GeroScience

https://doi.org/10.1007/s11357-019-00124-9

METHODS

WormBot, an open-source robotics platform for survival

and behavior analysis in C. elegans
Jason N. Pitt & Nolan L. Strait & Elena M. Vayndorf & Benjamin W. Blue &
Christina H. Tran & Brendon E. M. Davis & Karen Huang & Brock J. Johnson &
Keong Mu Lim & Sophie Liu & Arash Nikjoo & Anuj Vaid & Judy Z. Wu &
Matt Kaeberlein

Received: 26 September 2019 / Accepted: 14 October 2019

# American Aging Association 2019

Abstract Caenorhabditis elegans is a popular organ- mation of C. elegans survival analysis even in small
ism for aging research owing to its highly conserved laboratory settings with modest budgets.
molecular pathways, short lifespan, small size, and ex-
tensive genetic and reverse genetic resources. Here we Keywords C. elegans . Automated . Survival .
describe the WormBot, an open-source robotic image Healthspan . Open-source
capture platform capable of conducting 144 parallel
C. elegans survival and behavioral phenotyping exper-
iments. The WormBot uses standard 12-well tissue cul- Introduction
ture plates suitable for solid agar media and is built from
commercially available robotics hardware. The Over the last 30 years, the roundworm Caenorhabditis
WormBot is controlled by a web-based interface elegans has provided key insights into the genetic control
allowing control and monitoring of experiments from of longevity. Many of the genetic pathways now known
any internet connected device. The standard WormBot to control aging were first identified in the nematode and
hardware features the ability to take both time-lapse then later determined to play a role in mammalian sys-
bright field images and real-time video micrographs, tems (Kenyon 2011; Uno and Nishida 2016; Bitto et al.
allowing investigators to measure lifespan, as well as 2015). Studies of aging in the worm have capitalized on
heathspan metrics as worms age. The open-source na- their small size, short lifespan, exquisitely defined cell
ture of the hardware and software will allow for users to lineage, and large genetic tool-set to identify specific
extend the platform and implement new software and genetic factors and the tissues in which they operate to
hardware features. This extensibility, coupled with the influence the rate of aging. Nearly all of this progress was
low cost and simplicity of the system, allows the auto- made using manual lifespan collection techniques, in
which investigators examined animals grown on solid
media in petri dishes daily to determine if worms had
Electronic supplementary material The online version of this died since the previous time point by prodding the worms
article (https://doi.org/10.1007/s11357-019-00124-9) contains
with a stick (platinum wire) to elicit movement.
supplementary material, which is available to authorized users.
More recently, several automated approaches to
J. N. Pitt : N. L. Strait : E. M. Vayndorf : B. W. Blue : performing survival analyses have been developed
C. H. Tran : B. E. M. Davis : K. Huang : B. J. Johnson : (Xian et al. 2013; Stroustrup et al. 2013; Puckering
K. M. Lim : S. Liu : A. Nikjoo : A. Vaid : J. Z. Wu : et al. 2017; Churgin et al. 2017). These typically rely
M. Kaeberlein (*)
Department of Pathology, University of Washington, 1959 NE
on photographic monitoring of spontaneous nematode
Pacific St, Seattle, WA 98195, USA movement; however, methods based on vital dyes or
e-mail: kaeber@uw.edu disruption of intestinal permeability have also been
 GeroScience

