The fusion of early endosomes induces molecular-motor-driven tubule formation and fission

In the endomembrane system there is a continuous remodeling of membranes through fusion and fission. Through these processes membrane dynamics are specifically controlled to maintain the membrane equilibrium and integrity from early to late endocytic compartments. The early endosome is a particularly dynamic compartment with a high fusion capacity (Gruenberg et al., 1989). Early endosomes fuse with clathrin-coated-derived vesicles and also with each other, modify their geometry through tubulation and membrane fission, move over long distances and transfer cargo to other organelles (Clague, 1998). Early endosomal homotypic fusion is a way of ensuring directionality of endocytic membrane trafficking between transport vesicles and the target organelle.

In addition to being a highly fusogenic organelle the early endosome also demonstrates a complex pleiomorphic organization of cisternal and tubular regions, membrane invaginations and multivesicularity. For the early endosome to exhibit this pleiomorphic distinctiveness, selective changes must occur in the curvature and organization of the bilayer during invagination or tubule formation (Mukherjee and Maxfield, 2000). Both extrinsic and intrinsic forces within the membrane drive the formation of tubules. Cytoskeletal elements have long been known to play some role in membrane traffic, not only by forming the structural scaffold and network controlling the vesicle trafficking, but also by directly deforming membranes (Lippincott-Schwartz et al., 2000; Vale and Hotani, 1988). Microtubule motors can pull a developing membrane tubule along a preformed microtubule track in vitro (Roux et al., 2002). Other cytoskeletal elements, such as actin filaments and membrane-tethered myosin motors, could similarly participate in membrane deformation (Buss et al., 2001; Morris et al., 2002). Additionally cytosolic proteins can alter the curvature of the lipid bilayer and participate in tubular formation (Peter et al., 2004).

To facilitate the visualization and analysis of endosomal interactions we enlarged the early endosomes by two well-described different approaches that increase fusion of early endosomes. This was done by high-level expression of either the major histocompatibility complex class II-associated invariant chain (Ii; also known as HLA class II histocompatibility antigen gamma chain or CD74) (Nordeng et al., 2002; Stang and Bakke, 1997) or Rab5 (Roberts et al., 1999). We observed an explosive tubular formation and fission immediately after the fusion, a process specifically controlled by microtubule motors.

Early endosomes are small fast moving heterogeneous compartments that are difficult to visualize. To better analyze interactions, larger endosomes were generated by high-level expression of the Ii or Rab5. The cytosolic tail of Ii is fusogenic and expression of Ii causes a prolonged endosomal pathway and enlargement of early endosomes (Engering et al., 1998; Nordeng et al., 2002; Romagnoli et al., 1993; Stang and Bakke, 1997). High-level expression of Rab5 has been found to increase endosome recruitment of EEA1 and enhance the early endosome fusion, resulting in larger endosomes (Bergeland et al., 2008; Gorvel et al., 1991; Li et al., 1995; Simonsen et al., 1998; Stenmark et al., 1994). cDNA for both molecules was subcloned into the inducible expression vector, pMep4, and as a reporter for the early endosomes we used the C-terminus of EEA1 (residues 1257–1411) fused to GFP (ctEEA1–GFP) (Bergeland et al., 2008; McBride et al., 1999). All constructs were stably transfected into Madine-Darby canine kidney cells (MDCK) and human fibroblasts (M1 cells).

In control cells the diameter of the endosomes was 0.3–1 μm, however, 3–5 hours after Ii induction we observed frequent fusion events and a pronounced increase in size of the ctEEA1-positive vesicles (Fig. 1A, white arrow; supplementary material Movie 1). The characteristic diameter of endosomes 8 hours after induction was 2–4 μm, however, endosomes as large as 10 μm were observed. As the early endosomes fused and increased in size, there was a corresponding decrease in the number of small endosomes, indicating a transition to fewer and larger endosomes. Similar qualitative effects were observed in cells induced to express high levels of Rab5, with a corresponding transition to a reduced number of enlarged early endosomes (data not shown). The fusion between enlarged endosomes was rapid and completed within a few seconds as demonstrated in Fig. 1B.

