Nanoparticle-assisted optical tethering of endosomes reveals the cooperative function of dyneins in retrograde axonal transport

Dynein-dependent transport of organelles from the axon terminals to the cell bodies is essential to the survival and function of neurons. However, quantitative knowledge of dyneins on axonal organelles and their collective function during this long-distance transport is lacking because current technologies to do such measurements are not applicable to neurons. Here, we report a new method termed nanoparticle-assisted optical tethering of endosomes (NOTE) that made it possible to study the cooperative mechanics of dyneins on retrograde axonal endosomes in live neurons. In this method, the opposing force from an elastic tether causes the endosomes to gradually stall under load and detach with a recoil velocity proportional to the dynein forces. These recoil velocities reveal that the axonal endosomes, despite their small size, can recruit up to 7 dyneins that function as independent mechanical units stochastically sharing load, which is vital for robust retrograde axonal transport. This study shows that NOTE, which relies on controlled generation of reactive oxygen species, is a viable method to manipulate small cellular cargos that are beyond the reach of current technology.


Conjugation of commerical iron oxide nanoparticles (INPs) with WGA and fluorophores:
Single core superparamagnetic iron oxide nanoparticles (INPs, 30±2.5nm, 50±5nm) with very low size tolerances were purchased from Ocean Nanotech. These INPs, with carboxylic acid surface functionalization, were coated with fluorescent Alexa555-WGA by standard carbodiimide chemistry to obtain fluorescent INP30-WGA and INP50-WGA. This reaction was carried out at 1:1000 ratio of INP:WGA, and multiple magnetic separations were done in a high gradient magnetic separator (SuperMag, Ocean Nanotech) to discard unconjugated WGA.

Synthesis of biotinylated silica-coated iron oxide particles and WGA conjugation:
Silica-coated core-shell nanoparticles of tunable size, with multiple iron oxide cores and quantum dot shells were synthesized as described earlier 1 . Briefly, 1 mL chloroform solution containing QDs (655nm emission, 9.0 ± 0.4 nm) and INPs (5.9 ± 0.3 nm) was injected into 1 mL DTAB aqueous solution (20 mg/mL in Nanopure water). The solution was thoroughly mixed by vortexing for 5 second. After removing the chloroform from the mixture by blowing Ar at room temperature, a clear QD-INP-micelle aqueous solution was obtained. This QD-INP-micelle solution was swiftly injected into 5mL poly(vinylpyrrolidone) (PVP, MW 55000) ethylene glycol (EG) solution (2 mM) under vigorous stirring (700 rpm for 30 min) at room temperature. 2 mg of the resulting core-shell superparticles (CS-SPs) were isolated by centrifugation and redispersed in 20 mL ethanol. Under vigorous stirring, 3 mL deionized-water (DI-water) and 1 mL NH 3 ·H 2 O solution was added dropwise into the reaction solution, followed by the addition of 50 µL TEOS. The final solution was kept stirring at room temperature for 20 min. The resulting silica-coated CS-SPs were isolated by centrifugation, washed three times with DI-water and finally dissolved in 2 mL DI-water.
Biotinylation: The silica-coated CS-SPs in DI-water (2 mL) were diluted in 10 mL DI-water. Under magnetic stirring, 50 µL NH 3 ·H 2 O solution and 20 µL TEOS were added into the solution, sequentially. After the mixture solution was heated to 70 °C, 1 mL Biotin-PEG-silane (MW 5000) aqueous solution (0.02 M) was dropped in. The reaction solution was kept stirring at 70 °C for 3 hrs and then stirred at room temperature for another 15 hrs. The resulting biotinylated silica-coated CS-SPs were isolated by centrifugation, washed three times with nanopure water and finally dissolved in 1 mL phosphate buffer solution (PBS, 1X) and stored at ~4 °C. Figure S1a shows the dynamic light scattering measurement of the silica-coated nanoparticles, which gave an average hydrodynamic size of 164nm. We functionalized these particles with streptavidin using a stoichiometric amount of streptavidin and subsequently conjugated with biotin-WGA (1:300 ratio) for the axonal transport studies.