reported (Gill et al. 2003; Rera et al. 2018). Generally, attempts to automate survival analysis using the LM
the imaging-based approaches fall into two categories. identified some issues which ultimately led us to develop
First, several microfluidic devices have been developed an alternative system. First, the system relies on modified
that allow worms to be kept in liquid media and assayed flatbed transparency film scanners for imaging. The
by light microscopy-based systems (Xian et al. 2013; scanners used in the initial publication are no longer
Banse et al. 2019a; Saberi-Bosari et al. 2018). manufactured, and while the designers of the LM have
Microfluidic systems have some benefits over tradition- issued updates to support newer scanners, we were con-
al solid culture. Through the placement of properly sized cerned that rapid changes in the digital photography
sieves, progeny can be flushed out of the observation world could affect the long-term viability of sourcing
chamber allowing for automated survival experiments components for the LM. Second, the temporal resolution
in the absence of 5-fluorodeoxyuridine (FUDR), which, of the scanned images is suboptimal. While transparency
while traditionally used in nematode survival experi- scanners have very high spatial resolution, in a LM at
ments, has been reported to affect worm physiology normal capacity, they are only capable of capturing an
and have genetic background specific effects image of a single experimental plate once an hour. Ad-
(Burnaevskiy et al. 2018; Van Raamsdonk and Hekimi ditionally, LM protocols call for a modified version of
2011; Wang et al. 2019). In addition, microfluidic sys- nematode growth media (reduced calcium) that differs
tems allow for more precise control over the environ- from media used in previous aging studies in the worm.
mental factors to which the animals are exposed. Al- Furthermore, under normal operation, an LM scanner’s
though microfluidic systems can be powerful for longi- light source travels the length of the scanner bed four
tudinal imaging, they also suffer from drawbacks that times an hour, taking between 10 and 12 min per scan on
impair their utility for studying aging in nematodes. the original model of scanner. This results in long expo-
First, they require culturing animals in an environment sures to intense light which has been reported to have
they typically do not experience in nature, as C. elegans effects upon lifespan in the LM (Banse et al. 2019b).
is not an aquatic species but instead live in soil and in Finally, the physically closed format of the scanner,
rotting fruit (Schulenburg and Felix 2017). Upon expo- which presented thermal management challenges to the
sure to liquid environments, worms activate multiple system’s designers (Stroustrup et al. 2013), also limits
stress response pathways and exhibit a fleeing behavior extension of the device to experiments that require the
often described as “thrashing” that may impact aging worms to be placed in alternative gas environments. We
(Laranjeiro et al. 2019). This also means that it may be have previously reported that hypoxia can increase
difficult to reconcile data generated from animals aged lifespan (Mehta et al. 2009; Leiser et al. 2013) and the
in a microfluidic system with the vast majority of prior limited gas diffusion of plates sealed on a scanner, as
literature in the field, which has been obtained from well as the inability to modify the atmosphere of the
animals cultured on solid agar media and fed E. coli plates sealed to a flatbed scanner was intractable for our
OP50 (Brenner 1974). Finally, microfluidic systems experimental needs.
generally require specialized equipment and expertise Another automated lifespan system that allows for
that most C. elegans laboratories are not equipped with, solid media culture of worms is the WormMotel
limiting utility among the broader community. (Churgin et al. 2017), which uses a high-resolution
The second category of higher-throughput lifespan camera to monitor single worms arrayed into individual
devices involves time-lapse microphotography to record wells in custom PDMS plate format. This system has
images of the worms at fixed time intervals throughout been coupled with a plate handling robot, and since it
the lifespan of the worm on solid media. The most relies on a camera, it is able to record the individually
developed of these systems, the Lifespan Machine housed animal’s movements throughout life to generate
(LM), uses an array of modified flatbed scanners to longitudinal measures. While the WormMotel system is
repeatedly scan low profile petri dishes sealed onto the elegant and powerful, we felt the in-house manufactured
glass bed of the scanner with a rubber gasket (Stroustrup custom plates and picking/seeding individual worms
et al. 2013). This system has been implemented by a into single wells would make it difficult to use for
small number of external labs, including the three sites of high-throughput studies such as drug screens. Addition-
the National Institute on Aging C. elegans Intervention ally, the necessity of the plate handling robot to increase
Testing Program (Banse et al. 2019b). However, our own throughput would limit its utility for many labs.
GeroScience

Here, we describe the WormBot, a new robotic sys- Apache web server (Laurie and Laurie 2003) (httpd.
tem for semi-automated lifespan analysis in C. elegans apache.org). Details of other code dependencies are
which addresses some of the shortcomings of other available via Github. Wormbot software can be installed
systems. The WormBot is built from consumer robotics by cloning the git repository (> git clone http://github.
hardware that allows for solid media culture and use of com/JasonNPitt/wormbot) followed by running the
standard 12-well tissue culture plates to provide high- install script (> sudo ~/wormbot/INSTALL). More
throughput automated survival analysis of C. elegans. detailed software installation instructions are available at
wormbot.org and in the supporting online information.

Methods Media preparation and RNAi experiments

Hardware construction Twelve-well tissue culture plates were purchased from

Genesee Scientific (cat no. 25-101, geneseesci.com, El
A MakeBlock XY plotter robotics kit (MakeBlock.com, Cajon, CA). Each well contained 3 mL of sterilized
Shenzhen, China) was purchased from amazon.com and Nematode Growth Medium (wormbook.org)
assembled according to the manufacturer’s instructions supplemented with 100 units/mL Nystatin to prevent
until page 21 of the build manual (https://github. fungal growth and a final concentration of 50 μM FUDR
c o m / M a k e b l o c k - o f f i c i a l / X Y- P l o t t e r - 2 . 0 to prevent progeny development. For RNA interference
/blob/master/XY%20Plotter%20V2.0%20 (RNAi) experiments, media also contained 100 μg/mL of
Assembly%20Instruction.pdf). Following this step, ampicillin, 10 μg/mL tetracycline and 2mM IPTG. For
assembly diverged and a full list of additional parts is all experiments, worms were synchronized by 10-min
available at http://wormbot.org/bom.xls and in the hypochlorite treatment (1% sodium hypochlorite, 250
Supporting Online Materials. Full video build tutorials mM potassium hydroxide and water) followed by four
are available at http://wormbot.org and on https://www. washes in M9 buffer to isolate eggs, followed by an
youtube.com/channel/UCcbtj864r6CXAUZOVe4 overnight hatch off in 6 mL of M9 in a 60-mm petri dish
vCSg, and in Supporting Online Materials. Optional, sealed with parafilm in a 20 °C incubator (Torrey Pines
additional 3d-printed components were printed using Scientific, Torrey Pines, CA). RNAi clones were taken
1.75 mm PLA filament on a Monoprice MakerUltimate from the Vidal and Ahringer RNAi feeding libraries
3D printer (Monoprice, Rancho Cucamonga, CA). STL (Rual et al. 2004; Kamath et al. 2001), sequence verified
files are available at wormbot.org, thingiverse.com, and and kept as − 80 °C frozen stocks prior to use. dsRNA
in Supporting Online Materials. A 24 × 24 × 3/8 inch induction was performed by growing RNAi feeding
acrylic panel was purchased from TAP plastics strains overnight in media containing antibiotics, but
(tapplastics.com, Oakland, CA) and machined by without IPTG. The following morning, stationary phase
Front Panel Express (frontpanelexpress.com, Seattle, cultures were diluted fourfold with fresh media contain-
WA). CAD files for machining are available at ing 2 mM IPTG and grown for 4 h at 37 °C on a shaker.
http://wormbot.org/wormbot.fpd and in Supporting Following induction, cells were pelleted and resuspended
Online Materials. in the original overnight culture volume (4× concentrate)
and seeded onto RNAi NGM media. Following 20 μL of
Software seeding with RNAi bacteria, 12-well plates were allowed
to dry for 20–60 min with their lids off in a laminar flow
All of the WormBot’s code is open-source and available hood to allow all of the seeding solution to evaporate.
at Github (http://github.com/JasonNPitt/wormbot). The RNAi experiments began by placing synchronized L1
WormBot codebase is primarily written in C++ and worms on 10 cm RNAi plates until they reached the L4
should function on any POSIX operating system; larval stage when they were either hand picked or liquid
however, all testing has been done using Ubuntu 16.04 transferred (in M9) to the seeded 12-well RNAi plates
(Canonical LTD, London, UK). Wormbot software relies containing FUDR. Throughout the experiment, plate hu-
on the OpenCV library (Kaehler and Bradski 2016) midity was maintained by filling the interstitial spaces of
(opencv.org), the FFMPEG project (ffmpeg.org), the the 12-well plate with double-distilled water and refilling
Arduino platform (D'Ausilio 2012) (arduino.cc), and the as necessary during the course of the experiment.
 GeroScience