In the endomembrane system there is a constant influx and efflux of membrane as a result of fusion and fission events. By following the homotypic fusions of the enlarged endosomal structures we detected tubular extensions coated with ctEEA1– GFP (Fig. 2A,D; supplementary material Movie 2). These dynamic tubular structures were especially apparent directly after fusion, seemingly as a direct consequence of the fusion. During the first minute after fusion we measured a 4.5-fold increase in tube formation from the endosomes (prior to fusion, 1.7±0.8 tubules per minute; after fusion, 7.8±2.6 tubules per minute; Fig. 2B). This particular tube formation subsequently decreased during the few minutes after fusion (Fig. 2B), if not followed by another fusion and a subsequent eruption of tubules. The average speed of the emerging tubes, measured after ten different fusions, was 0.9 μm/second, and they could stretch to as long as 10 μm. The tubules were dynamic and some could extend and then disappear within 1 second, whereas others were more stable and could stay elongated for several seconds. By rapid four-dimensional (4-D) imaging we could also detect fission of the tubules as small free ctEEA1–GFP-positive vesicles (Fig. 2C; supplementary material Movie 3).

To study whether endocytic cargo could be transported through the tubules, Alexa-Fluor-555-labeled epidermal growth factor (EGF–Alexa-Fluor-555) was added to M1 cells stably expressing Ii and ctEEA1–GFP. Internalization was detected after 2 minutes and labeled EGF localized to early endosomes, where we subsequently observed accumulation of labeled EGF in microdomains on the enlarged endosomal structures (Fig. 2E; supplementary material Movie 4). Transport to the enlarged structures was observed to be through ctEEA1-positive vesicles or individual EGF-labeled vesicles (data not shown). Interaction of the incoming endosomes with the enlarged vesicles occurred through numerous fusion events and consequently numerous tubule formations were observed. The majority of the observed tubules formed were ctEEA1–GFP positive; however, 20±3.6% of the tubules contained EGF–Alexa-Fluor-555 (Fig. 2E, indicated by white arrows). We observed complete fission from the tubules, as shown in Fig. 2E. Apparently, EGF can be transported through these tubules and this can work as a transport step for proteins or receptors in early endosomes. Given that increased endosomal tubule formation and fission occurred immediately after fusion, this suggests that the fusion facilitates tubule formation and subsequent vesicular fission for intracellular vesicular transport.

A working endosomal pathway depends on an intact cytoskeleton and endosomal transport is microtubule (MT) dependent (Nielsen et al., 1999). The average velocity of emerging tubules in our experiments was 0.9 μm/second, and this is in the range of the known speed of microtubule motors (Coy et al., 1999). To test for a link between emerging tubules and microtubule motors MDCK and M1 cells with enlarged Ii-induced endosomes were treated with the microtubule depolymerizing agent nocodazole. Independent experiments with MDCK cells stably transfected with tubulin–EYFP showed depolymerized microtubules when incubated with 10 μM nocodazole on ice for 45 minutes (data not shown). An impaired MT network did not inhibit fusion between the enlarged endosomes, on the contrary, the size of the early endosomes continued to grow dramatically, indicating an increased fusion activity (Fig. 3A; supplementary material Movie 5). In non-treated cells the average fusion rate of enlarged endosomes (diameter >2 μm) was 5±1.4 fusions per hour, whereas in cells treated with nocodazole the fusion rate was calculated to 16±3.9 fusions per hour (means ± s.d.; Fig. 3B). Interaction with microtubules thus seems to have regulatory effect on enlarged homotypic early endosome fusion.

Without intact microtubules the post-fusion, ctEEA1–GFP-positive tubular extensions were no longer observed. Instead, intralumenal ctEEA1–GFP-positive membrane extensions were detected and occasionally free ctEEA1–GFP-positive vesicles appeared in the endosomal lumen (Fig. 3C; supplementary material Movie 6). Consequently the dramatic increase in the size of early endosomes after depolymerization of microtubules is a result of both an increased fusion rate and impaired tubular fission.

Because a lack of intact microtubules leads to increased fusion of enlarged endosomes this effect should also be visible studying normal sized endosomes. We therefore performed identical experiments as above but the MDCK cells were transfected with ctEEA1–GFP alone. After nocodazole treatment, the spatial distribution changed and the EEA1-positive endosomes moved to the cell periphery, as expected because the microtubules were no longer intact. Moreover, the size of the ctEEA1–GFP-positive vesicles increased rapidly by active fusion (becoming three to four times larger than their original size), and no tubule formation or fission was observed after fusion (Fig. 3D; supplementary material Movie 7). Together, our data indicate that microtubules regulate endosome fusion and fission processes independently of vesicle size.

