Automated Tracking of Biopolymer Growth and Network Deformation with TSOAX

Ting Xu, Christos Langouras, Maral Adeli Koudehi, Bart E. Vos, Ning Wang, Gijsje H. Koenderink, Xiaolei Huang, Dimitrios Vavylonis

TL;DR

This paper introduces TSOAX, an open-source software tool for automatically tracking biopolymer networks in 2D and 3D microscopy time-lapse sequences. By combining global and local correspondence algorithms with active contours, TSOAX enables quantitative studies of dynamic biopolymer behavior, such as filament growth, network deformation, and cellular processes like ring constriction, which were previously challenging to analyze.

What Problem Does This Paper Address?

Understanding the dynamic behavior of biopolymer networks is crucial for comprehending their function in biological systems, from tissue mechanics to cellular processes. These networks, composed of semi-flexible polymers like actin filaments, are constantly changing – growing, shrinking, deforming, and forming complex structures. Analyzing these dynamics requires tracking individual components and their collective behavior over time in microscopy images. However, existing methods struggle with several challenges: low signal-to-noise ratios in images, the sheer volume of data generated by time-lapse sequences, and the complex topological changes that occur within these networks (e.g., filaments intersecting, forming loops, or deforming under stress). Previous approaches often focused on tracking individual filaments or static networks, leaving a gap in robust, automated tracking of dynamic, interconnected biopolymer networks.

Methods

TSOAX builds upon the SOAX software, which uses "snakes" (stretching open active contours) to extract 2D and 3D biopolymer networks from static images. The key innovation in TSOAX is its extension to time-lapse sequences, incorporating two main algorithmic advancements:

  1. Global k-partite matching for curved filaments: This extends previous graph-based matching techniques to handle the curved nature of biopolymer filaments, allowing for temporal correspondence between detected curves across multiple frames. It aims to minimize the total dissimilarity between curves linked in a track.
  2. Local matching for temporal consistency: To address the issue of inconsistent network topology detection between frames (where the same physical structure might be detected differently), TSOAX introduces a local matching step. This step enforces temporal consistency by linking dissected curve segments at junctions based on both their local geometry and their relationship to structures in the previous frame. This helps maintain a coherent network structure over time.

The overall workflow involves a detection phase (centerline extraction, dissection at T-junctions) and a matching phase (local and global matching) to generate "curve tracks" for each filament or segment.

Datasets

The paper demonstrates TSOAX's capabilities using a variety of datasets:

  • Simulated 2D actin filament images: Generated to mimic elongating and intersecting actin filaments, allowing for controlled evaluation of tracking performance against known ground truth.
  • Experimental 2D actin filament images: Data from Fujiwara et al. (2007) was used to validate TSOAX against previous manual and semi-automated tracking results.
  • Live cell microscopy images: Used to track ring formation and constriction in mutant fission yeast cells.
  • Confocal rheology image sequences: 3D fibrin networks undergoing controlled shear deformation were analyzed to measure nematic ordering.
  • Simulated 3D polymer networks: Used to test tracking of networks undergoing shear and rotation.

The paper does not introduce a new benchmark dataset but rather applies TSOAX to existing and simulated data to showcase its versatility.

Key Insights and Ideas

  • Bridging static and dynamic analysis: TSOAX effectively bridges the gap between static network extraction (like SOAX) and the need for dynamic tracking, enabling quantitative analysis of processes that were previously difficult to study automatically.
  • The importance of local matching: The introduction of a local matching step is crucial for maintaining topological consistency across frames, significantly improving tracking accuracy and robustness, especially when detection algorithms produce slightly different results between frames.
  • Versatility across biological systems: The software demonstrates applicability to diverse biopolymer systems, including actin filaments, fibrin networks, and cellular contractile rings, highlighting its broad utility in cell biology and biophysics.
  • Quantitative insights from complex dynamics: TSOAX enables the quantification of complex phenomena like filament elongation rates, ring constriction speeds, and the nematic ordering of strained networks, providing valuable data for understanding mechanical properties and cellular functions.

Key Results

  • Accurate tracking of elongating filaments: TSOAX successfully tracked many elongating actin filaments in simulated and experimental 2D images, with measured elongation rates closely matching previous studies (e.g., 11.0 subunits/sec vs. 11.2-11.3 subunits/sec).
  • High precision and recall in 2D tracking: For experimental 2D images, TSOAX achieved a precision of 97.9% and a recall of 93.1% when local matching was enabled, significantly outperforming results without local matching (80.3% precision, 73.4% recall).
  • Quantification of ring constriction: The software accurately tracked the constriction of contractile rings in fission yeast cells, measuring a nearly constant constriction rate of 0.24 µm/min.
  • Measurement of nematic ordering in 3D networks: TSOAX was used to quantify the alignment of fibrin fibers in 3D networks under shear deformation, showing a monotonic increase in the nematic order parameter with increasing strain.
  • Accurate recovery of affine transformations: In simulated 3D networks undergoing shear, TSOAX accurately recovered the imposed affine transformation matrix, with deviations primarily attributed to detection and tracking errors.
  • Reduced tracking errors with grouping: The "grouping" option, which helps link segments at junctions, was shown to reduce tracking errors, particularly in complex scenarios.

Limitations and Future Work

  • Signal-to-noise ratio dependency: The accuracy of filament detection, and thus tracking, is limited by the signal-to-noise ratio of the input images, similar to the underlying SOAX algorithm. An order of 5 for the signal-to-noise ratio is suggested for reliable detection.
  • Tracking errors: While TSOAX shows high accuracy, tracking errors can occur, especially with sudden changes in topology or when filaments stop elongating at junction points. These errors manifest as smaller peaks in length change histograms.
  • 3D tracking complexity: Tracking accuracy in 3D images is generally lower than in 2D, with a higher rate of false negatives, due to increased complexity.
  • Future work: The authors suggest that TSOAX could be applicable to a broader range of systems, including mitochondrial networks, amyloid and cellulose fibers, and carbon nanotube networks. They also envision its use for tracking more complex events like filament merging, splitting, and overlapping. The paper also notes that for speed optimization, parameters like the number of snake points and maximum iterations can be reduced.

Summary updated

Abstract

Studies of how individual semi-flexible biopolymers and their network assemblies change over time reveal dynamical and mechanical properties important to the understanding of their function in tissues and living cells. Automatic tracking of biopolymer networks from fluorescence microscopy time-lapse sequences facilitates such quantitative studies. We present an open source software tool that combines a global and local correspondence algorithm to track biopolymer networks in 2D and 3D, using stretching open active contours. We demonstrate its application in fully automated tracking of elongating and intersecting actin filaments, detection of loop formation and constriction of tilted contractile rings in live cells, and tracking of network deformation under shear deformation.
BibTeX
@inproceedings{xu2019automated,
  author = {Ting Xu and Christos Langouras and Maral Adeli Koudehi and Bart E. Vos and Ning Wang and Gijsje H. Koenderink and Xiaolei Huang and Dimitrios Vavylonis},
  title = {{Automated Tracking of Biopolymer Growth and Network Deformation with TSOAX}},
  booktitle = {Scientific},
  year = {2019},
  doi = {10.1038/s41598-018-37182-6},
}

Loading related content...

Figure