Available via license: CC BY 4.0
Content may be subject to copyright.
Systems biology
NMJ Analyser: a novel method to quantify neuromuscular
junction morphology in zebrafish
Jaskaran Singh
1
, Yingzhou Edward Pan
1
, Shunmoogum A. Patten
1,2,
*
1
Institut National de la Recherche Scientifique (INRS), Centre Armand Frappier Sante´ Biotechnologie, Laval, QC H7V 1B7, Canada
2
De´ partement de Neurosciences, Universite´ de Montre´ al, Montreal, QC, Canada
*Corresponding author. INRS—Centre Armand Frappier Sante´ Biotechnologie, 531 Boul. des Prairies, Laval, QC H7V 1B7, Canada. E-mail:
kessen.patten@inrs.ca (S.A.P.)
Associate Editor: Pier Luigi Martelli
Abstract
Motivation: Neuromuscular junction (NMJ) structural integrity is crucial for transducing motor neuron signals that initiate skeletal muscle con-
traction. Zebrafish has emerged as a simple and efficient model to study NMJ structural morphology and function in the context of developmen-
tal neurobiology and neuromuscular diseases. However, methods to quantify NMJ morphology from voluminous data of NMJ confocal images
accurately, rapidly, and reproducibly are lacking.
Results: We developed an ImageJ macro called “NMJ Analyser” to automatically and unbiasedly analyse NMJ morphology in zebrafish. From
the Z-stack of a zebrafish hemisomite, both presynaptic and postsynaptic fluorescently labeled termini at NMJs are extracted from background
signal, with larger clusters of termini being segmented into individual termini using an unbiased algorithm. The program then determines whether
each presynaptic terminus is co-localized with a postsynaptic terminus and vice versa, or whether it is orphaned, and tabulates the number of or-
phan and co-localized pre- and postsynaptic termini. The usefulness of this ImageJ macro plugin will be helpful to quantify NMJ parameters in
zebrafish, particularly during development and in disease models of neuromuscular diseases. It can enable high-throughput NMJ phenotypic
screens in the drug discovery process for neuromuscular diseases. It could also be further applied to the investigation of NMJ of other develop-
mental systems.
Availability and implementation: NMJ Analyser is available for download at https://github.com/PattenLab/NMJ-Analyser.git.
1 Introduction
Functional and anatomical studies of the neuromuscular sys-
tem among the vertebrates have formed the basis of our un-
derstanding of motor neuronal control of muscle contraction
and locomotion (D’Elia and Dasen 2018,Luxey et al. 2020,
Xu et al. 2022). The zebrafish model has proven to be a pow-
erful tool for studying the vertebrate neuromuscular system. It
is a widely used model organism in scientific research due to
its advantageous features, such as high transparency and ex-
ternal development of embryos. These features make zebra-
fish an ideal candidate for microscopic imaging techniques,
such as confocal, two-photon, and super-resolution micros-
copy. These non-invasive imaging techniques in the zebrafish
provide researchers with the ability to visualize the internal
anatomy, development, and behavior of the organism at the
cellular and molecular levels. Furthermore, the genetic simi-
larity between the zebrafish and humans and the ease of ge-
netic manipulation of the zebrafish genome have made it a
powerful tool for understanding human biology and disease
(Howe et al. 2013). In addition, the ability to perform large-
scale genetic screens in the zebrafish makes it an effective tool
for identifying novel disease genes and studying their func-
tion. The ability to perform high-resolution imaging in a liv-
ing organism has revolutionized the field of biological
research, and the zebrafish is playing a critical role in advanc-
ing our understanding of the underlying mechanisms of devel-
opment and disease. Several studies have capitalized on the
advantages of the zebrafish and provided important insights
in motor axonal growth (Myers et al. 1986,Chen et al. 2012,
Sainath and Granato 2013), neuromuscular junction (NMJ)
development (Jing et al. 2010,Bailey et al. 2019), synaptogen-
esis (Jontes et al. 2000,Hutson and Chien 2002,Panzer et al.
2005,Son et al. 2020), muscle development (Skobo et al.
2014,Dubi
nska-Magiera 2016,Chen et al. 2017,Yin et al.
2022), and in NMJ deficits underlying neuromuscular pathol-
ogies (Patten et al. 2017,Butti et al. 2021,Lescouze` res et al.
2023). In the last few decades, the zebrafish has particularly
emerged as a valuable tool in studying the pathogenesis and
development of therapeutic interventions for various neuro-
muscular disorders (Singh and Patten 2022). Disturbances in
NMJ morphology is one of the characteristic hallmarks of
most neuromuscular diseases (NMDs) (D’Ydewalle et al.
2011,Mart
ınez-Herna´ndez et al. 2013,Butti et al. 2021).
Various fluorescent reporter and transgenic zebrafish lines
have been developed to model NMDs, which are used for
high-throughput drug screening assays (Giacomotto et al.
