1
Supplementary Material for:
Massively parallel C. elegans tracking provides
multi-dimensional fingerprints for phenotypic
discovery
Michele Perni1+, Pavan K. Challa1+, Julius B. Kirkegaard2+, Ryan Limbocker1, Mandy
Koopman3, Maarten C. Hardenberg3, Pietro Sormanni1, Thomas Müller1, Kadi L. Saar1,
Lianne W. Y. Roode1, Johnny Habchi1, Giulia Vecchi1, Nilumi Fernando1, Samuel Casford1,
Ellen A. A. Nollen3, Michele Vendruscolo1*, Christopher M. Dobson1* and Tuomas P. J.
Knowles1*
1Centre for Misfolding Diseases, Department of Chemistry, University of Cambridge,
Cambridge CB2 1EW, UK 2Department of Applied Mathematics and Theoretical Physics, University of Cambridge,
Cambridge CB3 0WA, UK 3University of Groningen, University Medical Center Groningen, European Research
Institute for the Biology of Aging, 9713 AV Groningen, The Netherlands.
+Those authors made equal contributions to this study
*To whom correspondence should be addressed: [email protected], [email protected], [email protected].
2
Supplementary Text
Development of the WF-NTP algorithm
In combination with our WF-NTP scope (Fig. S1), we also generated a custom made software
and associated GUI (Fig. S2) to set up image processing and experimental parameters, in
order to facilitate the analysis by the users. The version of WF-NTP used in this study is
available as Supplementary Software 1 and Python is required to run and to modify code. In
order to allow the WF-NTP to be widely used in many different laboratories we also provide
an installer package of all the required Python libraries, which can be automatically
downloaded online, and can help users that are not familiar with Python to install and use our
code. Together with a user-friendly graphical interface (GUI), this can help researchers that
are unfamiliar with Python to install and run our code.
As a first step, our code initially detects and subtracts the background signal, consisting of
non-moving objects such as small particles and shadows from the agar plate, and offers an
alternative approach to removing the background signal without using temporal information.
The method begins by defining a Gaussian adaptive threshold of the image; in this way all
worms are identified but with many false-positives. Afterwards, all pixels marked as worms
are recalculated by interpolating from pixels not marked as worms; the procedure avoids
removing immobile worms as part of the background. The software design is highly robust
against false positives and indeed allows a definition of the number of worms that is very
close to the number present on the plates (as obtained by manual counting) (Fig. S3a).
Morphological operations are then employed in our software to remove speckles and to close
holes that can appear in the images after thresholding (Van der Walt et al., 2014). These
operations can be easily modified by changing the ‘opening’ and ‘closing’ parameters in a
GUI. Subsequently, all regions of the thresholded image are labelled, at which time all objects
that are too small or too big to be a single animal are size excluded. In addition, modifications
of the ‘minimum size’ and ‘maximum size’ parameters allow the removal of background
particles and the selection of objects that are of sizes comparable to that of worms. This
procedure is highly customizable, easy to manipulate, and gives visual readouts to ensure
accurate and precise tracking of all worms prior to the analysis.
3
After this operation, the remaining labelled regions are identified as individual worms and the
positions of those regions are then stored for each frame. These regions are linked across the
frames by mean of a standard tracking algorithm (Allan et al., 2014), which takes into account
a worm ‘disappearing’ or overlapping in a few frames; this issue can be controlled using the
parameter ‘memory’. This parameter brings an additional edge on our tracking procedure and
a further aid to minimising the ‘collision’ problem; this works in combination and in addition
to the use of our wide-surface screening procedure that reduces the chance of the above-
mentioned events (Supplementary Video 1 and Fig. S3b). This issue cannot readily be
addressed by more conventional algorithms that discard the worms tracks during collisions
events, or require a much higher image quality, the attainment of which results in lower
throughput and larger video files.
In order to design the experiments better, we also estimated the chances of collisions as:
𝑝 = [𝑤] × (!)!
! Eq. S1
Where [p] is the average number of collisions in a random frame, [w] is the effective worm
area, [A] is the disk area, [N] is the number of worms.
The parameter [p] aids significantly the design of experiments on the basis of the number of
worms and the size of the plates.
Finally, we convert the worm images to single-pixel skeletons, which can then be pruned to
remove spurious features. The eccentricity of each tracked worm, a measure of the ratio of the
major and minor ellipse axes, can then be used to estimate the extent of worm bending as a
function of time. We also allow deletion of objects that are not worm-like (i.e. not
anisotropic) by placing a threshold on this eccentricity (called the worm-like parameter).
