenced by an ‘alignment filtering’ inherent in the selection of the master/slave channels, which is illustrated in the following simulation (C): We generated a set of 1000 50sec-lifetime intensity time courses (framerate 2sec) that start at t=0 and end at t=50s by perturbation of a ground truth intensity (blue curve) with Gaussian noise of SD=0.3 (green curves). (D) If we average these noisy time courses in their perfectly aligned state, we reproduce the ‘reconstructed ground truth’ (Panel D, blue ‘ground truth’ curve). In practice, however, the noisy intensity time courses are not aligned perfectly to their ‘true’ start and end points; rather, there is some variation in the temporal alignment. In the master channel, the individual intensity traces are aligned based on their first and last detectable time point. This means that the first time point at which the intensity exceeds the detection threshold is automatically identified as t=0. When we selectively re-align the 1000 noisy individual time courses according to this principle (i.e. they are aligned not to their true first/last time point, but to the first/last time point at which the intensity exceeds the detection threshold), the intensity time courses are shifted relative to their true first time point up to several frames. The average of this selectively realigned dataset (Panel D, yellow ‘master channel’ curve) is shortened compared to the ‘true’ average, and it appears sharpened in the vicinity of the first and last time point due to this intensity selection. In a second step, we generated 1000 slave intensity time courses, using the same ground truth and the same noise level. One slave time course is associated with each master time course, and the slave channel follows the realignment dictated by the noisy master intensities. Since the intensity noise in the master and slave channel are similar in magnitude but mutually independent, the alignment shifting in the slave channel will appear random (i.e. unrelated to intensity variations in the slave channel). When we re-align the 1000 noisy slave time courses according to this second principle, the resulting new average (Panel D, red ‘dependent channel’ curve) is broadened compared to the ‘true’ average, and details are smeared out.

Supp Fig S3. Supplementary Figure 3: Clathrin and AP-2 dynamics obtained using two-color, dual mode (TIRF/Epi) illumination.

(A) LCa-mCherry and σ2-EGFP (master) time courses, TIRF illumination. (B) LCa-mCherry and σ2-EGFP time courses, Epi-illumination. (C) Time points of TIRF intensity maxima in 4-channel LCa-EGFP/σ2-EGFP TIRF/Epi movies. (D) Time points of Epi-illumination intensity maxima in 4-channel LCa-EGFP/σ2-EGFP TIRF/Epi movies with AP-2 TIRF master channel. (E) Time points of Epi-illumination intensity maxima in 4-channel LCa-EGFP/σ2-EGFP TIRF/Epi movies with clathrin TIRF master channel. (F) LCa-mCherry relative axial distance time courses. For purple traces, time zero corresponds to the last detectable time point of the trajectories; for brown traces, time zero corresponds to the first detectable time point.