2015,Patten et al. 2017,Singh and Patten 2022,Lescouze` res
et al. 2023). High-throughput studies involve evaluating the
Received: 27 September 2023; Revised: 9 November 2023; Editorial Decision: 13 November 2023; Accepted: 5 December 2023
V
CThe Author(s) 2023. Published by Oxford University Press.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which
permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
Bioinformatics,2023, 39(12), btad720
https://doi.org/10.1093/bioinformatics/btad720
Advance Access Publication Date: 7 December 2023
Original Paper
Downloaded from https://academic.oup.com/bioinformatics/article/39/12/btad720/7460225 by guest on 13 December 2023
impact of thousands of potential drugs or toxic substances on
NMJ morphology. Automated microscopic processes lead to
the production of thousands of images and large amounts of
data. This has led to a growing need for automated image
processing and analysis to generate accurate results and elimi-
nate the time consumption and biases in the manual analysis
process. Unfortunately, the necessary image analysis goes be-
yond the capabilities of standard commercially available solu-
tions, such as the software provided with microscopes or is
confined to the specific format of images generated by some
microscopes, which are available at higher costs and are not
accessible to everyone. Thus, the creation of a specialized im-
age processing technique for zebrafish NMJ analysis is critical
in realizing the full potential of the data contained in the
images collected.
Zebrafish NMJ images acquired via confocal microscopy
are typically large, which makes processing, and analysis a
hefty task (Fig. 1). For instance, analyzing the NMJ morphol-
ogy in a single 6 days postfertilization (dpf) zebrafish requires
at least two channels to label the pre- and postsynapse. To ob-
tain a comprehensive z-stack, at least 80–100 slices per chan-
nel must be taken at 40magnification, with an optimal slice
distance of 0.44 mm. This results in nearly 300 slices (presy-
napse þpostsynapse þmerged) with numerous puncta on
each slice to be analyzed. Furthermore, when multiple zebra-
fish are analyzed for various drug treatments at various time
points, these numbers rapidly increase, making the analysis
process extremely complex and difficult.
To address this hefty task, we developed an automated im-
age analysis program specifically focusing on the analysis of
zebrafish NMJ morphology. In order to make our method
more accessible and applicable, we have integrated it into an
open-source image analysis software Fiji (Schindelin et al.
2012). Fiji provides a custom graphical user interface and
comprehensive documentation to guide users through the pro-
cess. An automated workflow pipeline has been established
for accessing confocal zebrafish NMJ files to create z-projec-
tions. Users are able to define their region of interest (ROI)
within the z-projection, typically a hemisomite region, and
an automated analysis will be run on multiple channels for
each defined hemisomite in the image. The analysis results
for each hemisomite include: (i) the number of presynaptic
puncta, (ii) the number of postsynaptic puncta, (iii) the num-
ber of presynaptic puncta co-localizing with postsynaptic
puncta, and (iv) the number of postsynaptic puncta co-
localizing with presynaptic puncta. Images of zebrafish NMJ
obtained through confocal microscopy frequently exhibit
clustering of puncta due to limitations in resolution, result-
ing in inconsistent manual counting of puncta and biased co-
localization analysis. This can compromise the accurate
analysis of NMJ morphology, which is crucial in the study
of many NMDs. To address this challenge, we have devel-
oped an automated image processing approach that incorpo-
rates an unbiased segmentation algorithm to detect and
segment clustered puncta. Our segmentation algorithm is
based on the analysis of three parameters: area, circularity,
and aspect ratio, to accurately identify clustered puncta. Our
program will enable users to perform high-throughput, high-
accuracy data analysis in a short time frame through an un-
biased approach.
Figure 1. Manual analysis of NMJ morphology is time-consuming and prone to bias. Confocal imaging of zebrafish NMJs, stained with presynaptic
marker SV2a and postsynaptic marker a-bungarotoxin (a-BTX), entails acquisition of 80–100 z-slices (N). Each z-slice contains an estimated 50–100
synaptic puncta (n). The manual quantification of individual puncta within separate channels (2) and the assessment of their co-localization necessitate a
substantial investment of time and effort, particularly when analyzing multiple samples representing diverse experimental conditions and varying time
points. Moreover, limitations in the resolution of confocal microscopy introduce the possibility of under or overestimation of synaptic puncta counts.
2Singh et al.
Downloaded from https://academic.oup.com/bioinformatics/article/39/12/btad720/7460225 by guest on 13 December 2023
2 Materials and methods
2.1 Zebrafish maintenance
Wild-type (AB/TL strain) (Danio rerio) and C9orf72-KD (C9-
miR) amyotrophic lateral sclerosis (ALS) model (Butti et al.
2021) were maintained at 28C at a light/dark cycle of 12/
12 h in accordance with Westerfield zebrafish book
(Westerfield 2000). Embryos were raised at 28.5C, and col-
lected and staged using standard criteria. All experiments
were performed in compliance with the guidelines of the
Canadian Council for Animal Care and the local ethics com-
mittee of INRS.
