Diffusion of Exit Sites on the Endoplasmic Reticulum: A Random Walk on a Shivering Backbone

Abstract

Major parts of the endoplasmic reticulum (ER) in eukaryotic cells are organized as a dynamic network of membrane tubules connected by three-way junctions. On this network, self-assembled membrane domains, called ER exit sites (ERES), provide platforms at which nascent cargo proteins are packaged into vesicular carriers for subsequent transport along the secretory pathway. Although ERES appear stationary and spatially confined on long timescales, we show here via single-particle tracking that they exhibit a microtubule-dependent and heterogeneous anomalous diffusion behavior on short and intermediate timescales. By quantifying key parameters of their random walk, we show that the subdiffusive motion of ERES is distinct from that of ER junctions, i.e., ERES are not tied to junctions but rather are mobile on ER tubules. We complement and corroborate our experimental findings with model simulations that also indicate that ERES are not actively moved by microtubules. Altogether, our study shows that ERES perform a random walk on the shivering ER backbone, indirectly powered by microtubular activity. Similar phenomena can be expected for other domains on subcellular structures, setting a caveat for the interpretation of domain-tracking data.

Introduction

The endoplasmic reticulum (ER) is a prominent organelle in eukaryotic cells with multiple vital functions: ribosome-decorated flat membrane cisternae in the cell center (“rough ER”) are responsible for the translation and translocation of about 10⁴ membrane protein species (1). Contiguous with these cisternae and responsible for lipid synthesis, the “smooth ER” extends as a wide ribosome-free network of membrane tubules throughout the cell (1, 2, 3) (see also Fig. 1 a for a representative image). Topologically, the smooth ER is mostly composed of three-way junctions that connect segments (i.e., membrane tubules) with a typical length of ∼1 μm (4). Notably, the smooth ER’s fishnet-like morphology is not only observed in living mammalian cells, but very similar topologies and geometries have been obtained via self-assembly in reconstitution assays (4, 5, 6).

Figure 1.

(a) Representative fluorescence image of a living HeLa cell with ERES and the ER being highlighted via Sec16-GFP and ssKDEL-RFP in green and red, respectively (see Materials and Methods for details). ERES are visible as a dispersed punctate pattern on the ER network, with an increased density in regions where ER membranes accumulate. (b) A magnification of the ER network (gray-shaded) with automatically determined positions of ER junctions and ERES—highlighted by red crosses and filled green circles, respectively—is shown. (c) The probability distribution function of distances between ERES and the nearest ER junction in a single cell, p(ξ), shows a pronounced peak with a mean $\langle \xi \rangle \approx 0.69$μm, suggesting that ERES are situated on ER tubules between two adjacent junctions (average distance between ER junctions: 1.01 μm). The experimental data is well described by a Wigner distribution, p(s) = πs⋅exp(−πs²/4)/2 with $s=\xi /\langle \xi \rangle$ (red line). The inset shows that the radial arrangement of normalized vectors between ERES and nearest ER junctions does not show an orientational preference, and the averages $\langle \text{cos}\theta \rangle$ and $\langle \text{sin}\theta \rangle$ of the associated angles are within the statistical bounds for random orientations. To see this figure in color, go online.

The smooth ER network also hosts distinct export gates for nascent proteins that need to travel along the secretory pathway: after clearance by the quality control machinery (7), properly folded proteins accumulate in stationary membrane domains, called ER exit sites (ERES), where they are packaged into vesicular carrier structures (see (8) for a recent review). The emerging transport intermediates are mostly 50 nm-sized vesicles (9) whose coat machinery is modulated, for example, by cargo proteins (10), sterols (11), and motor-associated factors (12, 13) for efficient cargo sorting and subsequent long-range transport.

Despite being fairly large and long-lived domains that can be monitored over extended timescales (14, 15), a thorough biophysical understanding of the self-assembly principles and the dynamic maintenance of ERES is still lacking. For example, how can mammalian cells feature ∼100 stationary ERES on the contiguous ER network (cf. Fig. 1, a and b) although their major molecular determinants, i.e., peripheral membrane proteins like Sec16 and COPII proteins, cycle rapidly (10, 16) between ER membranes and the cytosol?

In fact, ERES have been shown to self-assemble dynamically de novo in the yeast Pichia pastoris (17) presumably because of an initiating clustering of the peripheral membrane protein Sec16 (18). However, equally conclusive data for the larger mammalian cells are lacking so far because of a more complex set of involved molecular constituents with very rich interaction patterns. Recent findings have indicated, for example, that ERES homeostasis in mammalian cells depends on proliferation and secretion (16), suggesting that ERES also act as integrator platforms for growth-factor signaling. Because of their rather complex composition and regulation on the molecular level, we will hence adopt a more mesoscopic perspective on ERES.