The software also outputs, in parallel with the analysis, a thresholded video (Fig. S2), which
allow the user readily to check for the robustness of the tracking of individual worms and of
the parameters used for the tracking.
4
Using this method, individual worms can be tracked over time, and plots of their movements
can be extracted to give visual information about their motility levels. It is important to note
that our analysis provides an upper limit on the errors by considering the maximum number of
worms detected in a single frame rather than the total number of worms detected in the entire
video. In particular this procedure avoids underestimation of the errors due to worms that
appear as new in later frames of the data and hence in the tracking analysis.
The length of the recording is also likely to influence the measures of worm fitness, as the
worms can become fatigued when swimming over time; we demonstrate, however, that
recordings up to 3 min or less do not show significant differences in the thrashing frequencies
(Fig. S3); our protocol is moreover based on standard manual thrashing assays where the
motility of single worms is manually scored with the aid of a stereoscope for up to 30 s
(Gidalevitz et al., 2009).
The WF-NTP software computes a variety of metrics and can provide population-averaged or
animal-by-animal output, as appropriate. Metrics include body bends / min, speed, paralysis
rates, areas per animal and mean errors. Averages of all the computed metrics are saved to a
hard drive.
In addition to statistics derived with the retention of paralyzed animals, the code outputs all
the aforementioned metrics for mobile worms only, an approach applicable to situations
where only the moving fraction is pertinent. In addition, metrics can be saved on an animal-
by-animal basis or visualized as a color-coding of the tracks of the animal on a 2D map. We
have also generated a set of tools to extend the capability of the platform, e.g. for exporting
files to TSV format, plotting tracks and fingerprinting, that can be used to visualise the data
produced by the tracker in a rapid and supportive way. The fingerprint tool creates a radar
chart of selected strains (Figs. 3-4) and parameters when loading the text files, which are
created during the analysis, into the fingerprint software.
Development of the screening protocol
As previously mentioned, we designed our screening protocol on the basis of a standard
thrashing assay, and further developed it with a particular focus on drug discovery. In the
manual thrashing assays, single worms are first picked (with a sterile worm picker) and then
5
transferred onto a drop of M9 buffer, after which they are left 30 s to acclimate, upon which
their individual motilities are screened for another 30 s under a stereoscope.
We implemented this protocol in a fully automated and high-throughput fashion. Our C.
elegans populations are cultured at 20 or 24 °C and developmentally synchronized from a 4 h
egg-lay. At 64-72 h post egg-lay (time zero) individuals are transferred to FUDR plates, in the
presence or absence of drugs, with a maximum number of 700 worms per plate to avoid
starvation.
At different ages (i.e. at selected time-points) the entire plate containing the 700 animals is
washed with M9 buffer and transferred to an unseeded 9 cm NGM plate. The motility of the
animals is then recorded for an amount of time defined by the user; although prolonged
observation times are likely to affect the worm motility, as the animals grow tired of
swimming. We have shown that the increase in recording times affects only minimally the
motility readouts and is robust for recordings up to 3 min (Fig. S3c). In general, our
recordings were carried out at 20 fps for 1 min. In order to increase further the statistical
significance of our experiments, we washed 3 FUDR plates containing 700 worms each,
considering these as a triplicate set of data, and screened them in parallel, by achieving a total
number of ca. 2100 screened worms, in less than 3 min; by comparison, in a standard manual
assay 5x3 worms are manually screened, for a total of 15 animals, a procedure taking more
than 20 min.
We also adapted our screening procedure for multiwells (i.e. purely in liquid) studies (Fig. 5)
in order to facilitate high-throughput and multiple condition studies; when screening the
multiwell plates, worms were stored at 20 °C prior to the L4 stage, and at 23.5 °C thereafter,
in order to induce a phenotype, depending on the disease strain. Prior to screening, a bench-
top plate shaker was used at 750 rpm for 1 min to distribute sedimented OP50 and induce full
worm motility. Immediately after shaking, the worms were staged on the WF-NTP platform
and the file collection was initiated 60 s after shaking for 1 min at 20 fps.
Definition of paralysis rates and experimental reproducibility
The videos acquired are analysed using our custom made tracking code which measures
different metrics, including body bends / min, swimming speed, and paralysis assays. We
6
decided that worms showing less than 5 body bends /min and moving at less than 0.1 mm /
min were to be considered as paralysed, although this parameter is tuneable by the user.
As an internal control, to evaluate if our platform was detecting only “non-moving” animals
instead of paralysed worms, we carried out experiments in which we exposed wild type
worms to increasing concentrations of two different paralysing agents, sodium azide and
levamisole, which are commonly used in C. elegans research. We observed a clear dose
dependent increase in the paralysis rates in worms exposed to levamisole (Fig. S3d) or
sodium azide (Fig. S3e), and complete paralysis was observed at 50 and 500 µM,
respectively.
