A. Strains used

For imaging CEP neurons in the head region, the BY200 [vtIs1 (dat-1p::GFP, rol-6)] strain was utilized. The 6-OHDA exposures used RB1600 [tub-1(ok192)], and RB2237 [tub-2(ok3024)], which were crossed with BY200. Worms were cultured on K-agar [28] plates seeded with OP50 E. coli strain OP50 at 20°C. Additional culturing details specific to different exposure scenarios are provided below.

B. Rotenone exposure

Worm populations were age-synchronized by extracting worm eggs from gravid adult worms using a 1% bleach and 0.1 M NaOH solution [29]. Worms were then grown from the L1 stage in 1 mL wells filled with complete K medium (2.36 KCl, 3 g NaCl, 5 mg cholesterol, 1 mL 1M CaCl₂, 1 mL 1M MgSO₄ in 1 L of water) [28]. Worms were kept well fed using bacterial strain HB101 at 50 mg/ml. Well plates were gently rocked and incubated at 25°C. Rotenone diluted in DMSO was dosed appropriately at either 0 μM, 0.03 μM, or 0.5 μM and added to the wells at hatching. Rotenone concentrations were re-dosed daily to account for absorption and chemical breakdown. Dilutions were made ensuring final DMSO concentration in the K medium was 0.125% by volume. Worms were removed from the wells once reaching the L4 stage (an extra 24 h was required for worms to reach L4 in 0.5 μM rotenone, as previously established by Mello et al. [30]), then transferred to slides for prompt imaging (see microscopy and imaging).

C. Cold shock exposure

Animals were age-synchronized by extracting worm eggs from gravid adult worms using a 1% bleach and 0.1 M NaOH solution [29]. To improve age-synchronization, extracted eggs were then placed in M9 buffer (3 g KH₂PO₄, 6 g Na₂HPO₄, 5 g NaCl, and 1 mL of 1 M MgSO₄ in 1 L of water) overnight. Newly hatched L1 worms were then transferred to a solid nematode growth media (NGM) [31] plate seeded with Escherichia coli OP50. Worm populations were then cultured at 25°C until they reached day 1 of adulthood. A subpopulation of worms was then imaged (see microscopy and imaging) prior to the 16-hour cold shock. Worms were then transferred to the 4°C refrigerator for 16 hours. Following the cold shock, plates were placed at room temperature for 1 hour prior to imaging.

D. 6-hydroxydopamine exposures

Worms were age-synchronized by transferring adults to K-agar plates seeded with OP50 for 3 hours and allowed to lay eggs. Afterwards, all worms were washed from the plates with K medium. The remaining eggs were allowed to hatch and age to adulthood. Beginning on day 1 of adulthood, they were transferred to fresh plates each day to separate them from progeny until adult day 5. On adult day 5 worms were exposed in K+ medium to either a vehicle control of ascorbic acid, or 6-hydroxydopamine (6-OHDA) in ascorbic acid for one hour. After exposure, animals were washed three times with 2 mL K-medium with 2 minutes for gravity settling between addition of K-medium and aspiration. They were subsequently plated on K-agar plates seeded with OP50 where they remained for 48 hours until imaging.