In the spirit of condensation phenomena, which have recently received considerable attention in the cell-biological context (19), ERES self-assembly may be modeled effectively on mesoscopic scales as two-dimensional droplet formation of some molecular determinant(s). In this approach, specific interactions are integrated into effective parameters like the Flory-Huggins interaction parameter, with effective ERES constituents being subject to diffusional motion. Yet, given that diffusion coefficients of membrane domains only depend logarithmically on size (20, 21), growing ERES should rapidly explore their host membrane, fuse with each other, and hence form few but large domains. Obviously, this is in strong contrast to the stationary, dispersed ERES pattern observed experimentally. However, a basic two-dimensional aggregation model combined with a rapid dissociation of ERES constituents from ER membranes has been able to reproduce the punctate ERES pattern (22) when a strong size-dependence of the domains’ diffusion coefficient was assumed ad hoc. Therefore, the mobility of ERES on ER membranes moves into focus as an important observable.

In the simplest case, one could imagine that ERES self-assemble on ER junctions and become topologically trapped at these loci by looping around the y-shaped crossing of three membrane tubules. As an alternative, ERES might self-assemble on ER tubules and remain mobile on this segment but are trapped there in the long run because the domains cannot overcome the adjacent junctions without major topological dislocations. Both mechanisms, albeit very different in detail, would result in an almost vanishing long-range diffusion coefficient, hence satisfying a key requirement for maintaining dispersed droplet-like domains (22) without contradicting basic fluid dynamics (20, 21). So far, however, it has not been tested which of these two possibilities is eventually implemented in living mammalian cells.

Moreover, in a broader perspective, one may view ERES as representatives of the many bona fide domains on the cell’s vast set of endomembranes. Gaining insights into the secret life of these domains often includes quantifying their motion pattern with respect to the cell’s center of mass but also relative to their host membranes. Quantifying ERES mobility in detail therefore can also serve as a benchmark for other membrane domains.

Here, we report our findings on ERES mobility as obtained from extensive single-particle tracking experiments. In particular, we have quantified and compared the random-walk properties of ERES in untreated and nocodazole-treated cells—i.e., after disrupting the microtubule cytoskeleton—in relation to the motion of ER junctions. As a result, we find that ERES show a microtubule-dependent and heterogeneous anomalous diffusion on short and intermediate timescales that is markedly different from the motion of ER junctions. Hence, ERES are not locked on ER junctions but rather seem to move diffusively on ER tubules. These results are corroborated by simulations in which ERES motion on ER tubules is mimicked in a simplified fashion by a freely diffusing flag on a fluctuating semiflexible polymer. Together with our experimental data, these simulations also indicate that ERES are not actively moved by microtubules. Changes of the random-walk properties of ERES rather are indirect consequences of ER network fluctuations that strongly depend on the presence of microtubules. We propose that similar superpositions of domain motion and (active) fluctuations of the host membrane need to be taken into account when interpreting single-particle tracking data of other domains.

Materials and Methods

Cell culture and imaging

HeLa cells (German Collection of Microorganisms and Cell Cultures GmbH, ACC-57) were cultured in Dulbecco’s Minimal Essential Medium (Invitrogen, Carlsbad, CA) with phenol red, supplemented with 10% fetal calf serum (FCS; Biochrom, Holliston, MA), 1% L-glutamine (Invitrogen), 1% sodium pyruvate (Invitrogen), and 1% penicillin/streptomycin (Invitrogen). Cells were incubated in T-25 flasks (Corning, Corning, NY) at 37°C in a 5% CO₂ atmosphere.

One day after plating in four-well dishes (ibidi, Martinsried, Germany), cells were transfected with GFP-tagged Sec16 (Sec16-GFP, a kind gift of Farhan) (16) using 2 μL Fugene6 (Promega, Madison, WI) and 1 μg plasmid DNA in 100 μL serum-supplemented Dulbecco’s Minimal Essential Medium (Thermo Fisher Scientific, Waltham, MA). For covisualization of the ER, an additional plasmid DNA encoding for ssKDEL-RFP red fluorescent protein (a kind gift of Lippincott-Schwartz) (23) was added to the transfection mixture (2 μL Fugene 6, 0.75 μg ssKDEL-RFP, 0.75 μg Sec16-GFP). The transfection mixture was incubated at room temperature for 15 min and added to the cells maintained in fresh culture medium. For checking the localization precision of our single-particle tracking approach, cells transfected with Sec16-GFP were fixed after 1 day of expression by treating them for 15 min with 4% paraformaldehyde in 1× Dulbecco's phosphate-buffered saline (Biochrom) at room temperature. Fixed samples were stored in 1× Dulbecco's phosphate-buffered saline containing 1% paraformaldehyde at 4°C.

