Molecular Brightness Analysis Reveals Phosphatidylinositol 4-Kinase IIβ Association with Clathrin-Coated Vesicles in Living Cells

Abstract

Mammalian cells express two classes of phosphatidylinositol 4-kinase (PI4K), designated as Types II and III, that phosphorylate phosphatidylinositol to generate PI4P. A number of studies have indicated that these enzymes are important for Golgi trafficking and both early and late stages of endocytosis. In this study, we focus on PI4KIIβ, a protein that is evenly distributed between membrane and soluble fractions, and is believed to participate in stimulus-dependent phosphoinositide signaling. Using molecular brightness analysis, we found that EGFP-tagged PI4KIIβ exists as two distinct species in the cytoplasm: a soluble monomer and a high-order complex enriched with multiple copies of PI4KIIβ. This observation was confirmed by an autocorrelation analysis that identified two species with distinct mobilities. We further demonstrate that the high-order complex enriched with PI4KIIβ is sensitive to inhibition of palmitoylation, indicating that it is associated with membranes, very likely vesicles. Indeed, we show that the high-order PI4KIIβ complex is sensitive to expression of dynamin 2 (K44A), a dominant-negative inhibitor of endocytosis. Using dual-color heterospecies partition analysis, we directly detected that PI4KIIβ comoves with clathrin light chain on vesicles. This analysis allows us to isolate the comobile species in the presence of strong background contribution from the monomeric pool of PI4KIIβ. Our results strongly suggest that PI4KIIβ is involved in an early stage of endocytosis and is associated with clathrin-coated vesicles. Moreover, we establish molecular brightness as a powerful tool for characterizing cellular cytosolic vesicles that are otherwise difficult to characterize by other techniques.

Introduction

Phosphatidylinositol 4-kinases (PI4Ks) catalyze the phosphorylation of phosphatidylinositol (PI) to generate PI 4-phosphate (PI4P), the major precursor in biosynthesis of the polyphosphoinositides PI(4,5)P₂ and PI(3,4,5)P₃, which are critical elements in numerous cell signaling and membrane trafficking pathways (1,2). Mammalian cells express two classes of PI4K, termed PI4KII and PI4KIII, each of which contains α and β isoforms (3–6). PI4KIIIα (∼230 kDa) and PI4KIIIβ (∼110 kDa) are peripheral membrane proteins that bear little sequence similarity to the Type II kinases (7). PI4KIIα and β are ∼55 kDa proteins that are anchored to membranes by palmitoylation of multiple cysteines in a motif (CCPCC) located within their highly conserved catalytic domains (8–10). Palmitoylated pools of PI4KII behave as integral membrane proteins, requiring detergent for extraction from membranes. Nearly all of PI4KIIα is acylated and hence integrally membrane-bound, whereas PI4KIIβ is almost evenly divided between membrane-bound and soluble pools, and only ∼50% of the membrane-bound pool requires detergent for extraction (11,12).

PI4KIIIβ is abundant in the Golgi apparatus and has been shown to play important roles in Golgi budding (13), endocytic recycling (14), and cytokinesis (15). PI4KIIIα is enriched in the endoplasmic reticulum, but its primary function may be to generate precursor for the synthesis of agonist-sensitive pools of PI(4,5)P₂ on the plasma membrane (16,17). PI4KIIα is also highly enriched in the Golgi apparatus, where it generates a pool of PI4P involved in recruiting the clathrin adaptor protein, AP-1 (18). It has also been implicated in the late stages of endocytosis, as RNA interference-mediated depletion of PI4KIIα inhibits lysosomal degradation of the EGF receptor but not its internalization from the plasma membrane (19). PI4KIIβ, the subject of this investigation, is the least understood of the PI 4-kinases, and no function has yet been assigned to this isoform. It is found on the Golgi apparatus and in cytoplasmic punctae, some of which costain with endocytic markers such as EEA1 and Rab5 (9,20). A portion of PI4KIIβ has been shown to redistribute from the cytosol to membranes, including the plasma membrane, in response to growth factor treatment, expression of a constitutively active Rac mutant, or inhibition of its interaction with heat-shock protein 90 (10,21).