2.2 Immunohistochemistry
Zebrafish 6 days postfertilization (dpf) larvae were fixed in
4% paraformaldehyde in phosphate buffered saline (PBS) at
4C overnight. The next day, the fish were rinsed 3in a
0.1% Tween-20 in PBS solution (PBS-Tween) for 15 min at
room temperature. The fish were then permeabilized in 1 ml
of 1 mg/ml collagenase (Sigma-Aldrich; C0130-100MG) in
PBS solution for 2.5 h at room temperature on a rotator. The
fish were then rinsed three times with PBS-Tween for 15 min
at room temperature on a rotator. The larvae were placed for
1 h at room temperature in blocking solution (1% bovine
serum albumin, 1% DMSO, 1% Triton-X, 2% normal
goat serum, in PBS). They were then incubated in a
(10 mg/ml) tetramethylrhodamine-conjugated a-bungarotoxin
(Thermofisher T1175) in 0.1% PBS-Tween for 30 min to stain
the postsynaptic acetylcholine receptors. The larvae were
rinsed several times with 1phosphate buffer with 0.1%
Tween 20 for 30 min and then incubated in freshly prepared
blocking solution containing primary antibody SV2 (1:200,
Developmental Studies Hybridoma Bank) overnight at 4C.
The next day, the fish were rinsed three times with PBS-
Tween for 15 min at room temperature on a rotator before
being incubated in a 1:1000 Alexa Fluor 488 goat anti-mouse
(Invitrogen; A10680) in blocking buffer for 4 h at room tem-
perature on a rotator. The fish were then rinsed three times
with PBS-Tween for 15 min at room temperature on a rotator
before being stored in PBS-Tween overnight at 4C without
rotation. The larvae were then mounted the next day in
fluoromount-G (Invitrogen 00–4958-02) on a slide and im-
aged using Zeiss LSM 780 confocal microscope (Carl Zeiss,
Germany). The fish were imaged in Z-stacks with a xy-
resolution of 1024 1024 at 0.21 mm per pixel and a z-reso-
lution of 0.44 mm.
2.3 Software
The macro presented in this program is written in the ImageJ
macro language using the macro editor in FIJI.
3 Results
3.1 Workflow for NMJ morphology analysis and
quantification
The automated image analysis program is designed to analyse
confocal images and identify synaptic puncta within the
images (Fig. 2). The program begins by opening the confocal
images from a user-selected folder and generating a z-projec-
tion for each file present in that folder (Supplementary Video
S1). The user is then prompted to define the ROI, which is
saved in a specified folder. The program then saves the num-
ber of slices present in each image for each channel and per-
forms this process for all the files present in the selected
folder. Next, each slice for each channel is opened and proc-
essed with a series of image processing parameters.
The first step in image processing is to apply a Gaussian
blur with a sigma of 1.0 to the image. To enhance contrast
and detect puncta, the histogram of the image is expanded.
This is achieved by analyzing the minimum and maximum
grayscale values of the image slice, then increasing the mini-
mum and decreasing the maximum grayscale values by 10%
of the dynamic range (max. to min. grayscale value) of the im-
age. After histogram expansion, Otsu thresholding (Otsu
1979) is applied to the image slice. The image generated by
Otsu thresholding is then despeckled by applying an area pa-
rameter of 0.043 mm
2
and removing all pixels selected by
Otsu that have an area of 0.043 mm
2
, to filter out noise.
The program uses a Fiji plugin to analyse particles and
store the saved ROIs of each particle (which represents a
puncta) detected in the slice. By manual analysis of multiple
single and clustered puncta, we identified three parameters to
which can differentiate clustered puncta from individual
puncta; area, circularity, and aspect ratio. Any particle
detected by Otsu whose area is >4mm
2
, circularity is <0.65,
and aspect ratio is >2.5 is considered to be a clustered punc-
tum and is further selected for segmentation.
Our automated analysis program reopens the image slice
and the ROIs of the clustered puncta for that particular image
slice. A Gaussian blur with a sigma value of 1.0 pixel is ap-
plied to the image slices. The ROIs marked as clustered
puncta are selected from the slice and the “Find Maxima”
command is run. If more than one “maxima” are reported
within a clustered punctum, the xand ycoordinates of the
“maxima” are recorded. Then, a new binary image of the
same dimension is created. The maximum coordinates in the
new binary image are assigned a grayscale value unique to
each “maxima”. The program then executes an unbiased ex-
pansion algorithm to segment the puncta into nparts, where
nis the number of “maxima” reported. The expansion con-
tinues until the initial ROI is completely distributed among
the nmaxima. The segmented puncta are then filtered based
on their area, circularity, and aspect ratio. If the area is
smaller than 4 mm
2
, the circularity is >0.65, and the aspect ra-
tio is smaller than 2.5, the segmented puncta are considered
to be good/single puncta. If the conditions are not met, the
puncta are further segmented until they meet the parameters.