E. Microscopy and imaging

Animals were placed on a glass slide with 2% agarose pads. To immobilize worms for high-resolution imaging, two methods were used. For the 6-OHDA exposure, worms were picked via stainless steel pick onto 2% agarose pads on glass slides and paralyzed with 500 mM sodium azide. Images were obtained with a Keyence BZ-X710 fluorescence microscope using a 40x objective and GFP excitation/emission filter (OP-87763, Keyence). Z-stacks images were acquired of the head region of each worm using a 0.5 μM step size. Alternatively, for cold shock imaging, worms were placed in a solution of 5 mM tetramisole and M9 buffer prior to transferring to the agarose pad on a glass slide. Worms naturally move in a lateral orientation that causes the four CEP neurons to overlap while imaging. Thus, immobilizing prior to mounting worms increased worm orientation variability and improved the chances that all CEP neurons in the head would be visible in a maximum projection image. The experimental exposure determined how the images were acquired. For the rotenone exposure experiments, images were acquired with an inverted Leica DMi8 widefield fluorescence microscope, a Hamamatsu Orca-Fusion camera using a 63x objective (NA = 1.40), and a metal halide light source. Worms from the cold shock experiments were imaged with an inverted Leica DMi8 widefield fluorescence microscope equipped with a spinning disk confocal head (Crest Optics X-light V2), a Hamamatsu Orca-Fusion camera using a 63x objective (NA = 1.40), and a Laser Diode Illuminator (89 North LDI). Z-stack images were acquired using a 0.75 μm step size and 35 steps.

F. System requirements and performance

AUDDIT v1.0 was scripted in MATLAB 2021b (MathWorks) with the Image Processing and Statistics toolboxes installed. The graphical user interface (GUI) was developed using the built-in App Designer. The GUI works as a standalone desktop application and does not require a MATLAB license. Both the GUI and the MATLAB scripts can be found in the Dryad Repository (https://doi.org/10.5061/dryad.cvdncjt82). For the data in this article, AUDDIT v1.0 was used. We plan to continue to improve performance of AUDDIT and will upload new updates to the Dryad repository as they occur.

The software package was tested on a desktop computer with the Apple Silicon M1 chip and 16 GB memory. On 16-bit images captured with the Hamamatsu Orca-Fusion (2304 x 2304 pixels), AUDDIT could analyze images at ~10 seconds per image. On the smaller, 8-bit images captured on the Keyence BZ-X700 (760 x 920 pixels), it could analyze images at ~2 seconds per image. It is important to note that displaying images while the algorithm is running significantly increases the processing time.

G. Statistical analysis

The statistical analysis throughout this study used the Statistics and Machine Learning Toolbox in MATLAB 2021b. For each exposure, each dendrite analyzed is counted as N = 1. Each exposure and metric were analyzed with a one-way ANOVA. For individual exposure analysis (Figs 3–5), the Bonferroni multiple comparison correction was used [32]. When comparing exposure patterns to the respective controls in Fig 6, the Dunnett’s correction was used [33]. All statistical output with related p values for each experiment are available in the Dryad repository (https://doi.org/10.5061/dryad.cvdncjt82).

A. Unbiased, automatic algorithm for neurodegeneration detection

AUDDIT aims to increase throughput and decrease bias in obtaining quantifiable neurodegeneration metrics. We sought to limit user input to a maximum projection image of the four fluorescently tagged CEP neurons in the C. elegans head and the pixel size of the user’s camera. The pixel size of the user’s camera is used throughout the algorithm to scale functions that are size-dependent, such as creating structuring elements or determining minimum separation between points. Once the pixel size is set and images are loaded, the algorithm completes a series of steps to detect, track, and report neurodegeneration metrics (Fig 1). AUDDIT was developed based on typical morphological features exhibited by CEP neurons exposed to neurodegenerative insults, such as 6-OHDA [9, 16]. We used sample images (S1 Fig) of degenerated CEP dendrites to define algorithm parameters for each of the main tasks performed: dendrite detection, feature detection, and dendrite tracing. Since AUDDIT is built to analyze maximum intensity projections of z-stacks, acquisition of the entire neuron and avoiding under sampling in the z-plane is essential.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Overall neurodegeneration algorithm schematic.

The algorithm requires the user to provide maximum intensity projection image(s] of the CEP neurons and the pixel size in microns. Once the images and pixel size are set, the algorithm automatically detects dendrite locations and crops the image accordingly. Next, the cropped image undergoes a preprocessing step, which permits feature detection and dendrite tracking. An error check follows this step to ensure the dendrites were correctly identified. Using the tracks and feature locations, dendrite loss and feature metrics are quantified and stored for an overall neurodegeneration report. Max projection scale bar = 15 μm.

https://doi.org/10.1371/journal.pone.0281797.g001

i. Dendrite detection.