In this study we used fluorescence correlation spectroscopy to characterize the behavior of green fluorescent protein (GFP)-tagged PI4KIIβ in the cytoplasm (22,23). Using molecular brightness analysis (24,25) and autocorrelation analysis (26), we identified a fraction of PI4KIIβ that is highly enriched in small molecular complexes. We devised an analytical technique, termed brightness spike analysis, to spectroscopically isolate this species from its monomeric counterpart. Although this technique is qualitative in its current form, it is able to monitor changes in bright-species abundance. Furthermore, we probed protein complexes using a dual-color approach (27), specifically by using a newly developed dual-color heterospecies partition (HSP) analysis (28). We were able to establish that PI4KIIβ codistributes with clathrin-containing vesicles. Taken together, our results indicate that PI4KIIβ is likely to participate in an early endocytic pathway that involves clathrin-coated vesicles.

Materials and Methods

Experimental setup

A mode-locked Ti:Sapphire laser (Tsunami; Spectra-Physics, Mountain View, CA) pumped by an intracavity doubled Nd:YVO₄ laser (Millenia; Spectra Physics) served as a source for two-photon excitation. The experiments were carried out with a modified Zeiss Axiovert 200 microscope (Thornwood, NY) as previously described (25,28). Two-photon measurements were taken for 60 s with a 63x Plan Apochromat oil immersion objective (N.A. = 1.4) at a sampling frequency of 20 kHz. The excitation power measured at the objective ranged from 0.8 to 1.5 mW at 1000 nm. These low powers were chosen to avoid saturation (29).

Cell lines and expression vectors

U2OS cells were obtained from American Type Culture Collection and maintained in 10% fetal bovine serum and DMEM media. Transfection was carried out using TransFectin reagent (Bio-Rad, Hercules, CA) according to the manufacturer’s instructions 24 hr before measurements were made. For the sake of brevity, we refer to mCherry as mCh when naming protein constructs. GFP-PI4KIIβ was constructed by inserting Homo sapiens PI4KIIβ amplified from polymerase chain reaction (PCR) into the pEGFP-C1 plasmid (Clonetech, Mountain View, CA). Clathrin-mCh was constructed by inserting Clathrin light chain into pmCherry-N1 plasmid. The PCR fragment encoding human adaptor-related protein complex 2 (AP-2) subunit mu1 was inserted into the mCherry-C1 vector to construct mCh-AP-2. The mCherry-labeled dynamin 2 dominant-negative mutant (Dyn2(K44A)-mCh) was constructed in two steps by first inserting rat Dyn2 into the mCherry-N1 vector to construct Dyn2-mCh, and then by converting lysine 44 in Dyn2 to alanine with QuikChange site-directed mutagenesis. All sequences were verified by automated sequencing.

Data analysis

Single-channel experimental data were analyzed with Q-analysis, where the brightness is determined at each second, ε_segment (24,30). The resulting normalized brightness, b, was calculated by taking the average brightness of ε_segment divided by the monomer GFP brightness that was independently calibrated at the beginning of the experiments, b=<ε_segment >/ε_monomer. The same raw data were also used to calculate the autocorrelation function, g(τ) (26). Experimental autocorrelation functions, g(τ), were fitted to theoretical functions using a three-dimensional (3D)-Gaussian beam profile to recover the residence time. Dual-color HSP analysis was performed according to our previously published method (28,31). HSP separates fluctuations that are relevant for heterointeractions from the fluctuations from proteins that are not relevant to heterointeractions. This approach provides a direct and robust identification of the degree of binding between two proteins. In addition, protein coexpression ratios were determined for every cell (32).

Results

Clustering of PI4KIIβ on cytoplasmic structures

Our first objective was to use molecular brightness analysis to monitor the oligomerization of GFP-PI4KIIβ in the cytoplasm of transfected U2OS cells. Fig. 1 A is a plot of the normalized brightness, b, as a function of GFP-PI4KIIβ concentration. Each data point represents a single-cell measurement. Surprisingly, the normalized brightness was not dependent on concentration, and instead showed considerable scattering from cell to cell. We plotted the normalized brightness according to the measurement sequence. The particular measurement set shown in Fig. 1 B comprises a total of 23 cells that were randomly picked for examination. As expected, the normalized brightness b was randomly scattered. This type of experiment was repeated many times. We also performed experiments at 37°C and observed no difference in molecular brightness from experiments performed at room temperature.