After the segmentation process, good puncta from each slice
and each channel of each image file are used for co-
localization analysis. The program creates two binary images
of the same image slice number for Channels 1 and 2, with
the puncta of each slice for their respective channel. The
Boolean function AND is run on both slices belonging to
Channels 1 and 2, and the number of single, co-localized, and
total puncta is reported in a tabular form for each hemisomite
or ROI for each image. A punctum is considered to be individ-
ual if the percent of its area covered by a punctum of the other
channel falls below a specific threshold and to be co-localized
if the covered area percent falls above a specific threshold.
The result table reports the population of individual and co-
localized punctum with progressively higher thresholds, from
>0% to ¼100% in increments of 10%, resulting in 11
thresholds.
3.2 Image processing and thresholding
After the ROI has been defined in the respective images, the
slices of the z-stack for each channel and each image are
NMJ Analyser 3
Downloaded from https://academic.oup.com/bioinformatics/article/39/12/btad720/7460225 by guest on 13 December 2023
stored in their corresponding folders. Subsequently, each slice
for each channel is subjected to image processing parameters.
The first step in image processing involves applying a
Gaussian blur with a sigma of 1.0 to the slice. Gaussian blur
is often used in scientific image processing to reduce image
noise and smooth out image intensity variations, which can
help improve the results of subsequent image analysis steps.
The Gaussian blur works by convolving the image with a
Gaussian filter, which has a bell-shaped intensity distribution.
This convolution effectively replaces the intensity values of
each pixel with a weighted average of its neighbors, which
has the effect of smoothing out sharp intensity transitions and
Figure 2. NMJ analysis workflow overview (Steps 1–7). Z-projection are created of the confocal images where users are prompted to define the ROIs.
Each z-slice is processed for contrast enhancement and punctum detection via Otsu thresholding. Clustered puncta are segmented into single punctum
and co-localization analysis is performed for each punctum in the defined ROI.
4Singh et al.
Downloaded from https://academic.oup.com/bioinformatics/article/39/12/btad720/7460225 by guest on 13 December 2023
reducing noise. By using a Gaussian blur, the features of inter-
est in the image slice, such as puncta, can become more appar-
ent and easier to detect.
Then, we utilized histogram expansion to enhance the
image’s contrast while avoiding the saturation of pixels. This
approach allows for the detection of puncta through Otsu
thresholding in Fiji. In the formula we developed for histo-
gram expansion, the minimum and maximum grayscale val-
ues of the image are analyzed, and the extreme 10% of the
grayscale values from the minimum and maximum pixel are
trimmed out (Fig. 3). The histogram expansion (LUT com-
pression) is calculated using the formula:
NMinGV ¼OMin GV þRange
10
NMaxGV ¼OMax GV Range
10
Range ¼OMaxGV OMin GV
NMinGV :New Minimum Grayscale Value
NMaxGV :New maximum Grayscale Value
OMinGV :Original Minimum Grayscale Value
OMaxGV :Original Maximum Grayscale Value
After processing the image with Gaussian blur and histo-
gram stretching, Otsu thresholding (Otsu 1979) is applied to
detect the puncta in the slice. Otsu is a computationally effi-
cient and simple method to separate objects in an image based
on intensity values. The method assumes that the image to be
threshold contains two classes of pixels: foreground and back-
ground. The algorithm tries to find a threshold value and cal-
culates the variance between the two classes and selects the
threshold that maximizes the variance between the fore-
ground and background pixels, which is equivalent to maxi-
mizing the separability between the two classes. The threshold
is calculated based on the intensity histogram of the image,
and the result is a binary image where the pixels with intensi-
ties above the threshold are set to one (foreground) and the
pixels with intensities below the threshold are set to zero
(background). This method is often used in scientific image
processing to separate objects of interest from the back-
ground, making them easier to analyse.
3.3 Puncta analysis
After the detection of puncta by Otsu thresholding, there
remains a limitation in the resolution of confocal microscopy
that can lead to the detection of multiple clustered puncta as
one punctum (Fig. 4). This creates a bias of overestimation or
underestimation of the number of puncta in the image which
can have a significant impact on the assessment of NMJ mor-
phology. To mitigate this issue, we conducted a manual
analysis of both single and clustered puncta in the image and
identified three parameters that define clustered puncta. The
three parameters to detect clustered puncta are:
1) Area >4mm
2
2) Circularity <0.65
3) Aspect ratio >2.5
Our automated analysis program applies these parameters
to each punctum in the defined ROI for each z-slice and chan-
nel in the image. If any of these conditions are satisfied, the
puncta are considered clustered and are further segmented by
Figure 3. Distribution curve demonstrating histogram LUT compression. Increasing the minimum and decreasing the maximum grayscale values by 10%
of the dynamic range (max. to min. grayscale value) of the image slice increases the contrast of the image for puncta detection. NM
in
GV: New Minimum
Grayscale Value, NM
ax
GV: New Maximum Grayscale Value, OM
in
GV: Original Mininum Grayscale Value; OM
ax
GV: Original Maximum Grayscale Value.