Toxin exposure transillumination arm carrying an LED, optical tube,

and reflector that provides off axis illumination and high
Potassium cyanide (KCN) was purchased from Acros contrast bright-field imaging (Figure S1). The system
Organics (ThermoFisher, Waltham, MA) and resus- supports any camera that provides Video4Linux driver
pended at 100 mM in 100 mM sodium hydroxide. support, but the current software release and data pre-
Hydrogen cyanide (HCN) exposure was performed as sented here are based on a 1080p (1920 × 1080 pixel)
described (Gallagher and Manoil 2001), with modifica- USB 2.0 web camera with an f/2 4.8-mm S-mount
tions for the 12-well plate format. Briefly, in a fume micro video lens that yields images with a pixel size of
hood, 12-well plate lids were ringed with petroleum 24.4 microns. While the system can work with higher-
jelly from a syringe and a 100-μL drop of the 100 mM resolution cameras (we have successfully used USB 3.0
KCN solution was placed in the interstitial space of the cameras with resolutions as high a 4208 × 3120,
12-well dish near a 100 μL drop of 200 mM hydrochlo- Figure S4), the extra storage required for the images
ric acid. The lid was sealed and the plate tipped to mix generated and decreased light sensitivity have led us to
the two drops liberating HCN gas into the sealed cham- favor standard HD cameras, which provide adequate
ber and placed immediately onto the robot. Strains used resolution to resolve movements in adult worms. Addi-
in this study were provided by the Caenorhabditis Ge- tionally, these 1080p cameras are affordable and widely
netics Center (Madison, WI) which is supported by the available.
NIH Office of Research Infrastructure Programs (P40 The main hardware of the WormBot system consists
OD010440). of an XY-plotter extruded aluminum robotics kit made
by the company MakeBlock (MakeBlock, makeblock.
Health span/behavior measures com, Shenzhen, China). The WormBot system simply
replaces the drawing head of the XY plotter robot with
The WormBot can record brief 30 frame/s movies each a camera and attaches a transillumination arm to the
day for each of the 144 wells in an experiment. To assay camera head, much like a standard compound
worm movement in Fig. 6c, one well from each of the microscope with a very long and very curved arm.
12 plates was randomly selected and the WormBot Due to the size and mass distribution of this arm, a
application plateExplorer was used to analyze 1250 3-kg lead diving weight is placed on the camera head
frames from the selected wells for all 20 days of the to serve as a counter-balance and reduce torsion on the
experiment. The resulting 240 plateExplorer output linear bearings affording free movement of the camera
images were loaded into the Gnu image manipulation head and transillumination arm. Movement of the cam-
program (http://www.gimp.org) with the center portion era head is MXL belt driven and powered by two
of the well containing the worms masked off and the NEMA 17 stepper motors running in an open loop
total number of blue pixels measured, divided by the control configuration by a clone of the Arduino UNO
total number of worms (green objects) observed, microcontroller board and stepper motor control boards
divided by the number of seconds of video used to that are provided in the MakeBlock kit. The
produce the image. This mean pixel/worm/s value was WormBot’s aluminum frame is attached to a 3/8-inch-
then converted to μm2 by photographing a micrometer thick acrylic panel that has been drilled to provide
with the WormBot to determine the actual pixel size. mounting holes for the robot chassis, four extruded
aluminum or 3d printed legs, and an array of 1/8-inch
steel dowel pins. These dowel pins serve to lock in 12
Results standard microtiter plates into position on the optical
table. If using 12-well plates, this allows the system to
Hardware overview analyze 144 individual wells. While this acrylic plate
could be drilled by hand on a milling machine, it can
The WormBot system (http://wormbot.org) is a also be ordered from an online machine shop (Front
benchtop gantry robot that moves a USB microscope Panel Express, Seattle, WA, see Supporting Online
camera over a transparent acrylic optical table to image Information for CAD files for ordering or drilling the
wells containing agar and nematodes in standard tissue panel). For complete video construction tutorials and
culture plates (Fig. 1a). The camera is coupled to a assembly documentation, see http://wormbot.org.
GeroScience