Figure 1.

(A and B) Representative molecular brightness titration data of GFP-PI4KIIβ as a function of protein concentration (A) or measurement sequence (B). All of the measurements were performed in the cell cytoplasm. A total of 23 cells were randomly selected and measured. The normalized brightness, b, is calculated as described in Materials and Methods.

To understand the origin of the cell-to-cell brightness scatter, we examined many individual data traces. Results obtained from cells 17 and 18 are shown as representative cases. Fig. 2 A displays the intensity trace of cell 17 with a time resolution of 1 s. The intensity trace from this cell was stable as a function of time and resembled the trace of a typical GFP measurement from cells. The resulting ε_segment, displayed in Fig. 2 B, gave an average of 2020 counts per second per molecule (cpsm). Because the calibration with monomeric GFP gave a brightness of 2000 cpsm (solid red line in Fig. 2 B), the normalized brightness from this particular cell indicates that PI4KIIβ is monomeric. We also examined cell 18, which yielded a normalized brightness of 3.9 (Fig. 1 B). Although its intensity trace was less stable than that of cell 17, it was still within the variations observed from cytoplasmic measurements (24). However, the ε_segment (Fig. 2 D) exhibited a number of brightness spikes with values that clearly exceeded monomeric GFP brightness (again displayed as a solid red line). As is evident in Fig. 2 D, the brightness spikes were often followed by monomeric GFP brightness values, indicating that the sample contained some bright, albeit rare, species coexisting with monomeric GFP-PI4KIIβ. In the absence of these bright species, the molecular brightness corresponded to the value for monomeric GFP, whereas in the presence of the bright species the ε_segment value fluctuated greatly, and even spiked to 50 kcpsm in one instance. The amplitude of the brightness spike varied, reflecting the stochastic nature of the diffusion path through the nonhomogeneously illuminated excitation volume created by the microscope optics. A particle passing through the center of the illuminated volume ignites a much stronger burst of photons than a particle passing along its edge. PI4KIIβ is believed to be involved in endocytic trafficking pathways, and our brightness analysis suggests that PI4KIIβ may associate with vesicles. If multiple copies of GFP-PI4KIIβ occupy a single vesicle, its brightness would be high, as multiple copies of the labeled kinase on a single vesicle would emit a much higher number of photons than an individual monomer of GFP-PI4KIIβ.

Figure 2.

Fluorescence intensity data and their molecular brightness analysis for cells 17 and 18. (A) Fluorescence intensity time trace of cell 17. (B) ε_segment analysis of the same intensity trace of cell 17. (C) Fluorescence intensity time trace of cell 18. (D) ε_segment analysis of the same intensity trace of cell 18. ε_segment analysis reveals that virtually all of the GFP-PI4KIIβ proteins exist as a monomer in cell 17, and bright but rare species are present in cell 18. Solid red line indicates monomeric GFP brightness, and dashed line is cutoff brightness value.

We calculate conventional brightness values by averaging the ε_segment over complete measurements (60 s), so brightness spikes are effectively averaged out when a few bright objects coexist with monomeric protein, as is the case for cell 18. Consequently, the normalized brightness b fails to reveal the existence of bright but rare complexes. Clearly, a better way to characterize the heterogeneity of the GFP-PI4KIIβ sample was needed, and we turned to brightness values in a segment as an alternative analytical approach. For example, the number of spikes in the brightness value as a function of second reveals the heterogeneity of the sample and the concentration of bright species. Because the number of spikes depends on the total measurement time, we introduced another parameter, the spike count rate (min⁻¹), to describe the heterogeneity of the GFP-PI4KIIβ sample. To determine the spike count rate, we must use a cutoff brightness that eliminates any brightness points below the cutoff value. Obviously, the value of the cutoff brightness needs to be chosen judiciously. If the cutoff brightness is too close to the monomer GFP brightness, then clean separation between the dim monomeric species and the bright species is not possible. If the cutoff brightness value is too high, then many brightness spikes will be missed. To identify the optimal cutoff value, we systematically increase the cutoff values and identify a plateau region where neither too many nor too few events are accepted or rejected. For example, 8 kcpsm is a factor of 4 times higher than the monomeric GFP brightness and is in the plateau region. Therefore, in this study we used a cutoff brightness of 8 kcpsm to calculate the spike count rate (Fig. 3 A). We also plot the average brightness ε_spikes of the spikes (Fig. 3 B). The brightness ε_spikes is calculated by averaging ε_segment of spikes identified by the cutoff filter. The value of ε_spikes varies from 20 kcpsm to 60 kcpsm across cells. Because the values of ε_spikes do not change significantly from cell to cell, the change in the normalized brightness b is mainly due to the spike count rate.

