Figures and data
Experimental Design and Raw Data from Motor-DNA Tensiometers.
(A) Schematic of a motor-DNA tensiometer, consisting of a dsDNA (burgundy) connected on one end to a kinesin motor through a complimentary oligo (blue), and on the other end to the MT using biotin-avidin (tan and gray, respectively). A Qdot functionalized with GFP binding protein nanobodies is attached to the motor’s GFP tag and used to track motor position. (Not to scale; motor and Qdot are both ∼20 nm and DNA is ∼1 micron) (B) Predicted force extension curve for a worm-like chain 3009 bp dsDNA based on a 50 nm persistence length. We note that our analysis of family-dependent mechanochemistry does not depend on the precise shape of the force-extension curve, only that the motors reach stall. (C) Representative kymographs of motor-DNA tensiometers for kinesins-1, -2 and -3. (D) Enlarged kymograph showing diffusion around the origin, ramp, and stall. (E) Example distance vs. time trace (kinesin-3), highlighting detached durations (red), ramps and stalls (black) where the motor has pulled the DNA taut, and transient slips during stall (red). (F-H) Representative distance vs. time plots for kinesin-1 (F), kinesin-2 (G) and kinesin-3 (H), corresponding to the kymographs in (C). Further examples are shown in Figure 1 – figure supplement 3.
Tensiometer Stall Durations Indicate Catch-bond Behavior for Kinesin-1 and -2.
Tensiometer stall durations are plotted for (A) kinesin-1 (blue), (B) kinesin-2 (purple), and (C) kinesin-3 (green). Unloaded run durations for each motor are plotted in gray. Distributions were fit with a single exponential function using MEMLET to generate time constants, representing the mean durations. (D) Comparison of unloaded and stall durations for the three motors, with error bars indicating 95% CI. Stall durations >20s were excluded from the fit (three events for kinesin-1 and two events for kinesin-2). All fit parameters are given in Table S1 and bi-exponential fits of all data including >20 s are shown in Figure 2 – figure supplement 1.
During Ramps, Kinesin-3 Detaches More Readily Than Under Zero Load.
Unloaded, ramp, and stall duration parameters were estimated using a Markov process model, coupled with Bayesian inference methods. Curves show the posterior probability distributions of the duration parameters for (A) kinesin-1, (B) kinesin-2 and (C) kinesin-3. Bars below each peak indicate the 95% credible regions for the ramp (green), unloaded (gray) and stall (blue) duration parameters. Notably, the estimated ramp durations are larger, the same, and smaller than the unloaded run durations for kinesin-1, -2, and -3, respectively. For the unloaded and stall durations, this estimation method produces almost identical values as the maximum likelihood estimates in Figure 2 (values provided in Table S1).
Restart Kinetics for Kinesin-1, -2 and -3.
(A) Fraction of slip, fast rebinding, and slow rebinding events for each motor, with example kymographs for each (top; scale bars are 0.5 μm and 0.2 s). Solid colors indicate slips during stall, where the motor resumes a new ramp within a single frame (∼40 ms), crosshatching indicates rapid reattachment events (100 ms) following fall to baseline, and open bars indicate slow reattachment events with >100 ms fluctuations around baseline. (B) Kinesin-3 stall durations, with unloaded run times in gray, stall durations terminated by slips in light green, and stall durations terminated by falling to the baseline (ignoring slips) in dark green. Unloaded and stall durations (replotted from Figure 2) were fit with single exponential functions in MEMLET. Stall durations ignoring slips were fit with a bi-exponential by least squares (ρ1 = 2.01 s [95% CI: 1.52, 2.35 s], A1 = 0.66 s [0.54, 0.83 s], ρ2 = 11.0 s [9.18, 13.70], A2 = 0.33 [0.23, 0.48]). Weighted average of the two time constants is displayed in plot for comparison to other time constants. Similar results for kinesin-1 and -2 are shown in Figure 4 – figure supplement 2. (C) Comparison of stall durations for kinesins -1, -2 and -3 with slips observed as stall terminations or ignored. (D-F) Distribution of restart times for each motor fit to a tri-exponential (least squares). Confidence intervals of parameters determined by bootstrapping with 1000 iterations are given in Table S3.
Chemomechanical Model of Proposed Catch-bond Mechanism.
(A) Diagram of kinesin chemomechanical cycle model consisting of strongly- and weakly-bound states that make up the stepping cycle, and slip and detached states that terminate runs and stalls. Note that two pathways of detachment from the slip state (and reattachment) are incorporated into the model, but only one pathway is shown for simplicity (see Supplementary Methods for details). (B) Table of rate constants used to simulate unloaded and stall durations and restarting times. All rate constants are derived from fits to experimental data, as described in Supplemental Methods. kS-W and kslip depended exponentially on load 
Fit results for unloaded, ramp, and stall durations.