Fig. 1 a Top down diagram of

the WormBot system. a Plates are
held in a 3-by-4 grid by steel
dowel pins on a clear acrylic op-
tical table. The camera head and
transilluminator arm’s motion is
powered by a pair of stepper mo-
tors that are controlled by a mi-
crocontroller board that is con-
nected a Linux workstation via
USB. b Software components of
the WormBot system. A pair of
daemons handles image acquisi-
tion and image alignment. User
interaction is handled through
web server based tools that can be
accessed by any web browser

The WormBot system is connected via two USB cables drives into this workstation as the 1080p cameras generate
(one for the camera and one for the robot) to a single Linux approximately 50 GB per day at full capacity while
workstation. It is advisable to install multiple large hard 4kUHD cameras generate up to 300 GB of data per day.
 GeroScience

Software overview a cgi-based web application that can run on any internet
connected device, like an iPad, smartphone, or web
The WormBot hardware communicates via USB to a set browser (Fig. 2d, cgi-bin/marker?loadedexpID=X).
of software daemons on the workstation (Fig. 1b). The Marker allows the user to manually scroll through the
controller daemon coordinates movement of the robot stack of time-lapse images and click on a worm when it
hardware and captures the image data from the camera. ceases all movements. Marking a death event produces a
The captured images are stored as PNG files and served red circle which the marker application can also embed
to the alignerd that uses an ECC image alignment and export as a time-lapse movie of the entire experi-
algorithm (Evangelidis and Psarakis 2008) built into ment (see supporting online VideoS1). The application
the OpenCV library (Kaehler and Bradski 2016) to align also lists basic experimental statistics and provides the
each time-point for a well to previous timepoints. This data in an OASIS compatible format (Han et al. 2016) to
step is necessary as the spatial resolution of the robot kit be further analyzed. We have created another applica-
stepper motors running in open loop configuration is tion, retrograde (Fig. 2e, /retrograde/retrograde.html),
limited; therefore, the images must be aligned to reduce which attempts to automate the process of identifying
frame to frame noise. the time of death, alleviating the need to manually
The WormBot is conceived to be used by an entire annotate images. Retrograde is a JavaScript application
laboratory group; therefore, we designed the software’s that must be run on a web browser and controlled with a
user interface to work inside of a web browser so that three-button mouse and provides all of the same basic
any internet connected device could be used to control manual scoring functionality as the marker application.
the robot and analyze data on the system from any For data to be fed into the retrograde prediction algo-
location. This also allows data to be shared easily be- rithm, the user goes to the end of the time-lapse image
tween lab members and with collaborators. The work- data and uses the mouse to drag boxes around each
station connected to the WormBot runs a standard worm corpse. The user then initiates the application,
Apache Web Server (The Apache Software Foundation, and the software detects the point at which the objects
Wakefield, MA) that is installed and configured auto- in the boxes began to move and marks them as the death
matically when the WormBot software package is points. Retrograde relies on OpenCV’s image
installed. Experiments are started and stopped on the thresholding and a Canny edge detection algorithm to
WormBot using the scheduler application (Fig. 2a, /cgi- detect worm-sized objects in the user-defined bounding
bin/scheduler). The scheduler page is broken into two boxes and determine when they have moved outside of
parts. The upper red portion lists currently running the noise thresholds that are required due to time-point
experiments that can be individually stopped or clicked to time-point image variability.
on to access the marker application (Fig. 2b). The lower In order to validate the ability of the WormBot to
green portion lists available plate slots on the WormBot detect differences in C. elegans survival, we used
and allows users to add new experiments to the robot’s RNAi to reduce expression of members of the well-
joblist. The WormBot performs two types of data col- described insulin/IGF signaling pathway (Kapahi
lection that can be used simultaneously. Time-lapse: et al. 2017; Finch and Ruvkun 2001) (Fig. 3). Spe-
where the robot takes a single image of a well every cifically, we knocked down expression of the gene
10 min, and Daily-Monitor: where once per day, the encoding the longevity promoting transcription fac-
robot will position the camera over the well and take a tor DAF-16 or the gene encoding the insulin-like
1–5-min video at 30 frames per second. We typically growth factor receptor DAF-2. Results from the
rely on time-lapse data for determining time of death WormBot were compared to similarly treated animals
and daily monitor videos for healthspan, behavior, or that were kept in identical 12-well plates stored next
other movement metrics, but they could be used inter- to the WormBot but were scored by hand using a
changeably. The experimentbrowser application (Fig. worm pick and dissecting microscope. Both in empty
2c, /cgi-bin/experimentbrowser) allows users to see all vector controls and in worms treated daf-16(RNAi),
of the experiments that are currently stored on the server results were not statistically different when scored by
and to back them up or delete them. traditional methods or the WormBot marker applica-
In order to generate survival data, the WormBot tion (Fig. 3). When scored by traditional methods or
system provides two different applications. Marker is with the marker application, daf-2(RNAi) resulted in
GeroScience

