In order to reconstitute bidirectional transport in vitro, we isolated neuronal transport vesicles from a transgenic mouse expressing low levels of the dynactin subunit dynamitin fused to GFP. In this line, GFP-dynamitin is efficiently incorporated into dynactin, without altering the motility of the purified dynein-dynactin motor complex [2]. In neurons from these mice, GFP-labeled dynactin is localized in a punctate pattern in the cell soma and axon of neurons in situ and in culture (Fig. 1A), consistent with the possible integration of the labeled protein into membrane-associated dynactin complexes.

We isolated membranous vesicles by differential centrifugation followed by flotation through a discontinuous sucrose density gradient [3]. Analysis of the purified vesicle fraction demonstrates the co-purification of MT motors dynein, kinesin-1, and kinesin-2 along with dynactin, including the GFP-labeled dynamitin subunit (Fig. 1B). Thus, a complement of motor proteins remains tightly-associated with the vesicles throughout the purification.

Proteins known to localize to the axonal transport compartment, such as synaptotagmin and synaptophysin, are also enriched in the isolated vesicles, as well as markers for late endosomes/lysosomes, including LAMP-1 and Rab 7. In contrast, Rab 5, a marker for early endosomes, is not preferentially enriched in this vesicle preparation (Fig. 1B). We used electron microscopy to examine negatively stained preparations of vesicles bound to MTs (Fig. 1C). Vesicles had a mean diameter of 90.0 ± 2.9 nm (± SEM, n>300), consistent with previously characterized axonal transport vesicles (50–150 nm) [4, 5].

Photobleaching was used to quantify the number of bound GFP-dynamitin molecules stably associated with purified vesicles. GFP-dynamitin integrates into dynactin at a ratio of 2.2 labeled subunits out of four total dynamitin subunits per complex [2]. Quantitative stepwise photobleaching of dispersed vesicles statically bound to the cover glass results in a bimodal distribution (Fig. 1D,S1A). The majority of the vesicles (69%) were quenched in fewer than 10 steps, while 31% of the population was quenched in 10 to 20 steps. A fraction of the population was very bright (>20 bleaches); this likely correlates with vesicle aggregates seen by EM that did not bind well to MTs in motility assays, and were excluded from further analysis. Given a mean of 7.6 ± 3.0 (±SD) bleaching steps per dispersed vesicle, we estimate that on average, 3.5 ± 1.9 (±SD) dynactin molecules are bound to each vesicle.

The ratio of dynactin, kinesin-1, and kinesin-2 to dynein was measured by quantitative blotting of purified vesicle fractions, comparing multiple independent vesicle preparations to dilution series of purified recombinant standards (Fig. 1E,S1B,C). We measured a ratio of 1.3 ± 0.1 (± SEM, P=0.01, n=3 independent vesicle preps) for dynactin to dynein, similar to the recently reported 1:1 stoichiometry of dynein to dynactin in yeast [6]. We found an average ratio of 0.16 ± 0.02 (± SEM, P=0.01, n=3 preps) for kinesin-1:dynein and a ratio of 0.63 ± 0.04 (± SEM, P=0.004, n=3) for kinesin-2:dynein. Combined, quantitative western blotting and photobleaching yield an estimate of 2.8 ± 1.6 (± SD) dynein molecules, 3.5 ± 1.9 dynactin molecules, 0.45 ± 0.27 kinesin-1 molecules, and 1.7 ± 1.0 kinesin-2 molecules per vesicle. These measurements of the number of total cargo-bound motors are remarkably similar to previous estimates for the numbers of engaged motors driving vesicle transport in vivo, which range from 1 to 4 for kinesin and 1 to 5 for dynein [4, 7–11]. Together, these results suggest that multiple motor types are stably bound to vesicles moving along neuronal processes, and that oppositely-directed MT motors are bound to cargo at low, but similar, numbers.

To analyze the motility of GFP-labeled vesicles along rhodamine-labeled MTs in vitro, we used TIRF microscopy with 16.3-nanometer resolution (Fig. 2A). Purified vesicles show ATP-sensitive binding to MTs (Fig. 3A,S3A). AMPPNP induced the stable binding of >70% of the vesicles to MTs, consistent with the formation of a rigor bond via kinesin. Depletion of ATP also leads to the formation of a rigor bond between the vesicle and the MT, presumably due to trapping of dynein in the strongly-bound, no nucleotide state.