NMJ Analyser 5
Downloaded from https://academic.oup.com/bioinformatics/article/39/12/btad720/7460225 by guest on 13 December 2023
our unbiased segmentation algorithm until the parameters of
a single punctum are met.
3.4 Segmentation
Any puncta marked as a clustered punctum will be further
segmented into smaller puncta (Supplementary Video S2).
The newly generated sub-puncta will recursively be analysed
according to the Puncta analysis parameters described previ-
ously and segmented until they either fulfill all the criteria or
the segmentation algorithm is unable to segment them further
(Fig. 5).
The segmentation algorithm begins by applying a Gaussian
blur with sigma ¼1.0 pixels to the image. The program then
goes down the list of puncta detected by puncta analysis and
segments those marked as clustered puncta. The “Find
Maxima ...” ImageJ command is used to detect the presence
of local brightness maximum within the punctum to be seg-
mented. If no or a single maximum is found, then the cluster
punctum is left as it is and the next punctum is analysed. If
multiple “maxima” are detected within the cluster punctum,
then the coordinates for each “maxima” is recorded in a list
to be used as expansion nuclei in the expansion algorithm.
The expansion algorithm uses the coordinates of each
“maxima” as the point from which to expand each sub-
punctum within the clustered punctum (Fig. 6A, Pixel 0 in red
and green). It first checks whether the pixel to the top left of
the first maxima is both empty and within the clustered punc-
tum. If yes, the top left pixel is marked as part of the sub-
punctum belonging to the first maxima and this pixel is added
to the list of pixels associated with the first maxima (Fig. 6B,
Red pixel 1). If the pixel is already taken by another maxima
(Fig. 6G–I in red), then the pixel is not added to the list. The
same process is performed for the other maxima in the clus-
tered punctum (Fig. 6C, Green pixel 1). Subsequently, the
pixel to the left, bottom left, below, bottom right, right, top
right, and above pixel 0 for both maxima (red and green) are
checked. Once this is complete, the algorithm will select the
next pixel (Fig. 6K, Pixel 1, red and green) as the center from
which to expand according to the same rules. The expansion
algorithm stops when the area within the clustered punctum
has entirely been distributed between the local maxima.
After the expansion algorithm stops, each contiguous re-
gion around each original local maxima is saved as a new
punctum (Fig. 6M and N). The old clustered punctum is re-
moved from the list of puncta while the new puncta are added
to the list and flagged for a new round of segmentation analy-
sis. The segmentation algorithm ends when it has checked
every punctum on the list and either segmented them or deter-
mined that they cannot be segmented any further.
3.5 Co-localization analysis
As a final output, NMJ Analyser will tally up the total num-
ber of puncta detected in each channel and the number of
puncta that do not overlap and the number of puncta that do
overlap (co-localize) with puncta of the other channel. To de-
termine whether a specific punctum is co-localized with the
other channel, NMJ Analyser uses fractional overlap as the
metric for co-localization. NMJ Analyser will measure the
fraction of each punctum’s area that overlaps with the other
channel and past a threshold of overlap, the punctum is con-
sidered to be co-localized. NMJ Analyser reports the fraction
of puncta in both channels that are co-localized based on 11
thresholds: >0%, where a punctum is considered co-localized
even if a single pixel overlaps with the other channel, through
>10%, >20% ..., 90%, to ¼100%, where a punctum is con-
sidered co-localized only if its entire area is covered by a punc-
tum of the other channel.
To perform this analysis, NMJ Analyser opens the binary
images representing the detected puncta generated by Otsu
thresholding earlier for the two channels on a single slice of
the image (Fig. 7A). A Boolean “and” operation is performed
with both images as inputs. The result is an image that marks
only the region where the two initial channels overlap
(Fig. 7B). The area representing each punctum is overlaid
onto the “and” image and the percent of each punctum cov-
ered by the other channel is calculated. NMJ Analyser that
tabulates all the puncta into a table (Fig. 7C) that tallies up
the number total number of puncta per channel as well as the
number of single and co-localized puncta (Fig. 7C). Whether
a punctum counts as a single or co-localized punctum depends
on whether the fraction of its area covered by the other chan-
nel reaches one of the 11 predefined thresholds.