Fig. 2 Web applications for the

WormBot system. a The
scheduler application allows
users to start and stop
experiments. b The marker
application is designed for
manual scoring of lifespan time-
lapse data and generation of
annotated time-lapse movie files.
c ExperimentBrowser displays all
of the experiments stored on the
server and allows for data
management. d The retrograde
application allows for manual or
automated lifespan scoring. Users
mark the corpses (green
rectangles) at the end of the time-
lapse data, and the software
automatically determines the
death point (red circles) of the
enclosed worm object

longer lifespan compared with EV controls, but daf- While the marker application yields results that are
2(RNAi) animals were even longer lived when similar to traditional methods, it still requires a degree of
assayed on the WormBot (Fig. 3). manual user input in order to generate a survival curve.
 GeroScience

The retrograde application attempts to remove much of short survival times on the order of hours, such as
this user effort by leveraging computer vision ap- toxicity assays or stress resistance assays. From a prac-
proaches to detect when worms cease movement. In tical drug or genetic screening standpoint, such assays
all cases, the retrograde application was able to correct- can be powerful because the shorter timeframes afford
ly detect the differences between the treatments (Fig. 3 the ability to greatly increase throughput. For example,
bottom right); however, retrograde yielded absolute resistance to high temperature stress has previously been
lifespan values that were significantly greater than those shown to positively correlate with increased longevity,
obtained by traditional hand scoring (Fig. 3, green and exposure to 35 °C will kill wild-type worms within
curves). 10 h while the long lived daf-2(e1370) strain can survive
Since the WormBot was able to detect changes in up to 18 h (Gems et al. 1998).
lifespan due to disruption of a canonical aging pathway, In order to test the ability of the WormBot to perform
we next evaluated the variation in lifespan of wildtype short-term survival studies, we exposed C. elegans to
animals across all of the wells of the robot in large the lethal toxin HCN. Previous work on the susceptibil-
number of individuals (Fig. 4a). Perhaps due to the open ity of worms to the human pathogen Pseudomonas
and symmetrical design of the WormBot, we find that aeruginosa (PAO1) identified bacterial production of
when binning wells across all 12 plates, we detect no HCN as being responsible for the fast paralytic killing
significant difference in mean lifespan in any of the 12 of the worms when exposed to the pathogen, and resis-
well positions (Fig. 4b,type I ANOVA, F = 1.24 p = tance screens identified the prolylhydroxylase EGL-9 as
0.27). Average temperature variation between the conferring resistance to lethal HCN exposure (Gallagher
WormBot wells and the surrounding environment was and Manoil 2001). Subsequent work showed this resis-
less than 0.04 °C with the system running 144 simulta- tance was due to activation of the hypoxia-responsive
neous experiments (Table S2). transcription factor HIF-1 and its downstream targets
One of the key features of robotic image acquisition (Budde and Roth 2010). We placed L4 larval stage
is that it allows for finer temporal resolution survival worms, with loss of function mutations in either egl-9
data than is practical with human acquired datasets. This or hif-1 into sealed 12-well plates in the presence of
feature is useful when performing experiments with 270 μg of HCN and placed them onto the WormBot

Fig. 3 Variation in wild-type survival due to inhibition of insulin plates starting at the L4 larval stage. Plates were scored either by
signaling by RNAi as analyzed on the Wormbot or by traditional traditional hand scoring with a dissecting microscope (black
methods. Top row and bottom row contain the same nine survival curves on top row and bottom left panel), using the marker
curves sorted by RNAi treatment (top) or by analysis type (bottom application from the WormBot system (magenta curves on top
row). X axis is days post hatching. Wild-type worms were fed row and bottom middle panel), or with the automatic feature of the
E. coli expressing double-stranded RNA for the insulin growth Retrograde application (green curves top row and bottom right
factor receptor DAF-2 or its downstream effector, the FOXO panel). * indicates P value < 0.05 using Log-rank test. See
transcription factor DAF-16. All worms were cultured in 12-well Table S3
GeroScience

Fig. 4 Experimental variability in wild-type aging. a 142 wells of rank comparisons and Figure S3 for histogram. b Box and whisker
N2 grown on OP50 with curves from binned plates (Gray lines, plots of mean lifespan for individual well positions (red dots, A1,
see also Figure S2) shown along with mean of all wells (Black A2, etc.) were binned and mean lifespans for binned groups were
line). X axis is days post L4. Fractional day output was used and found not to differ significantly (Table S3)
timepoints are separated by 10 min. See Table S1 for pairwise Log-

(Fig. 5). After 2000 min (200 time points), the data were moribundity in animal models, are unlikely to be
analyzed. While only two egl-9 animals died during the translationally relevant. While the health span concept
observed period, all of the hif-1 animals were dead presently lacks the same rigorous definition as lifespan,
within 1300 min and 75% of wild-type animals were central to this concept is the ability to measure “health”
dead by 2000 min. These data show that the WormBot metrics as the organism’s chronological age increases
can provide high-resolution survival data for even ex- (Kaeberlein 2018). Health metrics that have been shown
tremely toxic and fast-acting compounds such as HCN. to decline during normative aging in C. elegans include
While the maximum and mean lifespan are useful crawling, feeding (pharynx pumping), reproduction,
metrics to assess biological aging, the geroscience com- mechanosensation, and olfaction (Hahm et al. 2015;
munity has recently begun to favor the concept of Russell et al. 2017; Bansal et al. 2015; Ewald et al.
healthspan as a useful metric to evaluate the effective- 2018; Pan et al. 2011; Leinwand et al. 2015; Rollins
ness of anti-aging interventions (Fuellen et al. 2019). et al. 2017). When equipped with a high-resolution
There is growing consensus that interventions that ex- camera, the WormBot is able to detect laid embryos
tend lifespan, but do not also reduce disease burden or (Figure S4), but our standard low-resolution cameras
 GeroScience