Synthesis of biotinylated polymer nanoparticles (b-PN) and WGA conjugation:
The b-PNs are synthesized according to previously published procedures [2][3][4] . Briefly, a mixed THF solution (1 mL) containing poly(9,9-dioctylfluorene-alt-benzothiadiazole) (F8BT) (0.25 mg/mL) and 1,2-distearoyl-sn-glycero-3-phosphoethanolamine-N-[biotinyl(poly(ethylene glycol))-2000] (ammonium salt) (DPSE-PEG-Biotin) (2.5 mg/mL) was used to prepare b-PNs by rapidly injecting it into distilled-deionized water (9 mL) under continuous sonication with a microtip-equipped probe sonicator. After sonication, THF was evaporated under nitrogen atmosphere. The aqueous solution was filtered through a polyethersulfone (PES) syringe driven filter (0.22 µm, Millipore), and washed three times using a 30K centrifugal filter unit (Millipore). The nanoparticle solution was finally concentrated to 0.08 mg/mL (based on the mass of PN) by ultrafiltration and stored in dark at 4 °C. Figure S1b shows the absorption and emission of the b-PNs. Figure S1c shows the dynamic light scattering measurement of the b-PNs, which showed an average hydrodynamic size of 31.3nm with a polydispersity of 0.176. We functionalized the b-SPNs with streptavidin using a stoichiometric amount of streptavidin and conjugated with biotin-WGA (1:5 ratio) to obtain fluorescent PN-WGA for axonal transport studies.

Oblique illumination imaging of axonal transport in microfluidic DRG neuron culture:
The use of microfluidic devices for primary neuron culture and imaging axonal transport had been documented by us elsewhere 5,6 . In this work, we used mature DRG neuron cultures (7-10 days old) for the QD-WGA and INP-WGA transport studies.
An inverted microscope (Nikon Eclipse Ti-U), customized with dual laser (488, 561nm) laser excitation and three-color imaging, was set for oblique illumination. Briefly, the laser beam was expanded to 3 cm diameter and focused at the back focal plane of the microscope objective (CFI Plan Apo TIRF, 100X/1.49NA) for a collimated illumination with beam waist ~120µm. Starting from total internal reflection geometry, the incident angle was gradually lowered below the critical angle till the QD-endosomes within the axons are made visible by the oblique angle illumination. Since the microfluidic channels in our device are 3 µm high, a penetration depth of 1-3 µm is sufficient to illuminate axonal fluorophores in multiple focal planes. The fluorescence collected by the objective was spectrally split into three different bands by two sets of dichroics. These bands were relayed and focused onto a third of an EMCCD sensor (Andor Ixon DU-897).
The axon terminals in the mcirofluidic culture are incubated with Alexa-WGA (0.5nM) or QD-WGA (0.5 nM) or INP-WGA (1-2nM) for 45 min, which was then washed off by the culture medium. Shortly before imaging, the culture medium was replaced by CO 2 independent medium and the culture was imaged on a water-heated custom microscope stage set to maintain the culture at 37 ºC. Imaging was started typically 2 hrs after the incubation start and restricted to < 45 min session. Time-lapse movies of endosome transport were acquired at 10-150 frames per second and the imaging laser power is varied in the range 0.13-11 mW (i.e. 1.4-114 W/cm 2 , for a beam waist ~120 µm) using neutral density filters. For imaging DiI stained retrograde endosomes in axons, we incubated the axon terminals with ~2µM DiI for 10min, and the imaging was done 1.5hrs after incubation as described above.

Single particle tracking of the endosome transport in time-lapse movies:
Semi-automated custom Matlab software is used to extract endosome trajectories (X(t),Y(t)) from the time lapse movies of axonal transport as described earlier 7 . Briefly, we used 2D-Gaussian fitting adapted to detect the locations of endosomes (with Alexa/QD/INP) in each movie frame. Intensity and signal-to-noise ratio cutoffs were adjusted to achieve localization accuracy < 20 nm. We then used a particle-tracking algorithm, adapted from Jaqaman et al. 8 , to link these detected locations into endosome trajectories. The flux of endosome transport, kept sparse by controlled incubation conditions, lead to minimal tracking errors. Occasional errors associated with crossing trajectories are corrected by manual inspection in the final step of processing. For each endosome trajectory, we extracted the underlying microtubule track (Q x ,Q y ) using an edge-based tracing algorithm. Using this microtubule track, the endosome trajectory is converted from camera pixels (X(t),Y(t)) to microtubule coordinates (Q || (t),Q ⊥ (t)) representing the motion parallel and perpendicular to the microtubule 7 . Since the microfluidic channels (and hence the axons) are aligned along the Y-axis of the camera, Y(t) is an approximation for Q || (t).