To depolymerize microtubules, HeLa cells were treated with 10 μM nocodazole (24, 25). To this end, a 2 mM stock solution of nocodazole (Sigma Aldrich, St. Louis, MO) dissolved in dimethylsulfoxide (Sigma Aldrich) was diluted in minimal essential medium without phenol red, supplemented with 5% FCS and 10% HEPES to working concentration. Nocodazole-treated cells were chilled on ice for 10 min before incubating at 37°C with 5% CO₂ for 15 min. Subsequent imaging was performed in the presence of the drug at 37°C or at room temperature. Successful depolymerization of microtubules with this protocol was confirmed by immunostaining with anti-α-tubulin mouse mAB (Cell Signaling Technology, Danvers, MA) as a primary antibody (Fig. S1).

Time-resolved two-dimensional imaging of transfected cells was performed with a spinning disk confocal setup, consisting of a Leica DMI 6000 microscope body (Leica Microsystems, Wetzlar, Germany) equipped with a CSU-X1 (Yokogawa Microsystems, Tokyo, Japan) spinning disk unit and a custom-made incubation chamber. Images were taken with a Photometrics Evolve 512 EMCCD camera (Photometrics, Tucson, AZ) using an HCPL APO 100×/1.4 NA (Leica Microsystems) oil immersion objective (excitation of GFP and RFP at 491 and 561 nm, respectively; corresponding fluorescence detection ranges were 500–550 and 575–625 nm). The setup was controlled by custom-written LabView software (National Instruments, Austin, TX). Imaging was performed at an interval of 200–260 ms with exposure times of 150–180 ms over a total period of ∼3 min (corresponding to 1000 frames). Live-cell imaging was performed at 37°C and at room temperature with cells being immersed in imaging medium (minimal essential medium without phenol red supplemented with 5% FCS and 5% HEPES). To extract two-dimensional ERES trajectories, image series were analyzed as described before (25) using the MATLAB Particle Tracking Code by Blair and Dufresne (available at http://site.physics.georgetown.edu/matlab). To avoid incorrect particle assignments, i.e., jumps to a nearby but different ERES when identifying a trajectory from individual positions, the maximal particle displacement between two frames was limited to three pixels (∼400 nm). Data acquisition, image analysis, and evaluation results of ER junction movement are reported and discussed in detail in (26).

Simulations

Simulations of (semi)flexible polymers were conducted with a Brownian dynamics approach using dimensionless quantities. In particular, a chain of N = 50 beads (bead radius r₀ = 1) was considered with nearest-neighbor beads i and j in a distance r_ij = |r_i − r_j| being coupled by a harmonic potential U_ij = k(r_ij − l₀)², with k = 50 and l₀ = r₀/2. In addition, the chain was equipped with a bending potential V_i = κ(1 − cosϕ) at each bead i, with the bond angle ϕ being defined via the scalar product $\text{cos}\varphi ={\hat{\mathbf{r}}}_{i-1,i}\cdot {\hat{\mathbf{r}}}_{i,i+1}$, where ${\hat{\mathbf{r}}}_{i,j}=\left({\mathbf{r}}_i-{\mathbf{r}}_j\right)/r_{ij}$. Random forces due to thermal noise were chosen such that every bead had a free diffusion constant of D = 10. Diffusion of a flag from bead to bead (representing the motion of an ERES on an ER segment) was obtained via a one-dimensional unbiased blind-ant scheme with total hopping probability p_hop to next-neighbor beads.