Fig. 5 High-resolution survival

data in HCN. Wild-type (N2), elg-
9(sa307), and hif-1(ia04) worms
were placed into a sealed 12-well
plate with an acidified solution of
KCN to liberate HCN gas. X axis
is days post hatching. The
WormBot’s fractional day output
was used to generate curves with
a 10-min (0.007 day) temporal
resolution starting from the time
of HCN exposure at L4 (1-day-
old animals). * indicates P value <
0.05 using Log-rank test

cannot resolve eggs unambiguously. However, sponta- increase and decrease lifespan, and that human annotat-
neous movement, as well as orientation to aversive or ed WormBot data are comparable to those obtained from
attractive odorants, could be quantified. In order to traditional manual lifespan methods, but at substantially
visualize the physical activity of animals, we used the reduced time and effort. In addition to quantifying mor-
dailymonitor feature of the WormBot software to ac- tality under standard conditions, we show that the
quire real-time video (30 frames/s) of the worms each WormBot can easily resolve survival differences under
day for the entire course of the experiment. These .avi toxic conditions such as hydrogen cyanide exposure and
video files can be used with any of the widely available can be used to quantify simple healthspan metrics such
worm tracking packages that have been previously de- as motility.
scribed (Husson et al. 2013). Because implementing We performed a direct comparison of data obtained
many of these trackers requires its own separate instal- on the WormBot to manual lifespan determination by
lation and batch processing procedures, we developed a traditional methods. In all cases, the WormBot data were
simple command line application for the WormBot sys- comparable to manually obtained lifespan data and suc-
tem called plateExplorer that processes all the cessfully discriminated short-lived (daf-16) and long-
dailyMonitor movies for an experiment and runs the lived (daf-2) conditions from wild-type control. The
same worm detection algorithms as the retrograde ap- fully automated retrograde survival analysis yielded
plication to produce a graphical output of the paths survival curves with greater median and maximum
worms take during the course of the recorded movies values than manual annotation of the WormBot data or
(Fig. 6a). These image files can then be mined with traditional hand scoring. We do not yet fully understand
standard image processing applications (see “Materials the reasons for this; however, because the relative dif-
and methods”) to quantify worm movement over the ferences in lifespan are maintained, we believe that the
course of the lifespan experiment (Fig. 6b). As worms automated system is suitable for screening purposes.
age, we find that their spontaneous movement peaks Because the WormBot data are stored as a series of
around days 2–3 and then declines (Fig. 6c), consistent image files, the data can always be re-analyzed using
with similar profiles observed in the WormMotel human annotation. The human annotated WormBot data
(Churgin et al. 2017). yielded lifespan effects that were not significantly dif-
ferent from traditional manual methodology for both
control and daf-16(RNAi) conditions, and slightly lon-
Discussion ger lifespan values for daf-2(RNAi). We speculate that
this may result from the damage caused by repeated
Here, we describe the WormBot, an open-source robot- manual prodding over the long lifespans of daf-2 ani-
ics system for high-throughput lifespan and behavioral mals when the experiment is carried out using traditional
phenotyping in C. elegans. We find that the automated methods. The WormBot design appears to be robust
lifespan scoring features of the WormBot software are against position effects and thermal variation from
able to differentiate between RNAi treatments known to well-to-well and plate-to-plate.
GeroScience

Fig. 6 Lifetime activity data.

Plate exploration was scored
using dailyMonitor movies from
the N2 control data in Fig. 4. a
Output of the WormBot
application plateExplorer used to
determine regions of the well
worms explored during the
recorded time frame. b Samples
of zoomed in traces from days 1,
5, 10, and 15 of the same
experimental well. Worm position
at the start of the movie is marked
in the green channel, the ending
position in the red channel and all
intervening frames (1250 in this
analysis) summed into the blue
channel. c Twelve wells were
randomly sampled from each of
the 12-plates (142 experiments) in
Fig. 4 and each of their daily
traces combined to generate the
composite curve showing the av-
erage number of μm2 explored by
a worm per second as a function
of age. See Supporting Online
information for example
dailyMonitor movies

