Standardization of a Novel Semi-Automatic Software for Neurite Outgrowth Measurement

The neurite outgrowth measurement algorithm is robustly capable of detecting neurites in both neural networks and single neurons. It generates a yellow mask that segments objects with high contrast, such as cell bodies, cellular debris, dead cells, tissue explants, and shadows. Additionally, a magenta mask appears on neurites of various thicknesses. Neurite length values are provided in mm/mm², indicating that the axonal length has been divided by the area of the image, which is 0.282739 mm² and constant for every scanning condition. Therefore, in order to obtain pure values of neurite length in mm, the numbers provided by the software need to be multiplied by the area of the image.

Semi-automatic versus manual method
The software used is a semi-automatic methodology to measure the total axonal length. To assess the accuracy of the software, we conducted measurements on the same neurons using the manual method with the NeuronJ plug-in as well. As depicted in Figure 1, the segmentation mask on the neurons is highly similar between the two methods (Figure 1A).

Additionally, we conducted statistical analysis on the values obtained to examine their correlation. The Spearman correlation analysis yielded a high coefficient r of 0.8526, hence providing strong evidence of the accuracy and precision of the algorithm (Figure 1B). Automatic measurement requires high standards of culture quality based on its cleanliness, density, and purity. The results obtained with semi-automatic segmentation are reproducible and not affected by individual judgment. Unbiased reproducibility is an issue for manual methodologies.

Sometimes, semi-automatic segmentation errors can occur because of different causes. In phase contrast images, dirt in the culture could be detected as neurites by the semi-automatic segmentation. Moreover, the presence of different cell types can disturb the segmentation process. Such issues do not arise with manual segmentation because it is performed by human eyes. Nonetheless, if such issues arise, they can be overcome by using immunocytochemistry images as a control.

Segmentation of neurites
For adult DRG neuron primary cultures, the optimal starting point for a reliable phase analysis is to have neurons uniformly plated in the well and clean culture. If errors occur during seeding and cells concentrate in one spot, as illustrated in Figure 2A-B, the values will be more of an estimation than a close reflection of reality. In such situations, the yellow mask will cover most of the neurites in between cells (Figure 2A-B), thereby resulting in the loss of neurite length. Moreover, the software will be significantly biased in the recognition of neurites, and it is very likely that a magenta mask will appear on objects that are not neurites (Figure 2D). In an optimal image, there should be up to 15 neurons at 20x magnification.

When neurons are correctly plated and the culture is clean, as illustrated in Figure 3A-B, it is advisable to adjust the segmentation slider towards the background (0.5 – 0.7; Figure 3C). This helps to reduce the yellow component that will appear on high-contrast objects in the image, such as branching points that should be in magenta. Moreover, if neurites are bold, a neurite sensitivity between 0.4 and 0.5 should be sufficient to cover most of them (Figure 3C-D).

Another common situation that can arise is a dirty culture with many cell debris and dead cells, as shown in Figure 4A-B. In such conditions, there are many high-contrast objects. Therefore, it is advisable to increase the size of the yellow mask by adjusting the segmentation slider towards the cell or by increasing the adjust size parameter (+1, +2, and so on; Figure 4C). Nonetheless, it would also be useful to decrease the neurite sensitivity slightly to prevent the software from incorrectly identifying as neurites objects that are not neurites as such. (Figure 4C-D).

At times, neurites can appear very thin and pale, as shown in Figure 5A-B, posing challenges for the software to accurately segment them (Figure 5D). In this case, it is advisable to increase the neurite sensitivity to at least 0.6 (Figure 5C). However, bear in mind that the higher the sensitivity, the greater the probability the software will incorrectly mark objects that are not neurites as such (Figure 5D). Some precautions can be taken to prevent the sensitivity bias from increasing too much, for example, by adjusting the segmentation slider towards cells. However, if neurites are too thin to be detected by the software, the neurite length values will be biased regardless.