Figure 3.

Spike count rate (min⁻¹) (A) and ε_spikes (B) of GFP-PI4KIIβ from the same data used for Fig. 1. The spike count rate (min⁻¹) is the number of events detected above the cutoff brightness value at a given minute. All of the data were taken for 60 s. Therefore, the maximum spike count rate (min⁻¹) is 60 and the minimum is 0. ε_spikes represents the average segment molecular brightness values of spikes identified by the cutoff filter. The spike count rate (min⁻¹) effectively enhances the contrast between complexes that contain multiple copies of GFP-PI4KIIβ from its monomeric fraction.

To that the spike count rate is a suitable indicator of the sample concentration of a bright species, we conducted a fluorescent sphere dilution experiment (Fig. S1 in the Supporting Material). Fluorescent spheres mimic complexes that carry multiple fluorescent proteins, because each sphere is bright and carries multiple fluorophores. We diluted 50 nm fluorescent spheres from an initial concentration of 200 pM. The spike count rate was linear at low concentrations (<100 pM), but deviated from linearity at high concentrations. The nonlinearity very likely is due to the multiple events that occur very close in time at high concentrations. Nevertheless, even at high concentration, the spike count rate still increases as a function of concentration. The spike count rate is therefore a suitable indicator of the sample concentration, and will be displayed instead of the normalized brightness (b hereafter).

We also performed an autocorrelation analysis on these data to characterize the mobility of PI4KIIβ in cells. For cell 17, we were able to fit the autocorrelation curve with a single-species model (data not shown) and recovered a residence time τ_D1 of 3.7 ms (Table 1). The autocorrelation curve for cell 18 could not be described by a single-species model, and a two-species model was needed to fit the data. We recovered residence times of τ_D1 = 2.6 ms and τ_D2 = 117 ms for each of the two species (Table 1). There is a clear correlation between the presence of two species in the autocorrelation function and a high spike count rate. For this data set of 23 cells, 20 cells required a two-species fit, and τ_D1 and τ_D2 were recovered. The remaining three cells (cells 7, 16, and 17) were described by a single-species model yielding a single residence time, τ_D1. A close examination of Fig. 1 B reveals that these three cells have normalized brightness very close to one. Fig. 4 displays the histogram of residence times, τ_D1 and τ_D2. The average residence time of τ_D1 is 2.7 ms with a relative standard uncertainty of 44%, and the average residence time of τ_D2 is 69 ms with a relative standard uncertainty of 50%. τ_D1 is very likely to be the residence time of monomeric PI4KIIβ, whereas τ_D2 is likely to represent the high-order complexes enriched with multiple copies of PI4KIIβ. For comparison, the average residence time of GFP is 0.8 with a standard uncertainty of 38% (Table S1). The great difference (a factor of 20) between τ_D1 and τ_D2 of PI4KIIβ suggests that the high-order complex is much bigger in size and potentially vesicle membrane bound.

Table 1.

Autocorrelation analysis of cells 17 and 18

Cell number	τD1 (ms)	τD2 (ms)
Cell 17	3.7
Cell 18	2.6	117.4

Autocorrelation function of cell 17 can be described by a single-species model, and recovered a residence time τ_D1 of 3.7 ms. The autocorrelation function of cell 18 requires a two-species model to describe the data. Residence times of τ_D1 = 2.6 ms and τ_D2 = 117 ms were recovered for each of the two species.