A major unmet challenge of the current WormBot precise time of death for each animal. While fully auto-
system, as well as other automated lifespan systems for mated survival analysis is a goal we hope to achieve in
C. elegans, is the inability to unambiguously identify the the future, human curation of the time-lapse image data
 GeroScience

currently provides survival curves that closely match WormBot hardware and hardware; Ron Musgrave and the UW
Physics Instrument Shop for advice and assistance when con-
traditional manual lifespan analysis. To do this, a person
structing the original WormBot prototype; and the developers of
scores the time point immediately following the last the various open-source software and hardware platforms on
observed spontaneous movement as a “death” point. which the WormBot relies.
While still requiring some human effort, we have found
that this approach requires less than 20% of the time Funding information This work was supported by the Univer-
sity of Washington Nathan Shock Center of Excellence in the
required for traditional manual lifespan experiments and
Basic Biology of Aging, NIH grant P30AG013280 to MK. JNP
can easily be performed by undergraduate students with and BWB were supported by NIH grant T32AG000057.
significantly less training required.
While the current WormBot platform is geared large- Compliance with ethical standards
ly toward survival analysis, it is a programmable robotic
plate scanning microscope and as such could be used for Conflict of interest JNP, BWB, and MK are shareholders of
GeroTech, Inc., a company seeking to provide a commercial
any assay that could be scored with a low-power bright-
version of the WormBot.
field microscope. For example, a WormBot was recently
built in order to study magnetotaxis in the worm
(Andres Gadea-Vidal, pers commun) and, conceivably,
any assay that involves determining the position of
worms on a plate could be adapted to the WormBot.
Obvious additional applications of the WormBot in- References
clude pathogen sensitivity/resistance and chemosensory
(avoidance/attractant) assays. The open-source nature of Baker M (2016) 1,500 scientists lift the lid on reproducibility.
the WormBot software and the flexibility of the open- Nature 533(7604):452–454
source hardware design will allow for future enhance- Bansal A, Zhu LJ, Yen K, Tissenbaum HA (2015) Uncoupling
lifespan and healthspan in Caenorhabditis elegans longevity
ments and upgrades of an already useful system. Thus, mutants. Proc Natl Acad Sci U S A 112(3):E277–E286
we anticipate that the WormBot has the potential to Banse SA, Blue BW, Robinson KJ, Jarrett CM, Phillips PC
accelerate the pace of C. elegans research in a variety (2019a) The Stress-Chip: a microfluidic platform for stress
of contexts. analysis in Caenorhabditis elegans. PLoS One 14(5):
e0216283
Finally, science is in the midst of a reproducibility
Banse SA, et al. (2019b) Automated lifespan determination across
crisis (Prinz et al. 2011; Baker 2016). The causes of this Caenorhabditis strains and species reveals assay-specific ef-
crisis are multifactorial, but range from simple statistical fects of chemical interventions. bioRxiv, p. 757302.
errors, like inadequate sample size, to outright data Bitto A, et al. (2015) Biochemical Genetic Pathways that modulate
fabrication. Systems like the WormBot that automatical- aging in multiple species. Cold Spring Harb Perspect Med
5(11)
ly create archival time-stamped digital observations that
Brenner S (1974) The genetics of Caenorhabditis elegans.
can easily be blinded, shared, and reexamined are an Genetics 77(1):71–94
important tool in addressing this crisis. In this aspect, Budde MW, Roth MB (2010) Hydrogen sulfide increases
automation itself is also beneficial, as it is designed to be hypoxia-inducible factor-1 activity independently of von
carried out by machines and is therefore necessarily Hippel-Lindau tumor suppressor-1 in C. elegans. Mol Biol
Cell 21(1):212–217
more highly controlled owing to the limitations of these Burnaevskiy N, et al. (2018) Reactivation of RNA metabolism
mechanical systems. Automation also allows for larger underlies somatic restoration after adult reproductive dia-
sample sizes, decreased experimenter bias, and fewer pause in C. elegans. Elife 7
human errors during data collection. All of these factors Churgin MA, et al. (2017) Longitudinal imaging of
Caenorhabditis elegans in a microfabricated device reveals
point to laboratory automation tools like the WormBot
variation in behavioral decline during aging. Elife 6
revolutionizing the scale, rigor, and scope of science in D'Ausilio A (2012) Arduino: a low-cost multipurpose lab equip-
the future. ment. Behav Res Methods 44(2):305–313
Evangelidis GD, Psarakis EZ (2008) Parametric image alignment
Acknowledgments We would like to thank Nick Terzopoulos using enhanced correlation coefficient maximization. IEEE
for media preparation; Shane Rea, Andres Vidal-Gadea, George Trans Pattern Anal Mach Intell 30(10):1858–1865
Sutphin, Christine Queitsch, and Makoto Horikawa for beta test- Ewald CY, Castillo-Quan JI, Blackwell TK (2018) Untangling
ing the WormBot in their laboratories; Matt Crane and other longevity, dauer, and healthspan in Caenorhabditis elegans
members of the Kaeberlein lab for discussions and testing of the insulin/IGF-1-signalling. Gerontology 64(1):96–104
GeroScience