Predicted force imposed on the motor during the ramp phase
Reattachment Statistics.
Kymographs of unloaded GFP-labeled kinesins.
GPF-labeled motors conjugated to their complimentary oligo were visualized via TIRF at a concentration of 1 nM at 5 fps. No neutravidin, Qdot or DNA are present in these unloaded controls. A fraction of kinesin-3 unloaded run durations were limited by the length of the microtubules, but fitting to a model that took into account missed events gave a similar mean duration as an exponential fit, and so no correction was made (Table S2).
Distribution of initial motor binding positions.
The initial positions were measured for events where the motor was clearly dissociated from the microtubule and fluctuated around origin on its DNA tether (N=141). The zero position was determined as the center point around which the detached motor fluctuated. The initial motor binding position was determined by the first start point of a ramp. The width of the gaussian distribution, quantified by the SD, demonstrates the large search space of the motor attached to the flexible ∼1-micron dsDNA tether. The mean of +11 nm likely results from some of the motors moving before the first frame acquisition, giving a small positive bias. The larger population seen at >+200 nm relative to <-200 nm may result from dissociation of motors that bind under assisting loads (negative displacements) and strengthening of motors that bind under hindering loads (positive displacements). Data are from kinesin-1 and kinesin-3 tensiometers.
Further kinesin tensiometer examples.
Distance versus time plots of (A) kinesin-1, (B) kinesin-2 and (C) kinesin-3 traces. Notably, some stalls are very stable, whereas other (particularly for kinesin-2 and kinesin-3) show fluctuations, presumably due to small slips and backstepping at stall. Other features to note include pauses in the motile segments, small changes in velocity, and repeated ramps for kinesin-2. Roughly 20% of tensiometers extended less than the expected 1 μm distance, stalling repeatedly at 500 nm or 800 nm. These apparently shorter DNA strands may result from DNA secondary structures or from primer binding at a secondary sequence. After in-depth comparison of the data we found that the stall durations, reattachment rates, and starting positions were unaffected by the shorter DNA length and thus included the shorter tensiometers in our data set.
Long stall durations are observed in the absence of Qdots.
Because the Qdots used in our experiments are functionalized with multiple GBP nanobodies, there is the possibility that the long stall durations observed were caused by multiple motors bound to the Qdots. To test this, we ran a control experiment where, instead of labeling the motors with Qdots, we fluorescently labeled our dsDNA by incorporating 5% dCTP-Cy5 during the PCR reaction to create fluorescent dsDNA. By removing Qdots from the system, the potential for multimotor events is eliminated. (A) Diagram of control experiment. (B) TIRF kymographs of representative bright kinesin-1 DNA tensiometers collected at 5 fps, showing the typical extension and stall profiles we observed in Figure 1, with the difference that the entire dsDNA is visualized rather than the Qdot. (C) CDF plot of the fluorescent DNA tensiometer stall durations of kinesin-1, fit with a single exponential function using a maximum likelihood estimator (MEMLET). Importantly, kinesin-1 continued to have much longer stall durations than its unloaded run durations (gray points with fit; reproduced from Fig. 2A), ruling out multi-motor interactions as the cause of the long stall durations. The longer stall durations here (5.26 s) compared to the Qdot stall durations (3.01 s; Fig. 2A) is attributed to the 5 fps frame rate used in here, which makes it more difficult to detect short slip events that are observed with the 25 fps Qdot movies.
Bi-exponential fits of stall durations reveal a longer duration sub-population for kinesin-1 and kinesin-2.
Tensiometer stall durations of (A) kinesin-1 and (B) kinesin-2 were fit with a biexponential function using a maximum likelihood estimator, MEMLET (https://michaelswoody.github.io/MEMLET/). Unloaded run durations are shown in gray for reference. Time constants (τ), relative amplitudes (A) and 95% confidence intervals for time constants are given in the accompanying tables. The rationale for why the motors would have two time constants is not clear, but it may suggest two alternative detachment pathways. Notably, both time constants are longer than the unloaded binding duration for both motors. Kinesin-3 stall durations were well fit by a single exponential function (see Fig. 2C).
Kinesin-1 control experiments.