Kymograph processing to extract the endosome jumps from time-lapse movies:
Considering the lower signal-to-noise ratio of the INP-WGA endosomes imaged at 150 fps, we used a semi-automated kymograph-based analysis to accurately capture the INP-endosome detachments (Fig. S2). First, we projected the intensities of all frames in the time-lapse movie to obtain a time-projection image, which gives the diffraction limited intensity profiles of the microtubular tracks of INP-WGA endosomes in axons (Fig. S2b). Then we manually traced the centerline of each intensity profile, which is refined by an automated centerline tracer, to give the underlying microtubule track Q(x,y) (red curve in Fig. S2b). We then generated the kymograph I(Q,t), which depicts the intensity variation along the microtubular track Q as a function of timeframe, due to the motion of INP-WGA endosomes (Fig. S2c). We fit the intensity profile at each timeframe I(Q) with a 1D Gaussian mixture model to obtain the INP-WGA endosome locations along the microtubular track at each time frame (Fig. S2d). These locations are then connected by 1D single particle tracking to obtain the INP-WGA endosome motion parallel to the microtubule i.e. Q || (t) (Fig. S2e).
The position uncertainties in Q || (t) (Fig. S2e, parenthesis) are computed from the integrated intensities of Gaussian fits using the calibration equation (Fig. 3b of main text). The calibration equation is obtained from the localization accuracy of INPs (with a wide range of intensities) stuck on coverslips as follows. We tracked the INP positions for over 3s (>400 frames) using the kymograph analysis described above. The standard deviation in the NP positions represented the localization inaccuracy in our system (imaging and processing). Figure 3b in main text shows the localization accuracy as a function of the integrated intensity of the INPs and the approximate power-law calibration curve. We selected the intensity cutoff to achieve < 25nm accuracy.

Solutions for the detachment model in Equation (2) of main text:
Ignoring the fast inertial and Brownian fluctuations compared to the motor driven endosome motion, Equation (1) in the main text simplifies to Equation (S1) (Equation (2) in main text).
Pre-detachment solution: The stochastic load sharing among dyneins leads to complex predetachment stall profiles of endosomes under load that cannot be described analytically. Under mean field assumption, where the load is shared equally by all dyneins, we approximated the average endosome velocity by a simple linear force-velocity relation in Equation (S2).
where v m is the unloaded velocity of the endosome in Stokes regime, k is the stiffness of the elastic tether, (q-q 0 ) is the stretch of the tether, nF s is the cumulative stall force of 'n' dyneins (with unitary stall force F s ) moving the endosome. The solution of S2 is given by Equation (S3), which is only applicable under the mean field assumption.
Post-detachment solution: Equation (S1) simplifies considerably after the detachment of leading-dyneins from the microtubule with F motor =0. The endosome recoil in the medium with friction coefficient γ is described by the solution given in Equation (S4), where V detach is the instantaneous recoil velocity of the endosome at detachment, and k/γ is the damping constant of endosome recoil.

Fitting the endosome jump to the motor-detachment model:
The primary objective of fitting an endosome jump is to obtain the instantaneous detachment velocity V detach , which is proportional to the cumulative stall force of leading-dyneins. In principle, V detach can be obtained by fitting the post-detachment recoil profile to Equation (S4). However, it is important to accurately identify the detachment time (t d ) and position (q s ) for accurate extraction of V detach by model fitting (see Fig. 3a in main text). Fitting the predetachment stall and post-detachment recoil profiles simultaneously, provides reliable estimation of (t d , q s ) as the intersection point of the fit curves.
Though Eq. S3 is not applicable to stall profiles of dyneins with stochastic load sharing, it is useful in fitting the gradually stalling segment prior to detachment as in Fig. 3c of main text. However, the parameters A and B obtained by fitting Eq. S3 to a partial range of stall profiles are not relevant for the interpretation of v m (unloaded dynein velocity), k (tether stiffness) and F s (unitary dynein stall force), as given in Equation (S3). In this analysis, Equation (S3) merely serves as a smoothly varying polynomial that aids in the objective estimation of (t d , x s ) in conjunction with Equation (S4), which is used to fit the post-detachment recoil profile.
We fit each of the INP100-endosome jumps to Equations (S3), (S4) by χ 2 minimization. The global fitting of the pre-detachment and post detachment curves weighted by the position uncertainties in INP100-endosome position yields t d , x s , V detach and k/γ. In many cases the endosome recoil showed sudden deviations from exponentiality within 40-70ms of detachment, presumably due to fast dynein rebinding or collisions with local cytoskeletal features. Therefore, we typically fit ~40-70ms (4-10 time points) of the post-detachment recoil profile depending on the jump profile. Figure S3 shows a random selection of a few INP100-endosome jumps and the model fits.
The damping parameter k/γ is expected to be constant for a given tethered endosome exhibiting jumps within a narrow stretch of the axon. However, as shown for the endosome in Fig. 3 of the main text, we notice a dispersion of k/γ values even for the same endosome due to A) the deviations from exponentiality in jumps and B) the standard errors in fitting a limited range of data (4-10 time points) within localization uncertainties. In order to mitigate these factors and improve the accuracy of the fit parameters (V detach and k/γ) we only selected the jumps that resulted in minimal standard errors (<15% of the fit parameter) in fitting V detach . These constituted 32% of the total number of jumps identified in our data (306 out of 961 jumps).