Figure 4.

Distributions of τ_D1 and τ_D2 recovered from the autocorrelation analysis of all 23 cells. Twenty cells required a two-species fit. The average residence time of τ_D1 is 2.7 ms with a relative standard uncertainty of 44%, and the average residence time of τ_D2 is 69 ms with a relative standard uncertainty of 50%.

To determine whether PI4KIIβ clustering is favored by its association with membranes, we took advantage of our previous observation that it is anchored into membranes by virtue of palmitoylation of a cysteine-rich motif (CCPCC) within the catalytic domain (11). We examined the effect of palmitoylation on PI4KIIβ brightness using a palmitoylation-defective mutant, PI4KIIβ-FFPFF, in which all four cysteines of the palmitoylation motif were replaced by phenylalanines. Fig. 5, A and B, display the spike count rate values that resulted from a brightness spike analysis of the wild-type (WT) PI4KIIβ and the PI4KIIβ-FFPFF mutant under otherwise identical experimental conditions. For a total of 20 cells measured, far fewer cells expressing PI4KIIβ-FFPFF displayed brightness spikes compared with those expressing WT PI4KIIβ. Even in PI4KIIβ-FFPFF-expressing cells that showed brightness spikes, the average value of the spike count rate was considerably lower relative to those expressing the WT enzyme.

Figure 5.

(A–C) Spike count rate (min⁻¹) of cells expressing GFP-PI4KIIβ-FFPFF (B) and GFP-PI4KIIβ coexpressed with Dyn2(K44A)-mCh (D); each experiment is also shown with a control experiment of GFP-PI4KIIβ under otherwise identical experimental conditions (A and C). A total of 20 cells were measured in each case. The spike count rate (min⁻¹) was greatly reduced in both samples, indicating a defect in forming GFP-PI4KIIβ-enriched complexes.

Association of PI4KIIβ with clathrin-coated vesicles

PI4KIIs have been implicated in endocytosis and Golgi budding (9,10). Therefore, we sought to determine whether the PI4KIIβ-containing structures could be endocytic or secretory transport vesicles. To inhibit the formation of transport vesicles, we used the dominant-negative K44A mutant of Dyn2, a ∼100 kDa GTPase required for both endocytosis and Golgi budding (33,34). Fluctuation analysis was carried out on cells cotransfected with GFP-PI4KIIβ and Dyn2(K44A)-mCh. Only cells that expressed both vectors were measured. As is evident from the plot of spike count rate in Fig. 5 D, PI4KIIβ-containing vesicles were virtually absent from cells that also express Dyn2(K44A), suggesting that these vesicles originate from a dynamin-dependent trafficking pathway. Under otherwise identical experimental conditions, WT PI4KIIβ coexpressed with mCherry displayed spike-count-rate values similar to those shown in Figs. 3 A and 5 A (Fig. 5 C).

Because many dynamin-dependent transport processes are mediated by clathrin, we wished to test whether the spike counting analysis could successfully identify clathrin-coated vesicles. Fig. S2 displays a plot of the spike count rate of clathrin-GFP transfected cells. Indeed, many brightness spikes were identified, indicating the presence of vesicles enriched with clathrin-GFP. We also transfected cells with GFP-AP-2, an accessory protein that also participates in receptor-mediated endocytosis from the plasma membrane. Fig. S3 displays the result of the spike count rate of AP-2. Overall, the spike count rate for AP-2 was less than that for clathrin-GFP, but was nevertheless discernible with the spike counting analysis. We also applied autocorrelation analysis to the same data sets. The resultant residence times are displayed in Table S1. Among 40 cells expressing clathrin, 31 cells required two-species fitting. The resultant τ_D1 was 2.2 ms with a standard error (SE) of 50%, and τ_D2 was 86.2 ms with an SE of 55.8%. Among 35 cells expressing AP-2, 24 cells required two-species fitting. We recovered a τ_D1 of 2.1 ms with an SE of 48%, and τ_D2 of 80.4 ms with an SE of 68.8%.