Because they show degeneration earlier than the cell bodies, dendrites are often the focus when studying neurodegeneration in CEP neurons [15, 20, 34–36]. Thus, the first step of the algorithm automatically crops the images such that only the dendrites are in view. This is feasible because the CEP dendrites, when correctly oriented, present a common morphology of four long, vertical strands. Preprocessing uses an initial global contrast adjustment to ensure the dendrites are bright enough to be detected, as their fluorescence is significantly lower than the cell body of the neuron. Since the cell bodies are often the highest intensity pixels, a binarization step with a threshold level aimed to only detect the brightest pixels is performed (Fig 2A). Once the cell body is detected, the image is rotated such that the dendrites are aligned with the vertical axis. The image is then split horizontally into two smaller images using the bounding box of the cell body as the reference point for cropping (Fig 2A). To determine which of the newly created images contains the dendrites, another binarization step with a less stringent threshold level is completed on both images. The resulting binarized images contain regions of dendrites and any other objects in the frame (Fig 2A). Various properties of the identified objects, such as orientation, major axis length, circularity, and area, are extracted using the built-in regionprops function in MATLAB. Next, to determine which image contains the dendrites, the region properties from the split images are compared. For example, the cell bodies tend to be circular, whereas the dendrites tend to be elongated and oriented along the vertical axis. Thus, the image with fewer circular objects is chosen as the image containing dendrites. Finally, once the correct side of the image is determined, border removal functions are used to exclude the cell bodies and any unwanted objects around the edges of the image. The image is further cropped to remove the extra space vacated by unwanted object removal (Fig 2A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Key steps in neurodegeneration detection algorithm.

A. Dendrite detection steps. The original image is first contrast adjusted and binarized to detect the location of the cell body (Scale bar = 25 μm). Next, the image is split into two images (Scale bar = 25 μm). Using the structures obtained through binarizing these images, the algorithm detects which image contains the dendrites (Scale bar = 10 μm). B. Feature detection steps. First, the cropped image undergoes local contrast adjustment to improve the subsequent global thresholding and binarization step. Feature locations are obtained using horizontal image erosion. Lastly, active contour segmentation of these features improves feature shape definition. C. Dendrite tracking. The local contrast adjusted image is used as a starting point to track dendrites. Each row of pixel intensities is analyzed to obtain 4 local maximum points, which correspond to the four dendrites. The resulting binary image is then closed with a vertical structuring element to connect points where no local maxima is detected. Next, for each row, every foreground point is binned and labeled into its respective dendrite and extra foreground points are removed. Lastly, rows with missing dendrites are interpolated and smoothed to create the final tracks.

https://doi.org/10.1371/journal.pone.0281797.g002

B. Rotenone exposure induces dendritic blebbing, breakage, and neuronal intensity loss

Rotenone is a potent inhibitor of mitochondrial electron transport chain Complex I known to induce oxidative stress [39–41]. Despite these dangers, it continues to be used as a pesticide outside of the United States, and as a piscicide inside of the United States [41]. Over two decades of research has shown that rotenone can consistently induce dopaminergic degradation both in cell cultures and live organisms [42–46]. We aimed to use our algorithmic platform to gain further quantitative insight into neuronal deterioration induced by rotenone exposure in C. elegans. Examples of healthy, moderately degenerated, and severe degeneration induced by control, 0.03 μM and 0.5 μM rotenone exposures can be seen in Fig 3A. The rectangles overlaid on top of the images depict the location of features detected by the algorithm.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Neurodegeneration induced by rotenone exposure in BY200 worms.