It is well known that the microtubule-associated motors kinesin and dynein control vesicular trafficking and tubule formation (Bourekas et al., 1999; Roux et al., 2002; Verhey and Hammond, 2009). Having established a role for microtubules in tubular formation we decided to investigate the role of specific microtubule motors already shown to have a role in trafficking of early endosomes. Kif5B, a kinesin-1 family member, is important for peripheral distribution of early endosomes, as well as endosome fission (Loubery et al., 2008; Nath et al., 2007). We constructed a mutant known to inhibit kinesin-1 function encompassing the first 330 amino acids of the human Kif5b fused to EGFP (Schepis et al., 2007), which we named mutKif5b. This mutant was transfected into MDCK cells expressing Ii-induced enlarged endosomes. In cells positive for mutKif5b–EGFP we observed that mitochondria changed from being uniformly distributed in the cytoplasm to become more clustered in the perinuclear region, which did not occur in control cells (Fig. 4A). This is an established effect of knocking down Kif5b (Tanaka et al., 1998) and indicated a successful inhibition of Kif5b activity. For an early endosomal marker the cells were transfected with ctEEA1–mRFP or Rab5–mCherry. After transfection of mutKif5b, the cells still had enlarged endosomes and the endosomal localization was not specifically altered. Fusions and extensive tubule formations were still observed, indicating that Kif5B is not required for the specific Rab5- or EEA1-positive post-fusion tubule formation (Fig. 4B,C; supplementary material Movie 8, data for ctEEA1–GFP is not shown).

Another candidate for pulling the post-fusion tubules is the endosome-binding microtubule motor Kif16b that belongs to the kinesin-3 family. Kif16b was reported to be responsible for movement of early endosomes on microtubules (Hoepfner et al., 2005). Two different YFP constructs, the wild-type Kif16B–YFP and the dominant-negative mutant Kif16B–S109A–YFP, were transfected into MDCK cells. Whereas both fusion proteins were localized to Rab5–mCherry-positive early endosomal vesicles, the wild-type one localized more to the cell periphery and the mutant to the perinuclear area, as reported earlier (Hoepfner et al., 2005) (Fig. 5A,C). The post-fusion Kif16B–YFP-positive tubules increased dramatically after fusion (Fig. 5A–C; supplementary material Movies 9 and 10), from 1.2±0.8 tubules per minute before fusion, to 8.2±2.6 tubules per minute in the first minute after fusion (Fig. 5D). In cells expressing the mutant Kif16B-S109A–YFP, by contrast, hardly any tubules were visible after fusion (Fig. 5C,D; supplementary material Movie 11). The serine to alanine mutation in Kif16B has been shown to impair its motor activity and the GFP fusion protein has a dominant-negative effect on the endogenous motor (Nakata and Hirokawa, 1995). Because tubulation after fusion was barely observed in cells expressing the Kif16B mutant this indicates an active role for Kif16B in pulling the tubules on microtubule tracks.

Post-fusion tubules were either Rab5–mCherry positive, Kif16b–YFP positive or labeled with both markers (examples of the different types are indicated by colored arrows in Fig. 5B). The three different tubules signified by the two markers, gives an indication of a complex tubular fission system. An additional interesting observation was that Kif16b-positive tubules predominantly elongated toward, and possibly made contact with, adjacent Kif16b–YFP-positive endosomes (Fig. 5A; supplementary material Movie 9).

