Fig 1.
Tuj1 staining accurately reflects growth cone morphology.
Co-staining of both collapsed (A) and extended (B) GCs with Phalloidin (red) and Tuj1 (green) shows that over-exposure of the Tuj1 channel accurately reflects the morphology of growth cones, down to the level of individual filopodia (arrowheads), with very a high signal-to-noise ratio. Scale bar in A represents 25μm.
Fig 2.
A) A sample chick LMC explant with an overexposed anti-tubulin antibody stain. B) Basic preprocessing (Despeckling) and thresholding to segment the image into foreground objects (black) and background (white). C) Zoom of 2 GCs used to set the user’s three spatial parameters: Axon width (blue) corresponding to the expected upper bound width of an axon; Axon minimum (green) showing the minimum accepted length for an axon, such that shorter objects are assumed to be filipodia or other structures instead; and Axon Retention (red), the length of axon desired to be retained, measured retrogradely from the GC core branch point (or tip if there are no branches). D) The low spatial frequency image, created by eroding the segmented version by the Axon Width value to leave only objects of greater minimal diameters. In this case, the cell bodies of the explant are found as they are large enough to meet the size criteria (blue) while everything else is removed (red); in a dissociated culture, somas would be found with the presence of a nuclear stain such as DAPI instead. E-J) Conographer’s main steps: spatial frequency filtering to identify axons and cell bodies, axon processing to expose putative GCs, and two verification steps to eliminate false positive axons and GCs. Discarded structures are colored red, those which are passed to the next step blue, and green demarcates sub-steps not shown. E) The high spatial frequency image created by subtracting the low frequency image from the segmented original (blue), consisting primarily of axons, thin GC processes, and noise; most of the latter two are removed with a size filter (red). F) Remaining spindles are cleared from the axons, as are other branch points and small branches (below the Axon Minimum length), to assure that no GCs remain on the image (red), leaving only uninterrupted stretches of putative axon (blue). G) Axons are dilated to connect disparate sections (green), and then eroded to the blue objects. This tip erosion will expose GCs in the next step. H) The explant cell bodies and axons of steps (D) and (G) are subtracted from the segmented original to expose noise particles, previously-unfound axon segments, and GCs. Another simple size filter removes very small structures, as well as those contacting the image edge (red). I) First verification: Putative GCs are checked against the axons (green). Those with more than 1, 1 inappropriately large, or zero axon interfaces are assumed to be parts of axons or, in the latter, noise, and are discarded (red). Those with one appropriately sized interface have that interface expanded until either the GC core or tip is reached, or the previously specified Axon Retention distance is reached. J) Second verification: The area around the cell bodies is scanned to remove too-close GCs (red); this is important as growth-cone-like objects adjacent to the cell bodies will not have been removed based on axon interfaces. K) The macro output: GCs (blue), Axons (green), and Cell Bodies (red).
Fig 3.
A-B) Quantification of the Conographer’s (CG) GC-finding ability (yellow outlines) relative to the 3 human observers (H1, H2, and H3; green outlines) in a 7-image dataset. All observers found roughly the same number of GCs, (210, 268, 258, and 219 respectively). The blue portions of the diagram correspond to its 93 false negatives—ROIs identified by at least two humans and not Conographer–while the red portion corresponds to the 33 false positives found only by Conographer. By our criteria the images contained 235 true GCs, of which Conographer found 177 total, while the humans found an average of 218. C) In cultures of dissociated DRG neurons, Conographer identified growth cones (blue) associated with DAPI-positive nuclei (red) by Tuj1-positive axons (green). Scale bars in A and C represent 50μm and 40μm respectively.
Fig 4.
Ten measurements comprehensively describe GC morphology.
A-C) Each panel contains a primary measurement and a derived ratiometric measurement with its graphical representation. A) The total GC Area and Roundness, which is derived by dividing the GC area (multiplied by 4) by a circle with the radius of the GC’s “major axis”. B) GC Perimeter, and Circularity, which is derived by dividing the GC area (multiplied by 4π) by the square product of the GC’s perimeter. C) Hull Area, which measures the area of a convex hull around the GC to approximate its spread, and Solidity, which measures the fraction of the convex hull covered by the GC. D-F) Non-ratiometric higher order measurements. D) Skeletonization, the total area of the skeletonized GC, as performed by the FIJI ‘Skeletonize’ function, and Branches, a count of the number of the skeleton’s terminal branches (arrowheads), excluding the axon. E) Thickness, which measures the number of binary erosions required to reduce the GC area to 0. F) Process Index, which approximates the number of processes as a count of domains between the GC’s perimeter and that of the convex hull, and Process Roundness, the average Roundness of these domains.
Fig 5.
Comparison and assignation of growth cone collapse state.
A) Distributions of Z-scores of the 10 Conographer-derived parameters in extended (blue) and collapsed (red) GCs. B) Concordance between human observers (black dotted line) and K-means clustering based upon PCA (red), and incorporation of multiple variables ranked by either their individual concordance (grey) or the difference between Z-score means (blue). C) Collapsed (red) and extended (blue) GCs, as determined by a human, plotted upon the first 2 principal components (PCs) derived from Z-scores. D) Collapsed (red) and extended (blue) GCs as determined by K-means clustering, plotted on the same axis as C; white-centred points indicate the GCs furthest from the centre of the cluster to which they do not belong. E) 5 extended and collapsed growth cones, as determined by K-means clustering. F) Growth cones misidentified by K-means clustering as either collapsed or extended.
Table 1.
Changes in parameters due to GC collapse.
Table 2.
Changes in parameters due to Netrin-1 treatment.
Fig 6.
Conographer-derived parameters change during physiological collapse.
A) Distributions of Z-scores of the 10 Conographer-derived parameters in Fc- (dark grey) and Netrin-1-treated (light grey) GCs superimposed over distributions of extended (pale blue) and collapsed (pale red) GCs. B) Plotting of GCs belonging to K-means derived clusters 1 (red) and 2 (blue) upon the first 2 principal components. C) Shapes of GCs belonging to clusters 1 (top) or 2 (bottom); cluster 1 GCs are simple, whereas growth cones belonging to cluster 2 are larger and more complex. D) Percentage of collapsed growth cones in explants from each treatment, as assessed by Conographer-derived measurements, PCA and k-means clustering. GC collapse percentage increased in the presence of Netrin-1 from 60.17±1.21% to 74.61±2.21% (p = 0.0046, Student’s two-tailed t-test). Scale bar in C represents 25μm.