We next turned to dual-color HSP analysis to spectroscopically test whether PI4KIIβ is associated with clathrin-coated vesicles. We previously used HSP analysis to obtain binding constants for hetero-protein interactions (28). Here we sought to apply HSP analysis to determine whether two proteins are simultaneously present on the same vesicle. HSP analysis returns the HSP-brightness vector, b = (b_green, b_red), with b_green and b_red representing the brightness of the green and red channels, respectively. Each HSP brightness corresponds to a point on a 2D brightness plot (Fig. 6 A). HSP values that fall on the green line represent vesicles containing only green protein. We refer to this line as the green species line. The slope in the line is simply due to the leakage of green species brightness into the red channel. Any other brightness values fall into a shaded area representing vesicles that carry both green and red proteins, and we call this area the green/red comobile zone.

Figure 6.

HSP analysis of cells that cotransfected with green and red fluorescently labeled proteins. (A) Dual-color brightness. The green line represents all possible brightness vectors that contain green species only. The gray-shaded zone represents all possible brightness vectors that contain both green and red species. (B) HSP-brightness vectors for cells coexpressing GFP-PI4KIIβ and mCh. (C) HSP-brightness vectors for cells coexpressing GFP-PI4KIIβ and clathrin-mCh. (D) HSP-brightness vectors for cells coexpressing GFP-PI4KIIβ and mCh-AP-2.

We examined the ability of the HSP analysis technique to isolate green-containing species by performing GFP-PI4KIIβ and mCh cotransfection experiments. Cells that coexpress both proteins were identified and measured. The HSP-brightness vector was calculated as described in the Supporting Material and displayed in Fig. 6 B. We expect that the HSP brightness should only fall on the green species line, because mCh is not supposed to associate with PI4KIIβ-containing vesicles. Indeed, the returned HSP-brightness vectors are all on the green species line. The b_green value varies from monomeric GFP values to high multimer values, indicating the presence of GFP-PI4KIIβ- enriched species, whereas the b_red value is simply proportional to the b_green value due to nonideal separation of green and red channels (Fig. 6 B). Fig. 6 B clearly demonstrates that the HSP analysis has the capacity to isolate green-associated species from green and red mixtures. Next, we examined the relationship between PI4KIIβ- and clathrin-containing vesicles. Cells cotransfected with GFP-PI4KIIβ and mCh-tagged version of clathrin light chain (clathrin-mCh) were examined. The HSP-brightness vectors were calculated from cells expressing both proteins, and are displayed in Fig. 6 C. The HSP analysis clearly detected green/red comobile vesicles, demonstrating that GFP-PI4KIIβ and clathrin-mCh reside together on the same vesicle. Finally, we tested whether PI4KIIβ comoves with a clathrin adaptor protein, AP-2, which is known to participate in receptor-mediated endocytosis from the plasma membrane. Fig. 6 D shows the HSP-brightness vector of cells cotransfected with GFP-PI4KIIβ and mCh-AP-2. Again, the HSP brightness varied from cell to cell, but overall was located in the green/red comobile zone. Taken together, these results strongly suggest that a pool of PI4KIIβ distributes to endocytic clathrin-coated vesicles.

Discussion

Molecular brightness in fluorescence fluctuation spectroscopy has become an important tool for studying protein oligomerization and interactions in vitro and in cells (23). When purified components are studied in vitro, the molecular brightness reveals the oligomerization state of the protein. When applied in cells, if the fluorescently labeled protein interacts with endogenous unlabeled proteins or is controlled by processes such as posttranslational modifications, the molecular brightness becomes a description of the copy number of the particular complex examined. Therefore, one must exercise caution when describing the oligomerization of fluorescently tagged protein in cells where endogenously unlabeled partners or posttranslational modification processes are not completely characterized. One of the strategies we use to deal with these issues is to perform concentration titration experiments. The concentration of the fluorescently labeled protein is determined by fluctuation measurements. If the oligomerization of this protein is driven by self-association, we would then expect the molecular brightness to follow the concentration of the expressed protein (25,28,32). In the past, we successfully applied titration experiments on nuclear receptors in cells where the concentration of endogenous receptor was negligible. In these cases, we were able to directly translate molecular brightness into protein oligomerization.