A. Example images and detection of (i) control, (ii) 0.03 μm rotenone exposed, and (iii) 0.5 μM rotenone exposed worms (scale bar = 10 μm). Each dendrite is labeled as a star, square, triangle, or circle. N = 1 represents a single dendrite. B-E. Selected neurodegeneration metrics for control (N = 204), 0.03 μM rotenone (N = 168), and 0.5 μM rotenone exposed worms (N = 208). B. Number of detected features (blebs) per dendrite length. (0.03 μM rotenone to 0.05 μM rotenone; p = 0.066). C. Dendrite lengths obtained from tracked dendrites D. Percentage of dendrite remaining and E. Dendrite intensities. The neurodegeneration results for each dendrite in panel (A) are overlaid. (One-way ANOVA. *p < 0.05; **p < 0.01; ***p < 0.001).

https://doi.org/10.1371/journal.pone.0281797.g003

We extracted various metrics of neurodegeneration, such as feature count (blebbing), dendrite length, dendrite remaining, and dendrite intensity. The extracted data from each dendrite labeled with a star, rectangle, triangle, or circle in the example images in Fig 3A is shown in Fig 3B–3D. Worms exposed to 0.5 μM rotenone exhibited significantly higher rates of blebbing per dendrite length than control worms (Fig 3B). Interestingly, AUDDIT revealed that worms exposed to 0.03 μM rotenone have shorter dendrites than controls (Fig 3C). We hypothesize that this is due to rotenone slowing development in worms, and thus making the dendrites shorter [30]. The worms exposed to 0.5 μM rotenone were grown for an extra day to account for slower development (as previously established by Mello et al. [30]), which permitted the dendrites enough time to reach the same length as the controls. Since AUDDIT relies on fluorescent images of CEP dendrites, it is not possible to determine whether length differences are caused by the exposure or if they are simply linked to animal size. We thus include two metrics that normalize for dendrite length (% of dendrite remaining, and feature count per dendrite length). Worms exposed to 0.5 μM rotenone also showed an increase in dendrite loss (Fig 3D). Interestingly, the 0.5 μM dosed worms exhibited significantly dimmer dendrites than the controls (Fig 3E). This may suggest that a decrease in fluorescent intensity driven by a dat-1 promoter [9] can be used as a marker of neurodegeneration. However, decreased dendrite intensity may also reflect decreased GFP protein production due to decreased energetics. The remaining neurodegeneration metrics for rotenone exposed worms can be found in S3 Fig. Ultimately, the neurodegeneration algorithm’s neurodegeneration report is consistent with previous literature showing that rotenone induces neurodegeneration in C. elegans [15].

C. Cold shock induces large, dispersed dendritic blebs

Acute cold shock has been shown to be harmful to worms, with various negative phenotypes being reported [47, 48]. Previous experiments have shown the prevalence of dendritic blebbing in the PVD neurons of worms following cold shock [22]. For this reason, we were interested in utilizing our algorithm to see if cold shock induced blebbing features in the CEP neurons as well. Thus, we exposed worms with fluorescently tagged CEP neurons to a 16-hour cold shock at 4°C. Example images of worms before (pre-cold shock) and after 16 hours of culture at 4°C (cold-shock) or 20°C (control), are shown in Fig 4A along the location of features detected by the algorithm. Interestingly, cold shocking worms for 16 hours significantly increases the number of features (blebs) on the dendrites (Fig 4B). Moreover, while control animals exhibit features mostly far from the cell body, features in cold shocked worms were significantly closer to the cell body (Fig 4C). Cold shocked worms also exhibited larger features (Fig 4D). Analysis of feature shape reveals that 16 hrs. of culture at 20°C (i.e., controls) results in smaller extents (Fig 4E). Extent is another metric to describe a feature’s shape, with bounds of 0 and 1. An extent of 1 refers to a filled rectangle, while an extent of approximately 0 refers to a cross. Thus, the algorithm detected that younger animals have fuller features. In addition, cold shock induced a significant increase in dendritic breakage compared to both pre-cold shock and control worms (Fig 4F). The other metrics extracted by AUDDIT for the cold shock experiments can be found in S4 Fig. Ultimately, the neurodegeneration algorithm revealed that cold shock induces dopaminergic degeneration, mainly by increasing the number and size of bleb-like features in C. elegans.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Neurodegeneration induced by cold shock at 4°C.