A further molecular motor found to be involved in controlling the trafficking of early endosomes is the dynein–dynactin complex (Burgess et al., 1999; Driskell et al., 2007). To test whether this complex is involved in tubule formation we transfected the dominant-negative mutant of dynactin in which the coiled-coil-1 (CC1) region of the p150^glued subunit (also known as dynactin subunit 1, DCTN1) was fused to monomeric RFP (CC1–mRFP) (Driskell et al., 2007). As a control for the CC1 dynein inhibition, MDCK cells were transfected with CC1–mRFP and labeled for anti-GM130, a Golgi marker. We could clearly detect a fragmented Golgi network in these cells, which was not seen in control cells transfected with wt–mRFP (Fig. 6A), evidence for the functionality of the construct (Christoforidis et al., 1999b). MDCK cells stably transfected with ctEEA1–GFP and Ii were induced to express Ii for 24 hours, resulting in the formation of many enlarged endosomes. These cells were then transiently transfected for 18 hours with CC1–mRFP or with mRFP as a control. In cells transfected with CC1–mRFP we could clearly observe fewer enlarged endocytic structures compared with control cells (Fig. 6B). By calculating the fraction of cells with enlarged endosomes, we identified a specific difference between CC1-transfected cells and control cells. Enlarged endosomes were found in 70.3% mRFP control cells, but only 25.4% of the cells transfected with CC1–mRFP contained enlarged endosomes (Fig. 6). Thus, our data indicate that the dynein motor is also involved in controlling the size of the enlarged early endosomes.

In the 25.4% of CC1–mRFP-transfected cells with enlarged endosomes we observed intraluminal ctEEA1–GFP-positive structures in the enlarged vesicles (Fig. 6E, white arrows). This is similar to the situation when the microtubules were depolymerized and there were no tubules formed after fusion (Fig. 3B, white arrows). This result indicates that tubule formation was inhibited and provides further evidence that dynein is involved in the formation of tubules.

In this work we took advantage of two different mechanisms to increase the size of the early endosomes in order to gain new insight into the dynamic interactions of these organelles in live cells. The fusion pattern of vesicles in living cells has previously been divided into ‘explosive’ and ‘bridge’ fusion (Roberts et al., 1999). In explosive fusion there is a rapid merger of vesicle membranes, leading to the formation of a single, enlarged compartment. In bridge fusion membrane merger proceeds slowly, and the transfer of membrane from the donor to the acceptor endosomal vesicle appears to occur entirely through a very narrow bridge between fusion pairs (Roberts et al., 1999). The fusion of the EEA1-positive endosomes observed in our experiments can then be described as explosive, because it is completed in a few a seconds. The more slow bridge fusion probably represents a fusion event taking place in later stages in the endocytic pathway (Roberts et al., 1999), and was not observed in our experiments.

Subsequent to the rapid early fusion of endosomes we were able to visualize and calculate extensive emergence of thin ctEEA1–GFP-coated tubules from various parts of the newly formed organelle. From Fig. 2B we can see that there are tubes being formed without a ‘detectable’ preceding fusion. This is an expected result because we cannot detect all the incoming fusions of smaller vesicles that characterize the tube formation background. From the very tip of the tubules we observed small, ctEEA1–GFP vesicles pinching off. We were not able to follow these vesicles for long, and could not detect their target membrane. However, after uptake of EGF–Alexa-Fluor-555 we found that only a specific fraction (1/5) of the EEA1-positive, post-fusion tubules contained the fluorescent EGF. These structures appeared to reflect a transport of cargo between early endosomal vesicles or towards the plasma membrane. EGF stimulation of EGF receptors directs the majority of the receptors to lysosomal degradation, although some are recycled (Roepstorff et al., 2009). Whether this is a specific tubular sorting of the EGF receptors alone is not possible to predict through our experimental setup. However, our observations indicate a transport option for cargo or the EGF receptor, concentrating them into tubules either for recycling or sorting within the endocytic pathway.

The stimulus for the extensive tubule formation may be that when two vesicles with identical volume fuse, the newly formed vesicle has 20% excess membrane, regardless of vesicle size (Pantazatos and MacDonald, 1999). This additional membrane has to be removed in order to balance the surface to volume ratio after fusion, which could be achieved by tubule formation. Even more importantly, the addition of membrane upon fusion also destabilizes the endosomal membrane, and subsequently lowers the force barrier connected with the initial deformation of the membrane (Koster et al., 2005). Molecular motors can utilize this transient instability in the membrane to form a tubule.

By depolymerizing the microtubules we removed the intracellular highways required for trafficking, fusion and fission (Murray and Wolkoff, 2003). Subsequent to the depolymerization of the microtubules no tubules were observed after fusion. On the contrary, we observed both tubular and vesicular intralumenal structures, reminiscent of observations made during in vitro experiments (Pantazatos and MacDonald, 1999). We hypothesize, that by depolymerizing the microtubules the excess membrane presumably invaginates into the luminal space in order to maintain the surface to volume equilibrium.