3.6 NMJ co-localization in WT and C9orf72 zebrafish
ALS model
ALS is a fatal neurodegenerative disorder characterized by the
selective loss of motor neurons. One of the primary pathologi-
cal processes in ALS is the disruption of NMJ morphology,
caused by dysfunction in both the pre- and postsynaptic com-
ponents. The most common genetic mutation associated with
ALS is the hexanucleotide repeat expansion GGGGCC (G4C2)
found in the first intronic region of the C9orf72 gene, present
in 40% of familial ALS cases. We recently developed a zebra-
fish ALS model with knockdown of the c9orf72 gene (hereafter
Figure 4. Image processing and Otsu thresholding for puncta detection. The original image was subjected to a Gaussian blur and contrast enhancement
algorithm to enable the detection of puncta via Otsu thresholding. However, the limitations in resolution posed challenges in properly separating clustered
puncta. The lack of resolution limits the clustered puncta to be separated. Dashed circle represents the clustered puncta; solid circle represents the single
punctum. Scale bar ¼2.5 mm.
6Singh et al.
Downloaded from https://academic.oup.com/bioinformatics/article/39/12/btad720/7460225 by guest on 13 December 2023
referred as the C9-miR model) that replicates the characteristic
features of ALS observed in patients (Butti et al. 2021).
The C9-miR zebrafish exhibited early motor impairments,
characterized by decreased swimming distance and velocity.
Furthermore, at 6 dpf, the C9-miR zebrafish displayed altered
NMJ morphology (Fig. 8A). We employed our program to as-
sess the NMJ morphology by automatically quantifying the
presynaptic (SV2a), postsynaptic (a-BTX), and co-localizing
punctae in WT versus C9-miR fish. Our analysis revealed a
significant decrease in the number of presynaptic (SV2a) and
postsynaptic (a-BTX) puncta in C9-miR zebrafish compared
to WT zebrafish (Fig. 8B and C). Additionally, we observed a
significant reduction in the co-localization of presynaptic
(SV2a) puncta with postsynaptic (a-BTX) puncta, as well as a
decrease in the co-localization of postsynaptic (a-BTX) puncta
with presynaptic (SV2a) puncta in C9-miR zebrafish com-
pared to WT zebrafish (Fig. 8D).
These results demonstrate the effectiveness of our program
in accurately quantifying defects in NMJ morphology. The
successful application of our program in assessing NMJ ab-
normalities in the context of ALS suggests its potential utility
for quantifying NMJ defects in other NMDs, such as spinal
muscular atrophy (SMA), Duchenne muscular dystrophy
(DMD), Charcot-Marie-Tooth disease (CMT), myasthenia
gravis (MG), and others. This highlights the broader applica-
bility of our program in investigating NMJ defects in various
neuromuscular disorders and advancing our understanding of
their underlying pathogenic mechanisms.
4 Discussion
In this study, we present a specialized image analysis program
for the analysis of zebrafish NMJ morphology. Zebrafish are
widely utilized in the study of neurodevelopment and NMDs,
Figure 5. Flowchart of Segmentation Algorithm. The segmentation algorithm compiles a list of all detected puncta into a list, marking the clustered
puncta. If a clustered punctum can be segmented, the segmented puncta are added to the list, the original punctum is removed from the list, and the
next punctum in the list is analysed. If a punctum cannot be segmented or was not marked as a clustered, the next punctum in the list is analysed. These
steps are repeated until all puncta have been analysed.
NMJ Analyser 7
Downloaded from https://academic.oup.com/bioinformatics/article/39/12/btad720/7460225 by guest on 13 December 2023
Figure 6. Expansion Algorithm. The expansion algorithm cycles through the eight neighboring pixels of each local maximum (A), beginning with top left (B
and C), then cycling through each direction (D–J), assigning the neighboring pixel to a local maximum if the pixel is both unclaimed and within the
punctum region. Once a cycle of eight directions has been completed, it is repeated with the first expansion as the origin of expansion (K). This process is
repeated until the entire region of the punctum has been assigned to a local maximum (L–N).
Figure 7. Co-localization analysis. Co-localization is based on percent overlap. (A and B) A Boolean “AND” operation is performed pixel-wise across the
two channels of an image, resulting in an image where only pixels that were common to both channels remain white. The program then calculates the
area in each punctum taken up by white pixels. A punctum is classified as “single” if the fraction of white pixels falls below the co-localization threshold
and “colocalized” if the fraction is above the threshold. (C) The program tabulates the number of “single” puncta, “colocalized” puncta, sorted according
to different thresholds from >0% to ¼100%.
8Singh et al.
Downloaded from https://academic.oup.com/bioinformatics/article/39/12/btad720/7460225 by guest on 13 December 2023
and the analysis of NMJ morphology is one of the key indica-
tors for motor function (Singh and Patten 2022). The genera-
tion of large amounts of data through genetic screening and
high-throughput drug assays poses a significant challenge for
the manual analysis of NMJ morphology, as it is a time-
consuming task. Moreover, the limitations in the resolution of
microscopes often result in images that introduce biases dur-
ing manual analysis. To address these challenges, we have de-
veloped an automated image analysis program integrated into
the open-source software Fiji (Schindelin et al. 2012). The
program is designed to address the challenges of manual
analysis and overcome the limitations of commercial solu-
tions. By integrating the program into an open-source image
analysis software Fiji, the method is made more accessible
and applicable for researchers.