Motion of the chain was simulated via the beads’ overdamped Langevin equation, integrated by a fourth-order Runge-Kutta scheme (time step Δt = 10⁻³). For an initial 2 × 10⁶ steps, the freely moving chain was equilibrated with κ = 20, resulting in the typical, slightly undulating shape of a semiflexible polymer. Then, the end beads were frozen in their position, and the bending stiffness κ was set to the desired value. With this approach, an extended ER segment with a well-defined stiffness but held in place by two immobile ends was modeled. After an additional equilibration over 2 × 10⁶ time steps, the motion of beads in the chain and of the flag (moving on top of the beads) were recorded for 2 × 10⁶ time steps. Internal beads therefore served as a representation of ER junctions. From these data, the time-averaged mean-square displacement (TA-MSD) of individual beads and the flag were obtained via Eq. 1. Fitting TA-MSDs with Eq. 2 in the range t ∈ [0.1,10] was used to extract the apparent anomaly exponent, α. By dissecting the recorded trajectories into 8000 distinct sequences of 250 positions (similar in length to experimental ERES trajectories), the average asphericity (Eq. 3) was determined. For each parameter set (κ, p_hop), 10 simulation runs were conducted to obtain the averaged quantities ($\langle {\alpha }_0\rangle$, $\langle A_0\rangle$) and ($\langle {\alpha }_{\text{flag}}\rangle$, $\langle A_{\text{flag}}\rangle$) for the motion of beads and flag, respectively.

Results and Discussion

To quantify the mobility of ERES on short and intermediate timescales, we utilized a well-characterized peripheral membrane protein, Sec16, coupled to a green fluorescent protein (Sec16-GFP) that is known to highlight ERES with only a low residual cytoplasmic background (16, 27, 28). In line with previous observations, fluorescence images of HeLa cells transfected with Sec16-GFP showed the previously reported punctate pattern that is characteristic for ERES in mammalian cells (Fig. 1 a).

As a first indication of where ERES are situated on the ER network, we inspected the distance between Sec16-GFP punctae and ER junctions in living cells using ssKDEL-RFP as a general ER marker (23). Following our previous approach (26), we determined the positions of ER junctions by skeletonizing the observable network (see image magnification in Fig. 1 b). Then, we determined the probability distribution of distances p(ξ)between ERES and their next-neighbor ER junctions. This simple imaging approach confirmed that ERES and ER junctions are close to each other but are on average separated by $\langle \xi \rangle \approx 0.69$μm (average distance between ER junctions: 1.01 μm) with no indication for an orientational preference (Fig. 1 c). However, given that the measured distances were near to the diffraction limit, we refrained from drawing too-bold conclusions about the relative positions of ERES and ER junctions from this simple imaging approach. Instead, we focused on the dynamic features of ERES, using rapid imaging and subsequent single-particle tracking (see Materials and Methods for details). In particular, we monitored ERES motion (at room temperature and at 37°C) in untreated cells and in cells in which the microtubule network had been disrupted by application of nocodazole. In agreement with our earlier work (22), we did not notice a change in the apparent size or number of ERES upon treatment with nocodazole.

For the sake of statistics, we only considered ERES trajectories with at least 300 consecutive positions, i.e., shorter trajectories or incomplete position sequences were discarded. For comparable statistics within the ensemble, all remaining trajectories were chopped to a length of N = 300 positions. This choice also prevented an under-representation of small versus large ERES because the latter could be tracked easily over several hundred frames because of a slower photobleaching-induced disappearance. For the resulting set of individual ERES trajectories, we first determined the TA-MSD:

$${\langle r^2\left(\tau \right)\rangle }_t=\frac{1}{N-k}\sum _{i=1}^{N-k}{\left\{\mathbf{r}\left(\left(i+k\right)\mathit{\Delta }t\right)-\mathbf{r}\left(i\mathit{\Delta }t\right)\right\}}^2.$$ (1)

Here, τ = kΔt, with Δt being the frame time of the imaging series. Representative TA-MSDs of single ERES in an untreated cell are shown in Fig. 2 a together with the ensemble average of all TA-MSDs within the cell (see Fig. S2 for a representative ERES trajectory). The data is well described by a simple power-law expression of the form

$${\langle r^2\left(\tau \right)\rangle }_t=K_{\alpha }{\tau }^{\alpha },$$ (2)

with 0 < α ≤ 1 denoting the diffusion anomaly exponent and K_α being the generalized diffusion coefficient (K_α = 4D for normal Brownian motion, i.e., α = 1, with the familiar diffusion constant D). As can be seen from Fig. 2 a, a sublinear growth with α ≈ 0.6 is observed for TA-MSDs and their ensemble average, i.e., ERES show a pronounced subdiffusion on short and intermediate timescales. Notably, subdiffusion has been observed frequently in biological and biomimetic specimens (see (29, 30) for reviews), and multiple stochastic processes have been discussed as a microscopic explanation (reviewed, for example, in (31)). Analogous to previous observations on telomeres (25), the subdiffusive TA-MSD of ERES assumed markedly lower values when microtubules were disrupted but remained well above TA-MSD data of ERES in fixed cells (Fig. 2 b). These findings highlight that intact microtubules render ERES more mobile.

