Reply to "Acquisition Frame Rate Affects Microtubule Plus-End Tracking Analysis"

Danuser replies:

Nicovich and Zhou¹ remind researchers in the cytoskeleton field of a well-known but often forgotten issue: parameters describing growth and shrinkage dynamics of cytoskeletal polymers—here, microtubules—vary with the frame rate of the image sequence. Given a time interval Δt between frames, any growth or shrinkage event lasting less than 2 × Δt will be detected ambiguously. Whether it is still detected as a growth or shrinkage event depends on the actual duration and rate of polymer assembly or disassembly. Regardless, temporal undersampling causes underestimated rates of microtubule growth and shrinkage and underestimated rates of catastrophe (switch from growth to shrinkage) and rescue (switch from shrinkage to growth). Although the undersampling usually does not preclude detection of changes in microtubule dynamics between molecular backgrounds as long as imaging conditions are identical, measurements originating from image sequences sampled at, for instance, 5 seconds per frame cannot be compared to measurements from image sequences sampled at 1 second per frame.

To make their case, Nicovich and Zhou use the plusTipTracker software², which measures microtubule dynamics by tracking fluorescent particles that mark the selective association of fluorescent protein fusions to plus-end tracking proteins (+TIPs) at growing microtubule ends. The authors conclude that accurate particle tracking requires subsecond sampling. Clearly, the shorter the Δt, the lower the ambiguity in particle assignment between frames, and thus the less error-prone the tracking. However, with fluorescence live-cell imaging, this rule has a cautionary footnote: faster sampling accelerates bleaching, which is in addition to the more general concern that cell physiology may be disrupted. Thus, with too fast sampling, it may happen that +TIP markers are lost before a growth period has terminated. The good news is that growth speeds are largely unaffected; however, catastrophe rates may be substantially overestimated. Moreover, the plusTipTracker software offers the feature of interpolating pause and shrinkage events where the marker particle disappears. Premature bleaching leads to a considerable underestimation of the shrinkage speed and rescue frequency.

To alleviate this problem, computer scientists have put much thought into developing accurate tracking solutions despite an increased ambiguity when using lower sampling. plusTipTracker, especially the upgraded version³, is equipped with a global optimization module for particle assignment and Kalman filter approaches that predict the immediate future of a microtubule plus-end trajectory on the basis of its past movement. This permits less frequent sampling; by how much depends on the image quality, microtubule density and microtubule dynamics. Defining the sweet spot between too fast and too slow sampling is a formidable task when starting a particle-tracking project. In Matov et al. we provide some practical guidelines (supplementary information in ref. 2). It should be noted that tracking parameters must be adjusted to the temporal sampling. Inspection of the data presented by Nicovich and Zhou (Fig. 1d in ref. 1) suggests that this may not have been fully accomplished. But this should not distract from the key message of this figure: parameters of microtubule dynamics cannot be interpreted without consideration of the sampling rate.