Quantification of retrograde transport velocities from endosome trajectories
We used a Bayesian parsing approach reported earlier 7 to identify the retrograde moving, pausing, and anterograde moving phases within each endosome trajectory (>10sec duration). Briefly, we parsed each endosome trajectory (10fps imaging, Q || (t) obtained by particle tracking and processing described in 1.6) into a series of constant velocity segments (minimum length 0.3s imposed). We then clubbed consecutive segments in the same direction to obtain retrograde runs, pauses, and anterograde runs. A run is a sustained period of motion (> 0.5s duration, ≥ 0.2 µm/s velocity) in the same direction that ends with a pause or a direction reversal. A pause is defined as a motionless segment > 0.5s duration with < 0.2 µm/s velocity.
In order to analyze the retrograde run velocity distribution, we acquired a large dataset of QD-WGA endosome trajectories (10fps imaging, N=1141). Figure S4a shows the distribution of the average speed of QD-WGA endosomes including pauses and direction reversals in motion. Figure S4b shows the distribution of retrograde run velocities of the QD-WGA endosomes. We do not notice any obvious multimodal pattern that is indicative of the number of dyneins on the endosomes.

Occasional jumps observed with retrograde QD-endosomes and DiI-endosomes in axons:
The frequency of endosome jumps/stalling showed a dramatic increase with laser power in the case of INP-endosomes. The endosome jumps were also occasionally exhibited by retrograde QD-WGA-endosomes and DiI-stained retrograde endosomes (Fig. S5, Movie S5, Movie S4). However, we do not notice any laser power dependence in the range of 1.4-114 W/cm 2 . Further, many QD-and DiI-endosomes resumed the retrograde motion after exhibiting a series of jumps (Movie S5). The characteristics of stall profiles and jumps are similar to that of INP-endosomes.

Analysis of endosome deformation during jumps:
It had been shown that endosomes of size > 500 nm stretched ~25% of their size during tugsof-war between dyneins and kinesins 9 . Axonal endosomes (50-250 nm) are expected to be more rigid due to their smaller size. In order to quantify the extent of endosome stretching at stall (i.e F opp =F dyneins ), we analyzed the jumps of DiI-stained fast retrograde endosomes. We found that DiI endosomes remained nearly diffraction limited and showed minimal deformation during jumps. Figure S6a shows the kymograph of DiI endosome jumps of ~1µm size. The peak endosome intensity and the Gaussian width (HWHM~350nm) remained nearly constant during the jump (Fig. S6c). Note that the HWHM is not an accurate metric for the absolute vesicle size but is a useful qualitative measure to assess the vesicle deformation at varying loads. On average the HWHM of DiI endosomes increased by < 7% from baseline to stall (Fig. S6e) and this increase contributed < 4% to the endosome jump size (Fig. S6f). This indicates that endosome deformation is contributing minimally to the elastic stretch responsible for the jumps in NOTE.

Considerations of nanoparticle heating under continuous laser illumination:
The role of local temperature spikes, due to the photothermal response of INPs, in causing the endosome jumps can be ruled out due to the following reasons. First, the local heating of an isolated INP by continuous laser excitation is negligible due to rapid diffusion of heat (<100ns) in the medium 10 . The steady state surface temperature of a 50 nm INP in water is estimated to be < 0.1K higher than that of the medium (see below). Second, the dynein stepping rate would be sensitive to INP heating 5,7 by virtue of rapid heat diffusion and the distance (<100nm) between the INP-endosome and the dynein head domains. However, we did not see any increase in endosome speed corresponding to endosome jumps, which suggests that the local temperature changes are negligible.
The thermal relaxation for a spherical particle with radius R is given by τ= R 2 /D, where D is the thermal diffusivity of the medium (10 -7 m 2 /s for water) 10 . For a particle of size 10-100nm the relaxation is fast and within 1-100ns. The steady state temperature difference at a particle surface upon continuous laser illumination is given by dT = σ abs I 0 /(4πκR), where σ abs is the absorption cross section of the particle, I 0 is the laser irradiance, and κ the thermal conductivity of the medium 10,11 . Given a maximum possible absorption cross section of 1e-14m 2 for an iron oxide nanoparticle (R=50nm), and the maximum laser irradiance tried in our imaging system (1150kW/m 2 ), we estimate dT~0.03K.      . The HWHM, which is a qualitative measure of the endosome stretching (beyond diffraction limit) along microtubule, remained nearly constant during the jumps. d) The HWHM of jumps computed at the baseline (blue arrows in (b) and the stall position (red arrows in b) of jumps. e) 74% of jumps (N=31) showed an increase (average < 7%) in HWHM from baseline to stall position, which is shown as a function of the jump size. f) Relative change in HWHM as a function of jump size shows that endosome stretching contributes <4% on average to the jump size.