The program provides an automated workflow pipeline for
confocal zebrafish NMJ files to create z-projections and allows
users to define their ROI within the z-projection. The results of
the analysis include the number of presynaptic and postsynaptic
puncta, as well as the number of presynaptic and postsynaptic
puncta co-localizing with each other. To ensure accurate analysis
of NMJ morphology, we have incorporated an unbiased segmen-
tation algorithm based on the analysis of three parameters, area,
circularity, and aspect ratio, to detect and segment clustered
puncta. Our program provides high-throughput, high-accuracy
data analysis in a short time frame through an unbiased
approach.
We applied our specialized image analysis program to as-
sess the NMJ morphology in a zebrafish model of ALS in-
duced by the knockdown of c9orf72 gene. Our laboratory
created a loss-of-function zebrafish model by targeted
miRNA gene silencing approach, resulting in the specific and
ubiquitous knockdown of the endogenous c9orf72 gene ex-
pression (Butti et al. 2021). At 6 dpf, the C9-miR fish exhib-
ited early motor impairments, including reduced swimming
distance and velocity, and disruptions in NMJ morphology
Figure 8. NMJ Morphology in Control versus C9-miR Zebrafish. (A) Co-immunostaining of zebrafish NMJs with presynaptic (SV2a; green) and
postsynaptic (a-bungarotoxin; red) markers at 6 dpf. Scale bar ¼100 mm. (B) Quantitative analysis demonstrated a significant decrease in the number of
presynaptic (SV2a) puncta in C9-miR zebrafish (363.9 6161.7) compared to WT zebrafish (492.7 6139.2). (C) Quantitative analysis demonstrated a
significant decrease in the number of postsynaptic (a-BTX) puncta in C9-miR zebrafish (430.0 6177.8) compared to WT zebrafish (616.9 6243.7). (D) Co-
localization analysis revealed a significant decrease in the co-localization of presynaptic (SV2) puncta with postsynaptic (a-BTX) puncta in C9-miR zebrafish
(131.7 655.2) compared to WT zebrafish (188.4 659.5). Similarly, co-localization analysis of postsynaptic (a-BTX) puncta with presynaptic (SV2) puncta
revealed a marked reduction in co-localizing puncta in C9-miR zebrafish (122.1 656.5) compared to WT zebrafish (180.3 662.6). n¼24–28; **P<0.001.
Data are presented as mean 6SD. nrepresents the number of hemisomites.
NMJ Analyser 9
Downloaded from https://academic.oup.com/bioinformatics/article/39/12/btad720/7460225 by guest on 13 December 2023
(Butti et al. 2021). Our program was used to quantify the
NMJ defects in the C9-miR fish, demonstrating its utility in
evaluating NMJ morphological defects in other NMDs, such
as SMA, CMT, MG, and DMD etc.
In order to enhance the applicability and versatility of our pro-
gram, we have incorporated an advanced mode. This mode
empowers users to adjust various parameters according to their
specific requirements. For instance, upon initiation of the pro-
gram, users are presented with the option to either conduct im-
age segmentation or skip this step. If they opt for segmentation,
they can either utilize the default parameters of punctum analy-
sis, which are optimized for Zebrafish NMJ as: (i) area >4, (ii)
circularity <0.65, and (iii) aspect ratio >2.5. Alternatively, users
can specify their own values for the parameters of area, circular-
ity, and aspect ratio for each channel. This advanced mode will
enable the analysis of not only Zebrafish NMJ morphology, but
also other scientific images, including mouse NMJ morphology,
different subcellular compartments, and surface receptors. This
level of customization allows for an unbiased and time-efficient
analysis of scientific images, increasing the utility and effective-
ness of our program.
In conclusion, the development of the automated image
analysis program for zebrafish NMJ morphology is a signifi-
cant step forward in the study of NMDs and the use of zebra-
fish as a model organism. Additionally, the wide application
of the program will provide a more efficient and accurate so-
lution for researchers for image analysis and will open up new
possibilities for the analysis of development and disease
mechanisms.
Supplementary data
Supplementary data are available at Bioinformatics online.
Conflict of interest
None declared.
Funding
This work was supported by the Natural Science and
Engineering Research Council (NSERC) (to S.A.P.); Canadian
Institutes of Health Research [CIHR-177940]; and the Anna
Sforza Djoukhadjian Research Chair in ALS and a FRQS
Junior 2 research scholar. J.S. is supported by a Pierre Auger
Morissette Capacity-Building Award in ALS Research from
Brain Canada Foundation.
References
Bailey EC, Alrowaished SS, Kilroy EA et al. NADþimproves neuromus-
cular development in a zebrafish model of FKRP-associated dystro-
glycanopathy. Skelet Muscle 2019;9:21.