Figure 2.

(a) Representative time-averaged mean-square displacement (TA-MSD), ${\langle r^2\left(\tau \right)\rangle }_t$, of individual ERES in untreated HeLa cells (thin gray lines). The ensemble average over all tracked ERES in the same cell (bold black line) agrees well with a scaling ${\langle r^2\left(\tau \right)\rangle }_t\sim {\tau }^{\alpha }$ (dashed blue line). (b) The same as before is shown but for cells that have been treated with nocodazole. Because of disruption of the microtubules, markedly lower TA-MSDs with a lower anomaly exponent as in the untreated case are observed. Red lines depict TA-MSDs of ERES in fixed cells, giving an estimate for the finite localization precision. The negligible increase of the fixed cells’ MSD by few tens of nm² on timescales > 10 s is attributed to minor fluctuations of the sample and the imaging setup. To see this figure in color, go online.

Let us briefly discuss at this point the dominant sources of error that may perturb the analysis of TA-MSD curves via Eq. 2: the finite localization precision of particle positions from images (“static error”) adds a positive offset to the MSD, whereas the finite camera integration time (“dynamic error”) contributes a negative offset (32). In our case, the static error can be estimated from the MSD in fixed cells (Fig. 2), resulting in an average offset of +7.5 × 10⁻⁴ μm². The average dynamic error is given by −2K_αΔt^α/((α + 1)(α + 2)) ≈ −7.3 × 10⁻⁴ μm² when using α = 0.6 and K_α = 0.004 μm²/s^α, i.e., for parameters that match the plain TA-MSD of individual ERES (cf. Table 1). Hence, both contributions basically cancel each other on average but may fluctuate for individual trajectories. Consequently, trying to correct individual TA-MSD for any constant offset becomes a somewhat arbitrary process for the data discussed here. We therefore have chosen to report data extracted from plain TA-MSD data that have not been corrected for a potential offset, bearing in mind that this may lead to a slight additional uncertainty in the values of α and K_α.

Table 1.

Summary of Experimentally Determined Diffusion Parameters for ERES and ER Junctions at Room Temperature and at 37°C, without and with Nocodazole Treatment

		RTa (−)b	RT (+)c	37°C (−)	37°C (+)
ERES	$\langle \alpha \rangle$	0.64	0.56	0.62	0.57
	$\langle K_{\alpha }\times 1{\text{s}}^{\alpha }\rangle$[× 10−3μm2]	5.50	3.16	7.30	4.37
	$\langle A\rangle$	0.43	0.40	0.42	0.39
	n	4005	3272	3004	3984
ERJd	$\langle \alpha \rangle$	0.51	0.35	0.53	0.33
	$\langle K_{\alpha }\times 1{\text{s}}^{\alpha }\rangle$[× 10−3μm2]	3.47	3.39	4.79	5.37
	$\langle A\rangle$	0.35	0.27	0.35	0.30
	n	1900	2291	1075	388

α and K_α denote the diffusion anomaly and the generalized diffusion coefficient, respectively; A and n refer to the asphericity and the number of evaluated trajectories. Data for ER junctions were taken from (26); insignificant changes with respect to the previously stated numerical values are due to a re-evaluation with fixed trajectory length N = 300. Standard errors of the mean were smaller than ±0.01 in all cases. Parameter value changes induced by nocodazole treatment and differences in parameter values between ERES and ER junctions at the same conditions were significant to the 1% level (Kolmogorov-Smirnov as well as Student’s t-test).

^aRT, room temperature.

^b−, without nocodazole treatment.

^c+, with nocodazole treatment.

^dERJ, ER junction.

To gain more detailed insights into the fundamental parameters of ERES motion, we evaluated large sets of trajectories for all conditions and calculated from these the probability distribution functions of the anomaly exponents and of the generalized diffusion coefficients, p(α) and p(K_α), respectively. As can be seen in Fig. 3 a, the distribution of anomalies varies significantly upon disrupting microtubules via nocodazole, i.e., the average anomaly exponent is shifted to a lower value (values are summarized in Table 1). This reduction is also observed at room temperature with only minor variations in the actual values of $\langle \alpha \rangle$ (Fig. 3 a; Table 1). Notably, anomaly values in the cell periphery and in juxtanuclear regions did not show significant differences, i.e., $\langle \alpha \rangle$ in proximal and distal regions were the same within an uncertainty range of ±0.03. Therefore, our data clearly indicate that the (anomalous) diffusion of ERES is enhanced by microtubules, i.e., the TA-MSD scales with an elevated value of $\langle \alpha \rangle$ when microtubules are intact.