When we sought to characterize GFP-PI4KIIβ in cells, we encountered a different situation. As demonstrated in Fig. 1, the normalized GFP-PI4KIIβ brightness displayed no concentration dependence, and instead was scattered across all concentrations examined. The lowest brightness appeared to be a unit of one, which indicates that GFP-PI4KIIβ can exist as a monomer. Numerous cells had brightness units greater than one. A close examination of the cells with higher brightness revealed that the ε_segment for these cells tended to be unstable and contained a mixture of monomer and high-brightness species (Fig. 2 D). In theory, data analysis techniques such as photon-counting histogram and fluorescence intensity distribution analysis can be used to quantify these species (35,36). In practice, these techniques require stringent conditions. For example, quantitative fluorescence fluctuation analysis is based on the assumption of a stationary signal. Drifts in intensity, such as those caused by large particles passing through the observation volume, lead to a nonstationary signal, which can introduce biases into the data analysis (24). This is especially relevant for the quantitative resolution of two or more species, which requires excellent signal conditions. If the application does not require detailed quantification, brightness peak analysis offers a robust alternative. We slightly modified an existing analysis method by applying a cutoff filter to selectively examine high-brightness species while simultaneously eliminating the contribution of monomeric GFP-PI4KIIβ.

Examination of heterogeneity with intensity spikes has been applied in a number of approaches, each of which takes advantage of species that display a clear intensity contrast (37,38). In the case of PI4KIIβ, the major contribution to intensity comes from the monomer species. The monomer species contributes a very stable intensity trace (Fig. 2 A) and obscures the intensity spikes that would otherwise be visible (Fig. 2 C). Thus, we had to introduce a brightness spike analysis to differentiate species by their brightness contrast. This method recovers two parameters, the spike count rate and the average brightness of the spikes. The spike count rate is related to the concentration of vesicles and was used in this work to follow the GFP-PI4KIIβ-containing vesicle fraction. The spike count rate depends on the integration time. We arbitrarily chose 60 s to reflect our typical measurement time. With this method, we are able to obtain the spike count rate (min⁻¹). The value of the spike count rate ranges from zero to 60. A value of zero indicates no spike and thus no high-order complexes visualized within the measurement time. A value of 60 means that high-order complexes are present in the excitation volume at every second.

Brightness spike analysis is ideal for applications in which a few bright complexes are embedded in a sea of low-brightness molecules and brightness spikes are easily identifiable (Fig. 2 D). Here, we use the qualitative relation that a decreasing spike count rate indicates a lower concentration of bright particles. In Fig. S1 we demonstrate that samples with low concentrations will have fewer brightness spikes than samples with high concentrations at a given cutoff brightness. We used this strategy to probe the effect of manipulating the concentration of PI4KIIβ-enriched vesicles by either mutating the PI4KIIβ palmitoylation sites or coexpressing with a dominant-negative Dyn2 mutant. Compared with controls, we observed a drastic reduction of spike count rate, a clear indication that the number of PI4KIIβ-enriched vesicles was reduced under these experiment conditions (Fig. 5). The exact concentration of complexes is not important for the work at hand, because our goal was only to demonstrate the presence of complexes under normal conditions and their absence in response to inhibitory interventions. In principle, a more quantitative theory can be developed, but this would have to account for the actual shape of the observation volume. A bright particle diffusing through the periphery of the observation volume would appear to be very dim, whereas a dim species diffusing through the center of observation volume would appear to be bright.

The overall concentration of expressed GFP-PI4KIIβ in our cells was relatively low, on the order of hundreds of nanomolars, suggesting the existence of a specific mechanism to enrich GFP-PI4KIIβ in the bright complexes. A potential mechanism for this enrichment would be concentrating PI4KIIβ on the surface of vesicles. Indeed, the other member of the PI4KII family, PI4KIIα, has been shown by conventional immunofluorescence microscopy to be enriched in vesicles (19,39). The existence of the bright species was intriguing, because the majority of GFP-PI4KIIβ exists as monomer in the cytosol. The intensity contributed by the majority of GFP-PI4KIIβ monomeric molecules precluded the characterization of GFP-PI4KIIβ-containing vesicles by intensity-only, whole-cell imaging methods. Techniques that allow examination of a very thin layer of the cell, such as total internal reflection fluorescence, are necessary to spatially isolate vesicles from monomeric background (40,41). Unfortunately, total internal reflection fluorescence is only applicable for examining vesicles bound to the plasma membrane. To examine vesicles that are highly mobile in the cytoplasm, fluctuation techniques with z-sectioning ability are among the best choices available.

