Fig 1.
Actin within the cell consists of both single monomers (G-actin) and branched filaments (F-actin). Filaments have an orientation, with a pointed end (-) and a barbed end (+). Polymerisation and disassembly can occur at both ends, but polymerisation typically occurs at a higher rate at the barbed end, while disassembly is more common at the pointed end. A range of proteins and signalling pathways contribute to the turnover of actin monomers, which drives the structure of the actin cytoskeleton and enables a high level of organisation and adaptation. The Arp2/3 complex is involved in filament nucleation and branching, while capping proteins bind to the barbed end to prevent both polymerisation and depolymerisation. Severing proteins such as ADF/cofilin can sever the filament creating an additional barbed and pointed end.
Fig 2.
Two parallel pathways are taken from the initial image (A) after background removal and filtering. The image had vessels enhanced (B) before thresholding into a binary image (C) before being skeletonised and rotated (D). Alongside this, the initial image was separately rotated before vessel enhancement (Brt) and thresholding (Crt) followed by skeletonisation (D). The skeleton results were combined to ensure no loss of information occurred in the rotation steps. From the combined skeleton, the labelling and relabelling processes were applied (E) to identify each filament and branch for further analysis.
Fig 3.
A visualisation of several of the measurements we define to quantify the actin network. The ROI size is measured as the area encompassed by the black line of the outer bounding box and is used in the calculation of densities such as the skeleton and branch point density. The cell orientation is measured as the angle of the bounding box relative to the horizontal axis. The curvature, both the signed filament curvature and the unsigned filament curvature, are averaged over the whole length of each filament and measured via a series of three points separated by a characteristic length scale, λ. The deviation measures the average distance between a filament and a straight line connecting the end points. The filament width is defined for each filament as an average over the filament length. The filament angle, θf, is measured relative to the major cell axis and is calculated both as a 2D and 3D (estimated) version. The branch angle, θb, is measured at every branch point and measures the deviation of the branch from the main filament; again, there is a 2D and 3D (estimated) version. Finally, the branch ratio is the number of branch points divided by the number of labelled filaments.
Table 1.
Quantitative network measures.
The seventeen quantitative measures that we calculate for each extracted network along with a description of each. Full details are given in S1 Text. N = single number, L = list of numbers.
Fig 4.
Testing network extraction performance.
Artificially-generated data was analysed with the DRAGoN algorithm and the results compared with the known ground truth. (A) To measure the effect of different signal-to-noise ratios, images were generated with three different levels of signal and noise intensities (100 each), showing that a critical SNR value of at least 4 is required to extract the most from the data in this assay. (B) To measure the effect of network density, images were generated with 5, 10 and 15 filaments (100 images for each), showing increased network density reduces sensitivity. (C) To explore the effect of the threshold parameter, four different threshold values were tested (100 images for each), highlighting the trade off between sensitivity and precision and that the optimal range is 87–92. (D) Finally, to investigate the effect of the numerical aperture (NA) of the lens, images were generated with a range of point-spread functions (100 for each), showing that even with high quality data, a minimum NA of 0.7–0.9 is required to achieve the resolution required.
Fig 5.
Arabidopsis hypocotyl cells labelled with GFP-Lifeact.
Three genotypes were tested with the DRAGoN algorithm: the wildtype (Col-0) and the loss-of-function mutants arp2 and formin4/7/8. Scale bar is 5μm.
Fig 6.
Comparison of the actin network in hypocotyls of the wild type, formin mutant and Arp2 mutant.
Four measures (structure density, skeleton density, average signed curvature and mean actin width) showed significant changes between the wild type and formin4/7/8 triple mutant. One of these (average signed curvature) also showed a significant difference between the wild type and arp2–1 mutant and another (skeleton density) displayed a significant difference between the arp2–1 and formin4/7/8 mutants. Pairwise testing was performed with Tukey HSD after a MANOVA test. Single/double/triple asterisks represents p < 0.05, 0.01, 0.001 respectively.
Fig 7.
Effect of cell tissue on actin cytoskeleton.
Four measures showed significant differences between wild-type hypocotyl and leaf cells. Significant differences are only shown between wild-type cells, although similar significance values were found between most genotype pairs. Pairwise testing performed with a Tukey HSD test after a MANOVA test. A double asterisk represents p < 0.01 and a triple asterisk denotes p < 0.001. Error bars show one standard deviation from the mean.
Fig 8.
Principal component analysis (PCA) of Blumeria graminis-infected and uninfected leaves.
Shown is the percentage of the variance that is explained by each of the principal components. The dotted line shows the mean percentage of variance over all principal components. The first four components contribute more than or equal to this mean and are therefore kept for future analysis.