Figure 3.

Probability distribution functions of (a) anomaly exponents, p(α), and (b) generalized diffusion coefficients, p(K_α × 1 s^α), for untreated and nocodazole-treated cells at 37°C (black and red histograms, respectively). Because the units of K_α depend explicitly on α, subfigure (b) refers to the typical area covered within 1 s, K_α × 1 s^α; please also note the logarithmic scale. Filled and open squares with error bars indicate mean ± SD at 37°C and room temperature, respectively (see Table 1 for numerical values). A significant shift of the mean values upon disrupting microtubules is observed for both quantities (see main text for discussion). To see this figure in color, go online.

The observation of a subdiffusive motion of ERES immediately raises the question of whether ER junctions show the same mode of motion. We have shown earlier that ER junctions also exhibit an anomalous diffusion with a clear subdiffusive signature and that the gross dynamics of the fractal ER network appears to be governed by fractons (26). Similar to ERES, the diffusion anomaly of ER junctions was seen to depend strongly on the integrity of microtubules (see summary of parameters in Table 1). However, comparing the actual values of the anomaly exponent for ER junctions and ERES, significantly larger values of $\langle \alpha \rangle$ are seen for ERES, indicating that they are not merely following the motion of ER junctions.

This conclusion is corroborated by the distribution of generalized diffusion coefficients K_α (Fig. 3 b). Please note that because the units of K_α depend explicitly on the anomaly, α, bare generalized diffusion coefficients cannot be compared directly. We therefore focus on the typical area covered by an ERES within 1 s, K_α × 1 s^α, but refer to this quantity for brevity as K_α in the remaining text. Similar to previous observations for ER junctions (26) and telomeres in the nucleus of mammalian cells (25), a roughly lognormal distribution for K_α was also obtained for ERES. Moreover, disrupting microtubules led to a clear shift of the distribution to smaller values (see also summary of parameters in Table 1). The same shift to a lower value of $\langle K_{\alpha }\rangle$ was also present at room temperature, albeit the actual values without and with nocodazole treatment were decreased ∼1.5-fold as compared to 37°C (Fig. 3 b; Table 1). Notably, ER junctions also exhibited a temperature-induced decrease of $\langle K_{\alpha }\rangle$ by the same factor (cf. Table 1). Comparing $\langle K_{\alpha }\rangle$ for ERES and ER junctions in untreated cells, significantly higher values are seen for ERES, giving additional support to the notion that these domains are not just passively following the motion of ER junctions.

Aiming at additional clues about the geometrical properties of the random walk of ERES, we also calculated the trajectories’ mean asphericity, defined as

$$A=\frac{\langle {\left(R_1^2-R_2^2\right)}^2\rangle }{\langle {\left(R_1^2+R_2^2\right)}^2\rangle },$$ (3)

with R₁ and R₂ being the principal radii of gyration for a trajectory, and $\langle .\rangle$ denoting an average over all trajectories. Contrary to the naive expectation, random walks have generically aspherical geometries at each instant of time with values between a sphere (A = 0) and a rod (A = 1), e.g., A = 4/7 for two-dimensional Brownian motion (33). The familiar isotropic sampling of space via (sub)diffusion is recovered via the stochastic reorientation of a single trajectory’s principal axis or via averaging over a particle ensemble.

As a result, we observed that ERES feature ensemble-averaged asphericities that are significantly higher than those observed for ER junctions (Table 1). This finding gives further evidence for the notion that ERES are not locked to ER junctions, i.e., they do not simply follow the motion of the ER network. Given that motion along a (subdiffusively moving) linear segment within the ER network is essentially rod-like, an increased value of $\langle A\rangle$ clearly signals that ERES travel on ER tubules.

Following previous work on RNA-protein particles (34), we have also probed to what extent ERES trajectories show a diffusion heterogeneity. To this end, we inspected the probability distribution function of normalized squared increments, p(χ²), i.e., we calculated for each ERES trajectory the squared distances taken between successive frames and normalized these values by their mean. Without heterogeneity, a Gaussian distribution of increments is expected, i.e., p(χ²) ∝ exp(−χ²), whereas deviations indicate changes of the diffusion process within trajectories (34). As can be seen in Fig. 4 a, significant deviations from the anticipated Gaussian are observed in untreated and nocodazole-treated cells, highlighting a significant diffusion heterogeneity. Little to no changes were seen when analyzing instead the increments taken within ten frames. Moreover, p(χ²) is virtually identical in untreated and nocodazole-treated cells, suggesting that the heterogeneity is independent of microtubule integrity. Most likely the observed diffusion heterogeneity is a consequence of the association and dissociation of Sec16 and/or COPII proteins as well as the formation and pinch off of transport vesicles at individual ERES, leading to fluctuations in the domain’s topology, size, and mobility.