N = 1 represents a single dendrite. A. Example confocal images of CEP neurons and detected features for pre-cold shock, control (16 hrs. at 20°C), and cold shocked (16 hrs. at 4°C) worms (Scale bar = 10 μm). B-F. Neurodegeneration metrics for pre-cold shock (N = 72), control (N = 56) and cold-shocked worms (N = 68). B. Feature count per dendrite length. C. Normalized feature location and distance from the cell body. D. Minimum caliper diameters of detected features. E. Feature extent. F. Percentage of dendrite remaining. (One-way ANOVA. *p < 0.05; **p < 0.01; ***p < 0.001).

https://doi.org/10.1371/journal.pone.0281797.g004

D. 6-OHDA-induced dendritic breakage is minimized in tub-1, tub-2 mutants

To test the algorithm’s versatility, we tested dopaminergic neurodegeneration after challenge with 6-OHDA in different genetic backgrounds, and with a different imaging set up. 6-OHDA is known to induce oxidative stress via free radicals and inhibition of mitochondrial electron transport chain Complex I and Complex IV [49]. Thus, we exposed wildtype nematodes (BY200) with no mutations, as well worms with deletions in tub-1 and tub-2, to 10 mM, 25 mM, and 50 mM 6-OHDA to induce dopaminergic degeneration. tub-1 and tub-2 are homologs of human TUBBY proteins that have previously been shown to increase fat storage [50]. However, in our quantification, tub-1 mutants showed decreased fat storage, whereas tub-2 mutants showed increased fat storage (S5 Fig, S1 Text). Images of the worms’ neurons were taken at a lower magnification and with a different camera than the images acquired for the rotenone and cold shock experiments. Remarkably, AUDDIT was able to extract the same quantitative metrics by only changing the input pixel size. Interestingly, the weighted feature count, or the features per dendrite remaining, did not change between most groups. There was a significant increase in blebbing features between wildtype unexposed worms and the 10 mM dosed worms, as well as the tub-2 mutant controls and the 50 mM dosed worms (Fig 5A). Moreover, dendrite loss seems to be a significant marker of neurodegeneration in wildtype worms when exposed to 6-OHDA (Fig 5B), as measured both by % of dendrite remaining and the total number of breaks (S6 Fig) Interestingly, the tub-1 and tub-2 mutations appear to be protective to dendrite loss caused by 6-OHDA exposure, with the tub-2 mutation possibly providing more protection (Fig 5B). This indicates that the quantity of lipids present in the worm does not alter susceptibility to neurodegeneration, but that tubby-like proteins or related signaling pathways may play a role in 6-OHDA induced neurodegeneration. Another interesting finding, which would not be possible to detect through manual neurodegeneration scoring, is that feature pixel intensity decreases with higher 6-OHDA exposures in all worm strains (Fig 5D). Again, the tub-2 mutation appears to be protective against feature intensity loss, as only the highest exposure concentration resulted in significant differences. The remaining neurodegeneration metrics detected by the algorithm can be seen in S6 Fig.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Neurodegeneration quantification of 6-OHDA exposed worms.

N = 1 represents a single dendrite. A. Representative images of CEP dendrites under 6-OHDA exposure. Insets show zoomed-in views of detected features. B-D. Neurodegeneration metrics for control (N = 52, 56, and 76 for wildtype, tub-1, and tub-2, respectively), 10 mM (N = 60, 68, and 56 for wildtype, tub-1, and tub-2, respectively), 25 mM (N = 48, 56, and 64 for wildtype, tub-1, and tub-2, respectively), and 50 mM (N = 32, 76, and 60 for BY200, tub-1, and tub-2, respectively) 6-OHDA exposures. B. Feature count on dendrites weighted by the amount of dendrite remaining. C. Percentage of dendrite remaining. D. Intensity of detected features. E. Comparison of the neurodegeneration algorithm’s detection compared to manual dendrite scoring. (One-way ANOVA. *p < 0.05; **p < 0.01; ***p < 0.001, additional statistical information provided in S2 and S3 Tables).