Finch CE, Ruvkun G (2001) The genetics of aging. Annu Rev Proteasomal regulation of the hypoxic response modulates
Genomics Hum Genet 2:435–462 aging in C. elegans. Science 324(5931):1196–1198
Fuellen G, Jansen L, Cohen AA, Luyten W, Gogol M, Simm A, Pan CL, Peng CY, Chen CH, McIntire S (2011) Genetic analysis
Saul N, Cirulli F, Berry A, Antal P, Köhling R, Wouters B, of age-dependent defects of the Caenorhabditis elegans touch
Möller S (2019) Health and aging: unifying concepts, scores, receptor neurons. Proc Natl Acad Sci U S A 108(22):9274–
biomarkers and pathways. Aging Dis 10(4):883–900 9279
Gallagher LA, Manoil C (2001) Pseudomonas aeruginosa PAO1 Prinz F, Schlange T, Asadullah K (2011) Believe it or not: how
kills Caenorhabditis elegans by cyanide poisoning. J much can we rely on published data on potential drug targets?
Bacteriol 183(21):6207–6214 Nat Rev Drug Discov 10(9):712
Gems D et al (1998) Two pleiotropic classes of daf-2 mutation Puckering T et al (2017) Automated wormscan. F1000 Res 6:192
affect larval arrest, adult behavior, reproduction and longev- Rera M, Vallot C, Lefrancois C (2018) The Smurf transition: new
ity in Caenorhabditis elegans. Genetics 150(1):129–155 insights on ageing from end-of-life studies in animal models.
Gill MS, Olsen A, Sampayo JN, Lithgow GJ (2003) An automated Curr Opin Oncol 30(1):38–44
high-throughput assay for survival of the nematode Rollins JA et al (2017) Assessing health span in Caenorhabditis
Caenorhabditis elegans. Free Radic Biol Med 35(6):558–565 elegans: lessons from short-lived mutants. J Gerontol A Biol
Hahm JH et al (2015) C. elegans maximum velocity correlates Sci Med Sci 72(4):473–480
with healthspan and is maintained in worms with an insulin Rual JF, Ceron J, Koreth J, Hao T, Nicot AS, Hirozane-Kishikawa
receptor mutation. Nat Commun 6:8919 T, Vandenhaute J, Orkin SH, Hill DE, van den Heuvel S,
Han SK, Lee D, Lee H, Kim D, Son HG, Yang JS, Lee SV, Kim S Vidal M (2004) Toward improving Caenorhabditis elegans
(2016) OASIS 2: online application for survival analysis 2 phenome mapping with an ORFeome-based RNAi library.
with features for the analysis of maximal lifespan and Genome Res 14(10B):2162–2168
healthspan in aging research. Oncotarget 7(35):56147–56152 Russell JC, et al. (2019) Electrophysiological measures of aging
Husson SJ, et al. (2013) Keeping track of worm trackers. pharynx function in C. elegans reveal enhanced organ func-
WormBook, p. 1-17. tionality in older, long-lived mutants. J Gerontol A Biol Sci
Kaeberlein M (2018) How healthy is the healthspan concept? Med Sci 74(8):1173–1179
Geroscience 40(4):361–364 Saberi-Bosari S, Huayta J, San-Miguel A (2018) A microfluidic
Kaehler A, Bradski G (2016) Learning OpenCV 3: computer platform for lifelong high-resolution and high throughput
vision in C++ with the OpenCV library. O’Reilly Media, Inc. imaging of subtle aging phenotypes in C. elegans. Lab
Kamath RS et al (2001) Effectiveness of specific RNA-mediated Chip 18(20):3090–3100
interference through ingested double-stranded RNA in Schulenburg H, Felix MA (2017) The natural biotic environment
Caenorhabditis elegans. Genome Biol. 2:1–10 of Caenorhabditis elegans. Genetics 206(1):55–86
Kapahi P, Kaeberlein M, Hansen M (2017) Dietary restriction and Stroustrup N, Ulmschneider BE, Nash ZM, López-Moyado IF,
lifespan: lessons from invertebrate models. Ageing Res Rev Apfeld J, Fontana W (2013) The Caenorhabditis elegans
39:3–14 Lifespan Machine. Nat Methods 10(7):665–670
Kenyon C (2011) The first long-lived mutants: discovery of the Uno M, Nishida E (2016) Lifespan-regulating genes in C. elegans.
insulin/IGF-1 pathway for ageing. Philos Trans R Soc Lond NPJ Aging Mech Dis 2:16010
B Biol Sci 366(1561):9–16 Van Raamsdonk JM, Hekimi S (2011) FUdR causes a twofold
Laranjeiro R, et al. (2019) Swim exercise in C. elegans extends increase in the lifespan of the mitochondrial mutant gas-1.
neuromuscular and intestinal healthspan, enhances learning Mech Ageing Dev 132(10):519–521
ability, and protects against neurodegeneration. bioRxiv p. Wang H, Zhao Y, Zhang Z (2019) Age-dependent effects of
633776. floxuridine (FUdR) on senescent pathology and mortality in
Laurie B, Laurie P 2003) Apache: the definitive guide. O’Reilly the nematode Caenorhabditis elegans. Biochem Biophys Res
Media, Inc. Commun 509(3):694–699
Leinwand SG et al (2015) Circuit mechanisms encoding odors and Xian B, Shen J, Chen W, Sun N, Qiao N, Jiang D, Yu T, Men Y,
driving aging-associated behavioral declines in Han Z, Pang Y, Kaeberlein M, Huang Y, Han JD (2013)
Caenorhabditis elegans. Elife 4:e10181 WormFarm: a quantitative control and measurement device
Leiser SF et al (2013) Life-span extension from hypoxia in toward automated Caenorhabditis elegans aging analysis.
Caenorhabditis elegans requires both HIF-1 and DAF-16 Aging Cell 12(3):398–409
and is antagonized by SKN-1. J Gerontol A Biol Sci Med
Sci 68(10):1135–1144 Publisher’s note Springer Nature remains neutral with regard to
Mehta R, Steinkraus KA, Sutphin GL, Ramos FJ, Shamieh LS, jurisdictional claims in published maps and institutional
Huh A, Davis C, Chandler-Brown D, Kaeberlein M (2009) affiliations.