Figure 4.

(a) Probability distribution function of normalized squared increments, p(χ²), for successive ERES positions in untreated (black squares) and nocodazole-treated cells (red circles). Clear deviations from a simple Gaussian (straight dashed line) are observed in both cases, indicating a diffusion heterogeneity. (b) The associated ensemble- and time-averaged Gaussianity, g(τ), (bold black and red lines) confirm this finding by assuming values between zero and unity (dashed lines indicating the curves’ SDs). See also main text for discussion. To see this figure in color, go online.

The diffusion heterogeneity of ERES is further substantiated by the ensemble-averaged Gaussianity of trajectories, $g=d{\langle r^4\left(\tau \right)\rangle }_t/\left(3{\langle r^2\left(\tau \right)\rangle }_t^2\right)-1$ (with d = 2 for two-dimensional trajectories). Although g = 0 for a Gaussian and g = 1 for an exponential distribution of increments, our experimental data on ERES are in between (Fig. 4 b), similar to observations for RNA-protein particles in bacteria and yeast (34).

To gain additional insights for the interpretation of our experimental data, we have performed simulations that mimic the motion of an ERES on an ER tubule in a simplified geometry (see Materials and Methods for details). To this end, we have modeled an ER segment as a semiflexible filament of beads with bending stiffness κ. Because of thermal fluctuations, all beads within the filament show a subdiffusive motion on timescales below the Rouse time beyond which all beads follow the polymer’s center-of-mass motion. Hence, any of the beads can be used as a representative for the motion of an ER junction. In addition, a flag, indicating the position of an ERES, can diffuse freely on this filament. Here, the flag’s hopping probability p_hop to next-neighbor beads encodes the relative diffusional mobility of an ERES with respect to the underlying and fluctuating host membrane. Consequently, p_hop tunes the relative weight of the host beads’ motion and the flag’s hopping along the chain, both of which enter the flag’s trajectory. Using this simplified model, we calculated for varying bending stiffnesses κ and hopping probabilities p_hop the mean anomaly and mean asphericity for individual beads ($\langle {\alpha }_0\rangle$ and $\langle A_0\rangle$) and for the moving flag ($\langle {\alpha }_{\text{flag}}\rangle$ and $\langle A_{\text{flag}}\rangle$). Because our simulations did not include the influence of a viscoelastic environment (which sets the actual values $\langle {\alpha }_0\rangle$ and $\langle A_0\rangle$ for ER junctions (26)), we focused our analysis on the differences between the flag and the beads, $\mathit{\Delta }\alpha =\langle {\alpha }_{\text{flag}}\rangle -\langle {\alpha }_0\rangle$ and $\mathit{\Delta }A=\langle A_{\text{flag}}\rangle -\langle A_0\rangle$ as a function of κ and p_hop (Fig. 5).

Figure 5.

(a) Color-coded map of differences Δα in the anomaly exponent between the hopping flag and a simple bead as a function of the hopping probability p_hop and the filament’s bending stiffness κ. The almost vertically striped appearance of the map indicates that κ only has a minor influence, whereas an increased hopping rate shifts the apparent anomaly exponent $\langle {\alpha }_{\text{flag}}\rangle$ of the flag toward larger values. (b) Similarly, the map of differences in the asphericity ΔA indicates more elongated random-walk shapes when the hopping probability is increased. To see this figure in color, go online.

As a result, we observed that the bending rigidity κ only had an almost negligible influence on Δα and on ΔA, whereas increasing values of the flag’s hopping probability led to a significant increase in both quantities. Given that the beads’ anomaly and asphericity values did not show major changes in the tested range of κ and p_hop, our simulation data basically show that $\langle {\alpha }_{\text{flag}}\rangle$ and $\langle A_{\text{flag}}\rangle$ increase when the flag’s mobility is increased relative to the filament’s beads. Therefore, our model simulations support the above reasoning that ERES are mobile entities on the shivering ER backbone.