Automated tracking allows us to observe the motility of the vesicle population as a whole, as well as to categorize subsets of motility within the population. Tracking of individual vesicles moving along MTs demonstrates that vesicles frequently change direction, with 86% of purified vesicles switching direction during an observation time of 40 seconds (Fig. S3B). An individual vesicle may often exhibit intervals of stationary, diffusive, and processive movement. To characterize intervals of each type of bidirectional motility within runs, we used the absolute value of run length between reversals, |L_Rev| (see Fig. 2B). Characterization of motility based on |L_Rev| is consistent with the standard definitions based on mean-squared displacement [12], as shown in Fig. S2.

Next, we compared the movement of purified vesicles along MTs in vitro to the motility of vesicles in live cells. We labeled vesicles in primary cultures of cortical neurons with Lysotracker, which preferentially labels late endosomal/lysosomal compartments, as the purified vesicles used in this study are enriched for markers of this compartment. Again, we analyzed motility using automated tracking to obtain unbiased sampling with the same spatial and temporal resolution as our in vitro assays (Fig. 2C). We find that there is a close correlation between parameters of motility observed in reconstitution assays and endogenous vesicles moving in live cells (Fig. 2). Processive motility occurs with approximately equal probability in either direction for both Lysotracker-positive vesicles in live cells and purified vesicles in vitro (Fig. 2E). In vitro, average velocities of processive runs were similar in either direction (Fig. 2F,G); consistent with rates measured for soluble motor proteins in single molecule assays [13]. Velocities observed for processive runs of vesicles in cells were somewhat higher (Fig. 2F,G) possibly due to differences between the in vitro and intracellular environment, such as nucleotide concentration or ionic strength. However, like the purified vesicles in vitro, Lysotracker-positive vesicles move with comparable velocities in both directions along processes. Both in vitro and in cells, ~20% of the total vesicle population was stationary, while a large fraction displayed apparently diffusive movement, defined by short run lengths between reversals and a linear dependence of MSD on time (slope of 1 in the log-log plot) (Fig. 2D, S2).

To further probe the interactions of opposite-polarity motors involved in bidirectional motility, we observed the effects of inhibitory antibodies to dynein and dynactin on vesicle binding and motility. Addition of polyclonal antibodies to dynein heavy chain (pAb-DHC) or intermediate chain (pAb-DIC) reduced vesicle binding compared to controls, 25% (P<0.05) and 40% (P<0.005) respectively (Fig. 3B). Both pAb-DHC and a monoclonal anti-dynein antibody (mAb-DIC) increased the relative fraction of processive motility without affecting velocity (Fig. S3E,I). In contrast, addition of pAb-DHC resulted in a change in the directional bias of processive motility, with 68% of processive motion directed toward the MT plus-end in the presence of the antibody (Fig. S3G). Statistical analysis of data from individual tracks indicates that this is a significant change from controls (P<0.0085, n=43).

To examine the role of dynactin in vesicle motility, we used antibodies to the CAP-Gly domain of the p150^Glued subunit of dynactin (mAb-p150), and a polyclonal antibody that binds to the extended coiled-coil domain of p150^Glued (pAb-p150). Addition of mAb-p150 significantly decreased the binding of vesicles to MTs (~90% compared to controls, P<0.0001; Fig. 3C), suggesting that the CAP-Gly domain of dynactin may play a key role in mediating an initial association of the vesicle with the MT. A second inhibitory polyclonal antibody to dynactin (pAb-p150) also decreased vesicle binding to the MT (P<0.05). Of the vesicles that did bind in the presence of the anti-CAP-Gly Ab (mAb-p150), there was a noticeable increase in the extent of processive motility along the MT as compared to controls (Fig. S3F). MT pelleting assays indicate that this Ab effectively inhibits the direct, nucleotide-independent binding of dynactin to MTs (Fig. S3K,L), but does not block the dynein-dynactin association (data not shown). Thus, the CAP-Gly domain of dynactin is likely to mediate some of the diffusive motion of vesicles along MTs observed in our in vitro assay. siRNA depletion and reconstitution experiments have shown that loss of this domain does not affect organelle distribution or rates of organelle motility in nonpolarized cells grown in culture [14, 15]. However, the CAP-Gly domain may mediate an initial interaction of vesicles with MTs [16] and therefore enhance the efficiency of vesicle transport, a possibility consistent with the observations reported here.

Several antibodies to kinesin-1 were tested including an antibody known to inhibit kinesin-1 motility in vitro (SUK4, [17]). None of antibodies to kinesin-1 tested had an apparent effect on vesicle binding. In contrast, an antibody to kinesin-2 (K2.4) strongly inhibited the binding of vesicles to MTs (Fig. 3D). The population of isolated vesicles described here are Rab7-positive (Fig. 1B) and kinesin-2 has been identified as the primary anterograde motor for late endosomes [18–20]. These data suggest that the plus end-directed transport of these vesicles is driven primarily by kinesin-2.