The residence time recovered from autocorrelation analysis serves as further evidence that PI4KIIβ is associated with complexes of a size that are on the order of small vesicles (42,43). Two-species analysis is always needed when there are brightness spikes in the segmental brightness analysis. The recovered residence times, τ_D1 and τ_D2, differed by >20-fold (2.7 ms vs. 69 ms). According to the Stokes-Einstein relation, the ratio of the diffusion time of two samples is equal to the ratio of the hydrodynamic radii of the diffusing particles, τ_D1/τ_D2 = r₁/r₂ (26,44,45). For untagged GFP, the residence time is ∼0.8 ms in U2OS cells (data not shown). The Stokes radius of GFP has been determined to be 2.82 nm (46,47). Using the Stokes-Einstein relationship, we estimated that the average Stokes radius of the first species is 9.5 nm, whereas the average Stokes radius of the second species is 240 nm. The average Stokes radius of the first species is two times larger than what would be predicted from its molecular weight. This is not surprising given that proteins are known to interact with other cellular components. The average Stokes radius of the second species is also unlikely to be accurate, because the cellular environment is known to be crowded and large particles (>30 nm) are more likely to be confined, and thus gives an apparent larger residence time than when unconfined (48). Thus, 240 nm will be an absolute upper limit of the particle; its actual size is likely to be smaller than the size estimated using an unconstraint diffusion model (49,50). Nevertheless, the second residence time, τ_D2, is considerably different from the residence time of the first species. If the first species represents the residence time of monomeric GFP-PI4KIIβ, then the residence time of the second species is consistent with that of a vesicular structure with GFP-PI4KIIβ enriched on the membrane.

Based on our finding that expression of a dominant-negative dynamin mutant suppressed the formation of GFP-PI4KIIβ brightness spikes, we suspected that the PI4KIIβ-rich particles could be clathrin-coated vesicles (although Dyn2 is also involved in some forms of clathrin-independent endocytosis). To test this possibility, it proved necessary to apply a newly developed dual-color brightness analysis technique, HSP analysis, which allowed us to spectroscopically isolate comixing species. For example, HSP analysis can isolate green-colored species in the presence of an excess of red-colored species (28). If the green species do not interact with the red species, then the resulting HSP brightness will fall on the green species line (Fig. 6 A). However, if the green species comoves with red species, then the resulting HSP brightness will fall into a shaded area, which indicates that these two species are cooccupying and move together as a molecular complex. Using this strategy, we found that PI4KIIβ comoves with clathrin light chain. From a comparison of Fig. 6, B and C, it is clear that the distribution of HSP brightness of GFP-PI4KIIβ/Clathrin-mCh is very different from that of PI4KIIβ/mCh. Most of the HSP brightness of GFP-PI4KIIβ/Clathrin-mCh falls into the green/red comobile zone, whereas the HSP brightness of GFP-PI4KIIβ/mCh falls on the green species line. Thus, it is clear that GFP-PI4KIIβ-containing complexes/vesicles do not contain mCh, but do contain Clathrin-mCh.

PI4KIIα is known to target the adaptor protein AP-1 to Golgi membranes (18), and is also a component of AP-3 derived vesicles (51,52). Here we show that PI4KIIβ codistributes with AP-2, the adaptor that functions primarily in receptor-mediated endocytosis from the plasma membrane (Fig. 6 D). Although our methods do not allow us to monitor the codistribution of three fluorescently tagged proteins simultaneously, our results strongly support a model in which PI4KIIβ is enriched in a population of clathrin-coated AP-2-containing vesicles. Given their small size and dependence on Dyn2 for their formation, it is likely that these vesicles represent elements of the initial stages of endocytosis, before they become uncoated and fuse with early endosomes. In the future work we aim to define the functional significance of the presence of PI4KIIβ in these early endocytic structures.