The average formation speed of the fusion-induced tubules was calculated to be similar to established vesicular trafficking speed along the microtubules (Coy et al., 1999), and the fusion tubules were indeed found to be dependent on intact microtubules. When microtubules were depolymerized by nocodazole treatment we additionally observed an increase in the endosomal fusion rate, concomitant with a further increase in the size of the already enlarged endocytic structures (Fig. 3A). This is a dual effect of increased fusion rate and obstruction of tubule formation and fission. Furthermore, a similar effect was detected with ctEEA1–GFP in normal sized endosomes in control cells (Fig. 3C). Our observations indicate that the strict microtubule control of endosomal fusion and fission is similar in cells with enlarged endosomes as in unperturbed cells.

Two families of kinesins have been found to interact with the early endocytic compartments, kinesin 1 and kinesin 3 (Hirokawa et al., 2009). We were specifically interested in Kif5b and Kif16b, both known to be involved in early endosomal trafficking and tubule formation (Hoepfner et al., 2005; Nath et al., 2007). High expression of the Kif5b mutant altered mitochondrial localization but had no observable effect on tubule formation. By contrast, we found that Kif16b–YFP localized to three different classes of post-fusion emerging tubules: Kif16b-YFP-positive tubules, Rab5–mCherry-positive tubules, and tubules positive for both Kif16b–YFP and Rab5–mCherry. This demonstrates the complexity of the sorting and trafficking from the early endocytic compartment. Expression of the mutant Kif16b inhibited post-fusion tubule formation and confirmed a specific role for KIf16b in tubule formation. Additionally, the most prominent tubules observed in the experiment were Kif16b–YFP positive. These tubules predominantly interact with Kif16b–YFP-positive early endosomes and presumably characterize an endosomal tubular cargo sorting induced by fusion.

The minus-end motor dynein was also demonstrated to be involved in regulating the size of the enlarged endosomes. In cells expressing CC1–mRFP of the dynactin p150^glued, the dynein–dynactin motor complex is impaired (Driskell et al., 2007). Only 25.4% of these cells contained enlarged endosomes, much less than in control cells, indicating impaired fusion and fission coordination. Additionally, we found ctEEA1–GFP-positive, enlarged multivesicular bodies in cells expressing CC1–mRFP, but empty enlarged endosomes in control cells expressing mRFP. This suggests impaired fission after fusion, similar to what we observed when the cells were incubated with nocodazole. Dynein has been shown to facilitate fusion of early endosomes (Driskell et al., 2007) and combined with our data this indicates that dynein plays a dual role facilitating both fusion and fission.

In conclusion, we show, through live cell experiments, a casual link between fusion and fission, where fusion functions as a membrane-destabilizing factor, facilitating spontaneous tubule formation and fission, controlled by molecular motors.

Complementary DNAs encoding CD74 (Bakke and Dobberstein, 1990) and Rab5 (Chavrier et al., 1990) have been described previously. These were subcloned into the pMep4 vector (Invitrogen) as KpnI–BamHI and HindIII–XhoI fragments, respectively (Gregers et al., 2003). ctEEA1–GFP has been previously described (Christoforidis et al., 1999a; Lawe et al., 2002; McBride et al., 1999).

Rab5–mCherry was made by amplifying Rab5 and mCherry separately by PCR using the following primers: for Rab5, 5′-AGAGAGGATCCATGGCTAGTCGAGGCGCAAC-3′ (the BamHI site is in bold) and 5′-AGAGACTCG-AGTTAGTTACTACAACACTG-3′ (XhoI site is in bold); for mCherry, 5′-AGAGAGGTACC ATGGTGAGCAAGGGCGAGGAG-3′ (KpnI site) and 5′-AGAGAGGATCCCTTGTACAGCTCGTCCATGCC-3′ (BamHI site). mCherry was ligated to the N-terminus of Rab5 by the BamHI site and subsequently subcloned into the pcDNA3 vector (Invitrogen) at the KpnI and XhoI sites.