We also measured the performance of our platform in the context of experimental
reproducibility. In order to define this point, we designed two different types of experiments.
In a first set of experiments, we analysed biological replicates by measuring the motility of
wild type worms after 4 independent age synchronization procedures. Here we noticed a
significant spread in the worm populations, underlining the need for a wide-sampling
approach. Indeed, we did not observe any significant differences between the 4 replicates
using the WF-NTP (P>0.1) (Fig. S3f). In an additional set of experiments, we assessed
potential variability between technical replicates, by growing a large population of wild type
worms and measuring the motility of sub-populations at day 4 of adulthood. As with the
biological replicates, we did not observe any significant differences between the technical
replicates (p>0.1). Thus, the WF-NTP allows for consistent measurement of C. elegans
motility between different experiments.
7
Figure S1. Assembly steps and component lists for different types of WF-NTPs. The WF-
NTP was developed to be easily assembled with standard optical components (A) in a fast and
user-friendly modular sequence (B-I). Different WF-NTPs can be built depending on the
available price range and required applications (J-K). Modified WF-NTPs can generate a
higher quality of imaging (L), which may be more suitable for specific types of investigation
(e.g. crawling studies) where more complex aspects of worm behavior need to be evaluated.
8
Figure S2. Graphical User Interface (GUI) of the WF-NTP and main steps of the WF-
NTP tracking algorithm. (A) Panel of the WF-NTP that can be used to upload large datasets
(left), and panel of the different parameters that can be tuned for the video analysis (right).
9
These parameters are: 1. Video - Specific videos and frames can be selected for the analysis,
and a pixel conversion factor can be inserted; 2. Locating - One of the two tracking
algorithms can be chosen, and the thresholding parameters can be tuned accordingly. The
opening and closing parameters allow the user to remove the noise and close small holes in
the images of animals, respectively. The skeletonizing function offers an alternative method
of analysis; 3. Filtering – The object size cut-off provides an additional filter for background
noise and the worm-like parameter allows the user to consider only ellipsoidal objects as
worms; 4. Forming Trajectories – Allows the user to select how many frames are needed to
keep the worm position in memory following collisions, and how many pixels a worm can
move between frames to be distinguished by noise. The Minimum Track Length allows the
user to discard worms which as been tracked for only a few frames; 5. Bends and Velocity –
Allows the user to select the threshold required to count a body bend and how many frames
are used to estimate the speed of the animals; 6. Dead Worm Statistics - Allows insertion of
the cut-off for paralysis evaluation; 7. Region of Interest (ROI) - Allows the user to select one
or more ROIs. The output of these ROIs are automatically sorted out in the result files. 8.
Output - Allows the user selection of the output folder and how many tracking example
frames will be produced (See Material and Methods for details). (B) Main steps of the WF-
NTP tracking procedure. The inserts show details of the images at higher magnification.
10
Figure S3. Details of the performance of the WF-NTP tracking platform. a) WF-NTP
enables the number of animals present on the plate (as obtained by manual counting) to be
tracked with high accuracy. b) The WF-NTP parameter “memory” provides an effective
solution to resolve the issue of the effects of collisions, by storing the information on the
tracks of individual worms and identifying the same animals after a collision. c) The
recording time of the experimental procedure was tuned on the basis of the manual screening
procedures used to measure BPM i.e. 30 s; the recording output was not affected by the length
of the recording up to 3 min. d-e) WF-NTP can measure paralysis rates generated by exposure
of the worms to increasing concentrations of two different paralysing agents, sodium azide
and levamisole, respectively. f-g) Reproducibility of measurements using WF-NTP for f)
11
biological replicas (i.e. different batches) and g) technical replicas (i.e. different plates). The
statistical significance was measured using a One-way ANOVA analysis.
Supplementary References
Allan D.B., Caswell T.A., and Keim N.C. 2014, Trackpy v0. 2.
Gidalevitz T., Krupinski T., Garcia S., and Morimoto R.I. Destabilizing protein
polymorphisms in the genetic background direct phenotypic expression of mutant SOD1
toxicity. PLoS Genetics 5, 2009, e1000399–e1000399.
Van der Walt S., Schönberger J.L., Nunez-Iglesias J., Boulogne F., Warner J.D., Yager N.,
Gouillart E., and Yu T. scikit-image: image processing in Python. PeerJ 2, 2014, e453–e453.
Top Related