In the case of immunocytochemistry images, the main issue lies in the background. Apart from the seeding conditions for which the aforementioned rules apply, the primary source of bias is the fluorescence itself. The software effectively recognizes very bright neurites while thinner, less intense neurites are left behind (Figure 6A-B). To prevent loss of neurite length, the neurite fine sensitivity can be increased up to 0.75 (Figure 6C). However, it is strongly advisable to reduce the neurite coarse sensitivity to at least 8-9 to prevent excessive detection bias by considering neurites in the background (Figure 6C-D). If the latter is not reduced, then all the background will be segmented, as depicted in Figure 7.

A common problem with fluorescence acquisition is the scattering of the light. Often, immunocytochemistry images present flashlights in the image, which significantly affect the quality and precision of the analysis (Figure 8A-B). In such a situation, not much can be done to improve the analysis, and values will be more of an estimation. Light scattering interferes with neurite recognition so that only very bright neurites will be detected (Figure 8C-D). Another issue in immunocytochemistry images is the quality of the staining per se. Frequently, due to human errors, axons can be broken (Figure 9A), and cell bodies can be torn away during washes (Figure 9B). These mistakes pose a critical problem as neurite length values lose precision and accuracy. Consequently, the interpretation of biological data is altered, leading to wrong conclusions.

For embryonic cultures, the situation differs. In this type of culture, the presence of glial cells is predominant. Consequently, errors significantly increase as the software also detects the linings of glial cells (Figure 10A-B). To minimize this issue, the segmentation slider should be moved towards cells, typically around values of 1.7-2 (Figure 10C-D). This approach ensures that most of the glial cells are covered by the yellow mask and thus not considered in the neurite length measurement. Another useful tip is to keep the neurite width at 2, as embryonic neurons in culture typically exhibit bipolar or unipolar shapes with thick neurites (Figure 10C-D). This precaution filters out most of the linings of glial cells that usually are very thin. Lastly, be careful not to increase the neurite sensitivity too much; otherwise, what has been filtered out by the neurite width parameter will be included again in the neurite length measurement.

In the case of embryonic cultures, where the glial cell component is predominant, immunocytochemistry might be the better option. By specifically staining neurons, the issue of segmentation of glia is resolved since glia will not be stained, making analysis much easier and more accurate (Figure 11).

Lastly, the software can also be exploited to evaluate the differentiation program of cell lines and/or iPSCs to assess their growth state. Thus, the application of similar parameters and precautions can be used for different goals.

Concerning the differentiation process, the software's robustness has been proved in the NSC-34 cell line (hybrid cell line constituted by mouse neuroblastoma cells and motoneurons derived from the spinal cord of mouse embryos) during the maturation into motoneurons-like cells. As for DRG primary cultures, the optimal starting point for good analysis is uniform cell seeding. The undifferentiated and differentiated cells, upon retinoic acid treatment, can be followed using acquisitions during the entire culture period or as shown in Figure 12, at the last time point.

Indeed, in addition to neurite length, the algorithm also provides the branch point parameter. However, it is important to note that the branch point parameter does not represent the exact number of branch points; rather, it indicates the density of branching in the image as it is expressed in mm/mm². This measurement is significantly influenced by debris in the culture and seeding concentration. Therefore, the density of neurons in the image and the cleanliness of the culture are crucial factors for obtaining reliable values. If the culture presents many cell debris, not filtered away by the yellow mask, they will be included in the neurite length as well as in the branch point measurement.

Consequently, it is recommended to normalize these values for the cell count, as the number of neurons in the image influences neurite length and branch point measurements.

Segmentation of cell bodies
Among all the parameters provided by the system, there are cell-body cluster and cell-body cluster area. However, these two parameters are not reliable to use as values for cell counting. As illustrated in Figure 4, the software segments high-contrast objects in the yellow mask, including shadows caused by medium movement in the well. Additionally, it also segments dead cells and cellular debris (Figure 4). To obtain a reliable cell count of the growing neurons, a manual method, such as the Cell Counter tool in Fiji (Figure 13), can be utilized.

A summary of the suggested analysis parameters sorted by type of culture is provided in Table 1 for phase images and Table 2 for immunocytochemistry images. Moreover, a summary of the suggested analysis parameters to solve specific issues is provided in Table 3.