(A) Kymograph of GFP-labeled kinesin-1 motors (green) transporting 3 kb Cy5-labeled dsDNA (red) along an unlabeled microtubule. Most motors do not have DNA attached, the one with DNA attached appears qualitatively similar to others. At top right, horizontal red line denotes free Cy5-labeled DNA diffusing near the microtubule for one frame. (B) Run durations for control experiments. Green circles: Unloaded GFP-labeled kinesin-1 on unlabeled microtubules; reproduced from Fig. 2A. From MEMLET fit, τ [95% CI] is 1.04 s [0.78, 1.31]. Red circles: Unloaded GFP-labeled kinesin-1 on biotinylated microtubules functionalized with neutravidin. Data were taken from excess non-DNA bound motors in DNA tensiometer experiments, thus matching the microtubules used for stall duration measurements. τ [95% CI] is 0.92 s [0.79, 1.04]. The similar run duration indicates that differences seen between unloaded run durations on unlabeled microtubules and durations of ramps and stalls in DNA tensiometer (on biotin-neutravidin functionalized microtubules) should not be due to effects of biotin-neutravidin. Blue circles: GFP-labeled kinesin-1 attached to 3 kb Cy5-labeled dsDNA moving on unlabeled microtubules (example shown in panel (A)). τ [95% CI] is 1.56 s [1.23, 1.90]. Thus, kinesin-1 transporting a 3 kb dsDNA has ∼50% longer run durations than kinesin alone. We interpret this difference to be due to the slower diffusion of the dsDNA, which enables undetected motor rebinding events that elongate the run length. Note that during ramps and stalls, any transient motor detachments should be detectable by a recoil of the stretched dsDNA, minimizing the effect of this slower diffusion of the dsDNA.
Correcting kinesin-3 stall duration for undetected slips.
Because of the frequent slips during kinesin-3 stalls, we asked whether undetected slips below our 60 nm threshold might be leading to an overestimation of the kinesin-3 stall duration. To estimate the fraction of kinesin-3 slip events below the 60 nm threshold, we fit an exponential function to the distribution of slip events > 60 nm at stall:
Because the data diverged at long distances, we only fit up to 400 nm and obtained dslip = 102 nm (95% CI: 98-106 nm). The estimated fraction of slip events larger than 60 nm is: 
Thus, if undetected slips are taken into account, the number of stall termination events for kinesin-3 should be corrected upwards by a factor of 1.42, or equivalently the estimated stall duration of 1.89 s should be corrected to: 1.89 s/1.42 = 1.33 s. Based on the dslip fit uncertainty (98-106 nm), the 95% CI is 1.30-1.35 s. We note that this corrected kinesin-3 stall duration of 1.33 s is still well above the 0.75 s (95% CI: 0.64-0.87 s) for the ramp duration parameter in Fig 3 (Table S1). Thus, the kinesin-3 stall duration is longer than the kinesin-3 ramp duration, meaning that force slows detachment. There were only a handful of slips from stall for kinesin-1 and kinesin-2, precluding estimation of the fraction of missed slip events for these motors. However, the fraction is expected to be small and to not affect our conclusions.
Stall durations are longer when slips are ignored.
(A) Kinesin-1 stall durations, with unloaded run times in gray, stall durations terminated by slips in light blue, and stall durations terminated by falling to the baseline (ignoring slips) in dark blue. Unloaded and stall durations (from Figure 2) were fit with single exponential functions in MEMLET. Stall durations ignoring slips were fit with a bi-exponential by least squares τ1 = 1.35 s [95% CI: 0.81, 1.91 s], A1 = 0.27 s [0.07, 0.62 s], τ2 = 4.99 s [4.45, 5.93], A2 = 0.77 [0.46, 1.00]. Weighted average of the two time constants is displayed in plot for comparison to other time constants. (B) Kinesin-2 is plotted similarly in purple, fitting of stall duration ignoring slips resulted in τ1 = 1.91 s [95% CI: 0.92, 2.78 s], A1 = 0.32 s [0.11, 0.79 s], τ2 = 7.60 s [6.74, 9.05], A2 = 0.68 [0.41, 1.00].
Model results incorporating both vertical and horizontal forces.
(A) Model results incorporating load-dependent kS-W and load-independent kslip. Very long stall durations are predicted that don’t agree with results, which motivated incorporating a load-dependence into kslip (Figure 5). (B) Diagram of single-bead optical tweezer experiment using 440 nm diameter bead and 35 nm long motor, which results in a 60° motor angle. With this geometry, a 6 pN stall force parallel to the microtubule is associated with a 10 pN vertical force on the motor. Note that figure is not to scale. (C) Evaluation of effects of vertical and horizontal load from different models. 1One-bead experimental data at zero and 6 pN from Andreasson 113,114 2Our model incorporating both horizontal and vertical forces into kslip (using δ‖ = 1.61 nm and δ┴ = 1.58 nm) is able to match the single-bead stall duration. 3Khataee model 114 also accounts for single-bead unloaded and stall durations. 4Kinesin-1 DNA tensiometer data. 5When modeling our DNA tensiometer data for kinesin-1, our model using only F‖ is able to match the stall duration, whereas the Khataee model101 overestimates the stall duration.