The simulation results also help to address another pending question that arises from our experimental data: are ERES directly shaken by microtubules or does their mobility only decrease in the absence of cytoskeletal elements because the ER network is not moved actively anymore? We first would like to emphasize that the experimentally determined anomaly α_ERES always reports on the superposition of tubule fluctuations and ERES diffusion on ER tubules irrespective of their relative importance. Therefore, decreased values of α_ERES upon nocodazole treatment are to be expected because of the concomitant reduction of the ER’s motion, reflected in α_ERJ (cf. Table 1). Yet, our experimental data also highlight significant changes for the diffusion anomaly difference between ER junctions and ERES. In particular, we find δα = α_ERES − α_ERJ ≈ 0.11 in untreated cells (average of values at room temperature and 37°C) and δα ≈ 0.22 in cells with broken microtubules. Therefore, ERES become more mobile in relation to their host membrane in the absence of microtubules. If microtubules were responsible for the motion of ERES relative to the ER network, no change or even a reduction of δα was expected. Furthermore, the asphericity difference ΔA also increases upon disrupting microtubules, i.e., ERES random walks appear more rod-like. Comparing these findings to our simulation data, the disruption of microtubules consistently translates into an elevated hopping probability, i.e., the relative motion of the flag (= ERES) on the filament becomes more pronounced. Therefore, the change in mobility observed for ERES when disrupting cytoskeletal elements does not seem to reflect an active pushing and pulling of microtubules on ERES. Rather the previously reported active, microtubule-dependent fluctuations of the ER network (the host substrate of ERES) add onto the domain’s microtubule-independent diffusive motion along ER tubules. It is even conceivable that intact microtubules might act against the free motion of ERES on ER tubules because direct interactions via the dynein-motor complex have been reported (12, 13). However, it is worth noting at this point that the role of microtubule-associated motors, e.g., dynein, cannot be dissected with our mesoscopic approach. Aiming at molecular insights and the interplay of components shared between the cytoskeleton and ERES, e.g., dynein, p150^glued, and COPII proteins, will have to rely on nanoscopic imaging approaches on the level of single ERES. At this point, the effects of crowding and sterical hindrance (35, 36) may also need careful consideration.

Neglecting molecular details and maintaining a mesoscopic view, our findings support the following model for ERES biogenesis: starting from an ER network devoid of ERES, e.g., after mitosis, ERES constituents explore their reticular host membrane via diffusion and start interacting with each other. Summarizing the constituents’ interplay by an effective Flory-Huggins interaction parameter, ERES formation may be described as a diffusion-driven condensation process. Growing ERES remain mobile on ER tubules during the entire process but eventually become effectively confined by ER junctions because overcoming these branch points of membrane tubules requires major topological changes of ERES domains. Therefore, ERES maintain a short-range diffusional mobility, but long-range diffusion along the ER network is suppressed. Hence, a coalescence of ERES to few but large droplet-like domains is prevented, in accordance with previous modeling results (22). As a result, a dispersed stationary ERES pattern is predicted and observed in mammalian cells in interphase.

Although probing the details of this model will require subsequent studies with major changes of the ER structure by drugs or RNA interference, our data on ERES also highlight an important and generic aspect of domain diffusion on a fluctuating organelle: because domains like ERES may move with but also relative to their host membrane, analyzing their motion requires special care and a thorough analysis that is capable of dissecting the superposition of the two transport processes. This is particularly important when bearing in mind that the two processes may be differentially governed by nonequilibrium driving forces (e.g., via the cytoskeleton and/or molecular motors) as seen here for ERES and the ER network. We hope that our data can serve here as a helpful and prototypical benchmark.

Conclusion

In summary, we have shown here that the ERES exhibit an anomalous diffusion behavior on short and intermediate timescales that is distinct from the subdiffusive motion of ER junctions. In particular, our experimental data and accompanying simulations provide strong evidence that ERES are diffusing on fluctuating ER tubules. Although the mobility of ERES clearly depends on the integrity of the microtubule cytoskeleton, the active driving appears to be indirect via the microtubule-induced motion of ER segments. We propose that similar phenomena need to be taken into account when interpreting the (activated) motion of other domains, e.g., respiratory chain complexes on mitochondrial networks (37) or sorting domains on endosomes (38). Analogous to the case of ERES, these domains may move with but also relative to their host membrane, hence requiring a thorough analysis of the available experimental data that dissects the different contributions.

Author Contributions

L.S. and K.S. performed experiments and analyzed the data. M.W. designed research, contributed analytic tools, and performed simulations. All authors contributed to writing the manuscript.

Editor: Claudia Steinem.