Figure 1: Correlation analysis between manual and semi-automatic neurite segmentation. (A) On the left, a representative phase image of a neuron segmented with the NeuronJ plug-in in Fiji. On the right, a representative phase image of a neuron is segmented with the semi-automatic software. (B) A simple linear regression run on 20 neurons was analyzed with both manual and semi-automatic methods. Spearman correlation coefficient r = 0.8526, p**** < 0.0001. Images were acquired 48 h after seeding. Magnification 20x. Scale bar, 50 µm. The figure was created with BioRender. Please click here to view a larger version of this figure.

Figure 2: Seeding error in adult sensory neuron culture. (A) Representative phase image of the seeding error. (B) Representative automatically segmented image. The yellow mask segments cell bodies, the magenta mask segments neurites. (C) Illustrative neurite outgrowth analysis definition parameters. (D) Top panel: zoom on the error of cell body segmentation (yellow mask) due to clustering of cell bodies. Bottom panel: zoom on the error of neurites segmentation (magenta mask) due to cellular debris. Images were acquired 48 h after seeding. Magnification 20x. Scale bar, 50 µm. The figure was created with BioRender. Please click here to view a larger version of this figure.

Figure 3: Ideal adult sensory neuron culture for neurite outgrowth analysis. (A) Representative phase image of an ideal seeding condition for neurite outgrowth analysis. (B) Representative automatically segmented image. The yellow mask segments cell bodies, the magenta mask segments neurites. (C) Illustrative neurite outgrowth analysis definition parameters. (D) Zoom on cell body segmentation (yellow mask) and on neurites segmentation (magenta mask). Images were acquired 48 h after seeding. Magnification 20x. Scale bar, 50 µm. The figure was created with BioRender. Please click here to view a larger version of this figure.

Figure 4: Disturbing elements in neurite outgrowth analysis in adult sensory neurons culture. (A) Representative phase image of cellular debris and shadows due to medium movements. (B) Representative automatically segmented image. The yellow mask segments cell bodies, the magenta mask segments neurites. (C) Illustrative neurite outgrowth analysis definition parameters. (D) Zoom on the error of cell body segmentation (yellow mask) and neurites segmentation (magenta mask) due to cell debris and shadows of the moving medium. Images were acquired 48 h after seeding. Magnification 20x. Scale bar, 50 µm. The figure was created with BioRender. Please click here to view a larger version of this figure.

Figure 5: Thin neurites in neurite outgrowth analysis in adult sensory neurons culture. (A) Representative phase image of neurons characterized by very thin neurites. (B) Representative automatically segmented image. The yellow mask segments cell bodies, the magenta mask segments neurites. (C) Illustrative neurite outgrowth analysis definition parameters. (D) Top panel: zoom on the loss of neurite length due to detection limits of the neurites segmentation (magenta mask). Bottom panel: zoom on the error of neurite segmentation (magenta mask) on foreign objects. Images were acquired 48 h after seeding. Magnification 20x. Scale bar, 50 µm. The figure was created with BioRender. Please click here to view a larger version of this figure.

Figure 6: Neurites' brightness in neurite outgrowth analysis in adult sensory neurons culture. (A) Representative immunocytochemistry image (neuronal marker Tuj1) of neurons characterized by very bright and thick neurites. (B) Representative automatically segmented image. The purple mask segments cell bodies, the blue mask segments neurites. (C) Illustrative neurite outgrowth analysis definition parameters. (D) Top panel: zoom on the cell body segmentation error (purple mask) due to the thickness of the neurites. Bottom panel: zoom on the loss of neurite length due to intense fluorescence brightness of thick neurites. Images were acquired 48 h after seeding. Magnification 20x. Scale bar, 50 µm. The figure was created with BioRender. Please click here to view a larger version of this figure.

Figure 7: Background fluorescence in neurite outgrowth analysis in adult sensory neurons culture. (A) Representative immunocytochemistry image (neuronal marker Tuj1) of high background fluorescence noise. (B) Representative automatically segmented image. The purple mask segments cell bodies, the blue mask segments neurites. (C) Illustrative neurite outgrowth analysis definition parameters. (D) Zoom on the neurite segmentation error (blue mask) due to the interference of the background fluorescence. Images were acquired 48 h after seeding. Magnification 20x. Scale bar, 50 µm. The figure was created with BioRender. Please click here to view a larger version of this figure.