Bidirectional motility along a MT has been modeled as a stochastic “tug-of-war” [21], whereby net transport is a consequence of the force-dependent dissociation kinetics of opposing motors in the absence of external regulation. By varying the number of plus- and minus-end directed motors or the kinetic parameters of the motors, the predicted patterns of transport exhibit a range of motility regimes; including bidirectional motility and unidirectional movement. We fit a mathematical model of bidirectional motility to our data using experimentally defined parameters for transport mediated by kinesin-1 and dynein [21] or kinesin-2 and dynein (Table S1), varying only the number of active plus- and minus-end directed motors. The results from this modeling predict that directionality is strongly modulated by the ratio of oppositely directed motors (Fig. 4A).

The predictions of the model were then compared to our experimental observations of processive motility (Fig. 4B, control); the model best describes the data when transport is driven by a ratio of 7 dynein to 1 kinesin-1 motor, or 3 dynein to 2 kinesin-2 motors. These predictions for the number of active motors are remarkably consistent with the experimental estimates for the ratio of total motors bound to vesicles, of 6:1 for dynein:kinesin-1 and 3:2 for dynein:kinesin-2.

Several models have been proposed to describe the bidirectional motion of cargo in the cell [8]. Muller et al. [21] developed a mathematical model of a stochastic “tug-of-war” in which bidirectional transport is regulated by the force-dependent kinetics of opposite-directed motors, in the absence of external factors. Here we show that this model accurately describes the fast switching between plus- and minus-end directed transport exhibited by vesicles in vitro (Fig. 4). Further, we find that the number of active kinesin and dynein motors predicted by the model closely correlates with estimates of the total number of bound motors through photobleaching and quantitative western blotting. Thus, the model suggests that bidirectional motility directly results from the unregulated interactions of oppositely-directed motor proteins simultaneously bound to the same cargo. In striking agreement with our findings, Soppina et al. [11] used optical trapping to estimate the number of motors driving the transport of Dictyostelium endosomes; 4–8 “weak” dynein motors (~1 pN stall force) and a single “strong” kinesin motor (~5–6 pN stall force) drive each ~500–1500 nm endosome. However, in contrast to the behavior of the larger Dictyostelium endosomes, we did not observe “triphasic” reversals and fission events in our purified vesicles in vitro or Lysotracker-positive organelles in live cells, suggesting these characteristics may be specific to certain cargos.

Analysis of the motility of Lysotracker-positive organelles in live cells indicates that bidirectional motility, characterized by frequent directional switches and pauses, is a common transport mechanism for at least some types of intracellular cargo. Both in vitro and in cells, we also see that significant fractions of the total vesicle population are either paused or moving diffusionally over short distances. Tracking of vesicles in live cells also reveals that in addition to bidirectional motility, a subset of vesicles move primarily unidirectionally, in either the anterograde or retrograde directions (data not shown). This subset of fast, directed motility observed in the cell was not seen in vitro, suggesting that additional regulatory factors are required to reconstitute long, unidirectional runs.

Intracellular regulation may occur through several pathways. Enhanced recruitment of one motor type to a cargo would result in a directional bias. Alternatively, scaffolding molecules have been proposed to regulate motors on cargo. Directional switching may also be regulated at the level of motor activity, whereby the activity of the motors is stochastically modulated to affect the net direction of motion (Fig. 4D). Kinesin-1 can fold into an inhibited conformation [22]; intramolecular autoinhibition through folding has also been demonstrated for other members of the kinesin superfamily [23]. While binding to cargo was initially proposed to unfold kinesin and activate motility, recent progress indicates that activation occurs in multiple steps [24, 25] and thus even kinesin motors tightly associated with cargo may alternate between active and inactive conformations.

Unlike for kinesin, no self-regulation scheme has been demonstrated for dynein-dynactin, suggesting only a single motor type may be actively regulated. In a dynein-dragging model, kinesin is regulated by auto-inhibition but dynein remains constitutively active. Because dynein is both less powerful and more flexible than kinesin motors, and can stay associated with the MT track even in the presence of obstacles [26], dynein could potentially enhance the processivity of cargos moving in either direction. Importantly, force measurements indicate that the stall forces of either kinesin-1 (~6 pN; reviewed in [1]) or kinesin-2 (~4 pN; our unpublished data) are significantly stronger than the opposing force of a mammalian dynein motor (~1.1 pN; [27]). Thus at a ratio of 1 kinesin-1 to 7 dynein motors or 2 kinesin-2 to 3 dynein motors, forces in either direction are balanced, and would be predicted to result in a tug-of-war with approximately equal fractions of plus- and minus-ended motility such as we observed.

01

02

03

04

05

06

07

08