Butti Z, Pan YE, Giacomotto J et al. Reduced C9orf72 function leads to
defective synaptic vesicle release and neuromuscular dysfunction in
zebrafish. Commun Biol 2021;4:792.
Chen F, Yuan W, Mo X et al. Role of zebrafish fhl1A in satellite cell and
skeletal muscle development. Curr Mol Med 2017;17:627–36.
Chen X, Huang C, Wang X et al. BDE-47 disrupts axonal growth and
motor behavior in developing zebrafish. Aquat Toxicol 2012;
120-121:35–44.
D’Elia KP, Dasen JS. Development, functional organization, and evolu-
tion of vertebrate axial motor circuits. Neural Dev 2018;13:10.
Dubi
nska-Magiera M. Zebrafish: a model for the study of toxicants af-
fecting muscle development and function. Int J Mol Sci 2016;17:
1941.
D’Ydewalle C, Krishnan J, Chiheb DM et al. HDAC6 inhibitors reverse
axonal loss in a mouse model of mutant HSPB1-induced Charcot-
Marie-Tooth disease. Nat Med 2011;17:968–74.
Giacomotto J, Rinkwitz S, Becker TS et al. Effective heritable gene
knockdown in zebrafish using synthetic microRNAs. Nat Commun
2015;6:7378.
Howe K, Clark MD, Torroja CF et al. The zebrafish reference genome
sequence and its relationship to the human genome. Nature 2013;
496:498–503.
Hutson LD, Chien CB. Wiring the zebrafish: axon guidance and synap-
togenesis. Curr Opin Neurobiol 2002;12:87–92.
Jing L, Gordon LR, Shtibin E et al. Temporal and spatial requirements
of unplugged/MuSK function during zebrafish neuromuscular devel-
opment. PLoS One 2010;5:e8843.
Jontes JD, Buchanan J, Smith SJ et al. Growth cone and dendrite dynam-
ics in zebrafish embryos: early events in synaptogenesis imaged
in vivo. Nat Neurosci 2000;3:231–7.
Lescouze` res L, Hassen-Khodja C, Baudot A et al. A multilevel screening
pipeline in zebrafish identifies therapeutic drugs for GAN. EMBO
Mol Med 2023;15:e16267.
Luxey M, Berki B, Heusermann W et al. Development of the chick wing
and leg neuromuscular systems and their plasticity in response to
changes in digit numbers. Dev Biol 2020;458:133–40.
Mart
ınez-Herna´ ndez R, Bernal S, Also-Rallo E et al. Synaptic defects in
type I spinal muscular atrophy in human development. J Pathol
2013;229:49–61.
Myers PZ, Eisen JS, Westerfield M et al. Development and axonal out-
growth of identified motoneurons in the zebrafish. J Neurosci 1986;
6:2278–89.
Otsu N. Threshold selection method from gray-level histograms. IEEE
Trans Syst Man Cybern 1979;9:62–6.
Panzer JA, Gibbs SM, Dosch R et al. Neuromuscular synaptogenesis in
wild-type and mutant zebrafish. Dev Biol 2005;285:340–57.
Patten SA, Aggad D, Martinez J et al. Neuroleptics as therapeutic com-
pounds stabilizing neuromuscular transmission in amyotrophic lat-
eral sclerosis. JCI Insight 2017;2:e97152.
Sainath R, Granato M. Plexin A3 and turnout regulate motor axonal
branch morphogenesis in zebrafish. PLoS One 2013;8:e54071.
Schindelin J, Arganda-Carreras I, Frise E et al. Fiji: an open-source plat-
form for biological-image analysis. Nat Methods 2012;9:676–82.
Singh J, Patten SA. Modeling neuromuscular diseases in zebrafish. Front
Mol Neurosci 2022;15:1054573.
Skobo T, Benato F, Grumati P et al. Zebrafish ambra1a and ambra1b
knockdown impairs skeletal muscle development. PLoS One 2014;
9:e99210.
Son J-H, Stevenson TJ, Bowles MD et al. Dopaminergic co-regulation of
locomotor development and motor neuron synaptogenesis is
uncoupled by hypoxia in zebrafish. eNeuro 2020;7:12.
Westerfield M The zebrafish book : a guide for the laboratory use of
zebrafish (Brachydanio rerio). M. Westerfield, Eugene, OR. 1993.
Xu J, Zhu J, Li Y et al. Three-dimensional mapping reveals hetero-
chronic development of the neuromuscular system in postnatal
mouse skeletal muscles. Commun Biol 2022;5:1200–14.
Yin X, Liu W, Chen H et al. Effects of ferulic acid on muscle develop-
ment and intestinal microbiota of zebrafish. J Anim Physiol Anim
Nutr (Berl) 2022;106:429–40.
10 Singh et al.
Downloaded from https://academic.oup.com/bioinformatics/article/39/12/btad720/7460225 by guest on 13 December 2023