The primer pair 5′-GCTCAGAATTCATGGCGGACCTGGCC-3′ and 5′-CAGATGGGCCCGACAAACTGTGTTCTTAATTGTTTTGG-3′ was used to amplify the coding sequence for the first 330 N-terminal amino acids of human KIF5B. PCR products were ligated into pEGFP-N1 (Clontech) using the EcoRI and ApaI restriction sites. Cells grown on 12 mm glass coverslips were transfected with plasmid DNA using FuGENE 6 (Roche) according to the manufacturer's protocol. KIF16B–YFP and KIF16B-S109A–YFP were kindly provided by Marino Zerial (Max Plack Institute of Molecular Cell Biology and Genetics Dresden, Germany) and used as described previously (Hoepfner et al., 2005). CC1–mRFP was kindly provided by Viki Allan (University of Manchester Faculty of Life Sciences, UK).

Madine-Darby canine kidney strain II (MDCK) and M1 (human fibroblasts) cells were stably transfected with Ii, Rab5-pMep4 and ctEEA1–EGFP. Ii and Rab5 expression was induced by the addition of 25 μM CdCl₂ in MDCK cells and 5 μM in M1 cells. The cells were grown in complete medium: DMEM (Bio Whittaker) supplemented with 9% FCS (Integro), 2 mM glutamine, 25 U/ml penicillin and 25 μg/ml streptomycin (all from Bio Whittaker) in 6% CO₂ in a 37°C incubator.

Stably transfected MDCK cells were grown in chambered coverglasses (Nalgene and Mattek), incubated with DMEM at 37°C overnight. The cells were incubated with CdCl₂ for between 6 and 12 hours. Immediately before imaging the cells were washed with PBS and then incubated with microscopy medium: DMEM without Phenol Red and sodium carbonate, supplemented with 3.5 g/l D-glucose to a final concentration of 4.5 g/l; 25 mM HEPES, adjusted to pH 7.5 (Bergeland et al., 2001), plus 10% FCS. EGF–Alexa-Fluor-555 (Molecular Probes) was added to the live cells on the microscope, at a final concentration of 100 ng/ml.

For fast and 4-D imaging, a Perkin Elmer Ultra View LCI/RS and Andor Revolution XD spinning disc microscope was used with an Olympus IX 71 with a PlanApo N 60×/1.42 NA oil immersion objective. The CSU22 spinning disc unit was synchronized with an iXon^EM+ 885 EMCCD camera. A stable cellular environment was maintained with a 37°C microscope chamber (Solent Scientific, Segensworth, UK).

The LCI system was equipped with a NikonEclipse TE 2000 microscope with an Astro Cam Ultra Pix CCD camera. Nikon Plan Fluotar, 100×1.30 NA oil immersion objective, and the RS scanning system was equipped with a Zeiss Axiovert 200 with a 63×/1.40 NA oil immersion objective and a Hammamatsu Ocra ER cooled CCD camera. Spinning disc technology, provided by Nipkow (http://www.yokogawa.com/scanner/products/csu22e.htm), used a multiple pinhole disc. Exposure time varied from 20 to 500 ms depending on the purpose of the experiment and sample intensity. Samples were exited with an argon–krypton, argon ion laser using the 488/568 lines.

For movie and image analysis, Imaris (Bitplane) and ImageJ software was used. The elongation speed was calculated from ten different experiments showing post-fusion tubule formations. ImageJ was used to manually track and calculate the speed of tubule elongation.

Projections and color-coded movies were produced with a macro written in IDL 5.5 (http://www.exelisvis.com/language/en-US/ProductsServices/IDL.aspx) by Timo Zimmermann, EMBL. The color-coded projection is in three colors: blue for the uppermost z-slices to the top of the cells, green for the middle and red for the region close to the coverslip.

Nocodazole (Sigma) were dissolved in dimethyl sulfoxide (DMSO) at concentrations of 10 μM. After the drug was diluted in cultures, the final concentration of DMSO never exceeded 1% (vol/vol). Nocodazole was added to the sample 30 minutes after incubation on ice, the rest of the reagents were added when the sample was on the microscope stage.

We thank Anne Simonsen and Gareth Griffiths for fruitful discussions and critically reading the manuscript. We also thank the Advanced Light Microscopy Facility at the EMBL Heidelberg and the Molecular Imaging Centre at the University of Bergen for the use of their instrumentation in the early stages of the project. The major part of the work was performed at the NorMIC Oslo imaging platform, University of Oslo.