Figure 8: Light scattering in neurite outgrowth analysis of adult sensory neurons culture. (A) Representative immunocytochemistry image (neuronal marker Tuj1) of the light scattering. (B) Representative automatically segmented image. The purple mask segments cell bodies, the blue mask segments neurites. (C) Illustrative neurite outgrowth analysis definition parameters. (D) Top panel: zoom on the loss of neurite length due to the interference of the light scattering. Bottom panel: zoom on the cell body segmentation error (purple mask). Images were acquired 48 h after seeding. Magnification 20x. Scale bar, 50 µm. The figure was created with BioRender. Please click here to view a larger version of this figure.

Figure 9: Quality-affecting errors of the staining procedure interfering with neurite outgrowth analysis of adult sensory neurons culture. (A) On the left, a representative phase image of adult sensory neurons. On the right, a representative immunocytochemistry image (neuronal marker Tuj1) of broken neurites due to washing in the staining procedure. (B) On the left, a representative phase image of adult sensory neurons. On the right, a representative immunocytochemistry image (neuronal marker Tuj1) of cell body removal due to washes of the staining procedure. Images were acquired 48 h after seeding. Magnification 20x. Scale bar, 50 µm. The figure was created with BioRender. Please click here to view a larger version of this figure.

Figure 10: Ideal embryonic (E13.5) sensory neuron culture for neurite outgrowth analysis. (A) Representative phase image of an ideal seeding condition for neurite outgrowth analysis in embryonic sensory neuron culture. (B) Representative automatically segmented image. The yellow mask segments cell bodies, the magenta mask segments neurites. (C) Illustrative neurite outgrowth analysis definition parameters. (D) Zoom on cell body segmentation (yellow mask) and on neurites segmentation (magenta mask). Images were acquired 24 h after seeding. Magnification 20x. Scale bar, 50 µm. The figure was created with BioRender. Please click here to view a larger version of this figure.

Figure 11: Ideal embryonic (E13.5) sensory neuron culture for neurite outgrowth analysis. (A) Representative immunocytochemistry image (neuronal marker Tuj1) of an ideal embryonic sensory neurons culture for neurite outgrowth analysis. (B) Representative automatically segmented image. The purple mask segments cell bodies, the blue mask segments neurites. (C) Illustrative neurite outgrowth analysis definition parameters. (D) Zoom on cell body segmentation (purple mask) and on neurites segmentation (blue mask). Images were acquired 24 h after seeding. Magnification 20x. Scale bar, 50 µm. The figure was created with BioRender. Please click here to view a larger version of this figure.

Figure 12: Differentiated NSC-34 cells have increased neurites and branching compared to undifferentiated controls. Representative images of NSC-34 cells. Top panel: undifferentiated NSC-34 cells (control). Bottom panel: differentiated NSC-34 cells after 96 hours of retinoic acid treatment. On the left side of both panels, phase contrast images, while on the right side of both panels, automatically segmented images (soma in yellow, neurites in magenta). Images were acquired 96 h after seeding. Magnification 10x. Scale bar, 50 µm. The figure was created with BioRender. Please click here to view a larger version of this figure.

Figure 13: Cell Counter tool on Fiji. Manual cell count performed on a phase image of an ideal seeding condition for neurite outgrowth analysis. Magnification 20x. The figure was created with BioRender. Please click here to view a larger version of this figure.

Table 1: Summary of suggested analysis definition parameters for adult sensory neuron cultures, embryonic (E13.5) sensory neuron cultures, and NSC-34 cultures in phase contrast images.

Table 2: Summary of suggested analysis definition parameters for adult sensory neuron cultures, embryonic (E13.5) sensory neuron cultures, and NSC-34 cultures in immunocytochemistry images.

Table 3: Summary of the suggested analysis definition parameters to solve specific issues in different types of cultures.