https://doi.org/10.1371/journal.pone.0281797.g005

To verify the robustness of the neurodegeneration algorithm, we sought to compare results of the code to a typical manual scoring method with images of 6-OHDA exposure [15] (Fig 5D). The images were manually scored as follows: no degeneration (0), blebs on less than half of the dendrite (1), blebs on more than half of the dendrite (2), less than 50% breakage (3), and more than 50% breakage (4). Blebs on half the dendrite means less than half the total length of dendrite contained blebs, such that in looking at the entire dendrite less than 50% of the length was consumed by blebs. Typically, this represents dendrites with less than 5–7 blebs. A score of 2 represents occasions in which several blebs but no breaks are present, such that over half of the dendrite is within blebs. This is usually greater than 5–7 blebs without breaks. A 3 represents (by eye) a break present with more than half of the dendrite remaining. A 4 represents a break present with less than half of the dendrite remaining. Regardless of the number of blebs present, if a break is identified the score is automatically at least a 3. We then binned the results from the code using the following procedure: more than 85% dendrite remaining and no blebs (0), less than 5 blebs (1), more than 5 blebs (2), between 50% and 85% dendrite remaining (3), and less than 50% dendrite remaining (4). The neurodegeneration algorithm is consistent with current manual methods for scoring dopaminergic neurodegeneration, which shows its readiness to be used in future dopaminergic neurodegeneration studies.

E. AUDDIT enables comparative degenerative analysis between exposures

After individually quantifying degeneration of rotenone, cold shock, and 6-OHDA separately, we sought to investigate whether different exposures result in distinct degeneration patterns. We quantified the z-score for neurodegeneration metrics of each treatment compared to their control. These z-scores along with its associated p-value are shown as heatmaps in Fig 6. It becomes clear each stressor causes similar directional effects for most metrics besides intensity and length, yet significance values differ between exposures. Interestingly, cold shock appears to affect feature morphology more than rotenone and 6-OHDA, as a cold shock exposure makes features larger and more circular when compared to their control. Alternatively, this difference could be a result of the improved resolution and reduced noise from confocal images as compared to those obtained with 63x epifluorescence (rotenone experiments) and 40x epifluorescence (6-OHDA experiments).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Experimental comparison across different exposure groups compared to their respective control.

A. Heat map of z scores for each exposure and its metrics obtained from AUDDIT to show directionality. B. Metrics that were significantly affected for each exposure with p < 0.05. (One-way ANOVA, Dunnett’s multiple comparisons).

https://doi.org/10.1371/journal.pone.0281797.g006

Most exposures significantly decreased the amount of dendrite remaining, which indicates that dendrite loss is a key metric in dopaminergic neuron degeneration. Dendrite and feature intensities, which would not be measurable through manual scoring, may also be important when attempting to differentiate degeneration. Interestingly, both rotenone and 6-OHDA significantly decrease both intensity measurements, while the cold shock exposure has no effect. This indicates that the chemical exposures to rotenone and 6-OHDA had a greater effect on the overall fluorescence of the neuron rather than feature morphology.

Moreover, both 0.5 μM rotenone and 10 mM 6-OHDA have similar degenerative effects in both significance and directionality. It is important to note that since each experiment was performed on worms of varying ages and culture conditions, definitive conclusions must be taken with caution. Thus, a more in-depth comparison of degeneration patterns resulting from exposures should be completed by standardizing worm age, culture conditions, and imaging setups. More broadly, however, these results indicate that the metrics obtained from AUDDIT have the potential to elucidate differences in degeneration patterns between different exposures, particularly if other parameters (e.g., microscopy) are held constant.