- Methodology article
- Open Access
A multi-animal tracker for studying complex behaviors
© Zaslaver et al. 2017
- Received: 9 February 2017
- Accepted: 6 March 2017
- Published: 6 April 2017
Animals exhibit astonishingly complex behaviors. Studying the subtle features of these behaviors requires quantitative, high-throughput, and accurate systems that can cope with the often rich perplexing data.
Here, we present a Multi-Animal Tracker (MAT) that provides a user-friendly, end-to-end solution for imaging, tracking, and analyzing complex behaviors of multiple animals simultaneously. At the core of the tracker is a machine learning algorithm that provides immense flexibility to image various animals (e.g., worms, flies, zebrafish, etc.) under different experimental setups and conditions. Focusing on C. elegans worms, we demonstrate the vast advantages of using this MAT in studying complex behaviors. Beginning with chemotaxis, we show that approximately 100 animals can be tracked simultaneously, providing rich behavioral data. Interestingly, we reveal that worms’ directional changes are biased, rather than random – a strategy that significantly enhances chemotaxis performance. Next, we show that worms can integrate environmental information and that directional changes mediate the enhanced chemotaxis towards richer environments. Finally, offering high-throughput and accurate tracking, we show that the system is highly suitable for longitudinal studies of aging- and proteotoxicity-associated locomotion deficits, enabling large-scale drug and genetic screens.
Together, our tracker provides a powerful and simple system to study complex behaviors in a quantitative, high-throughput, and accurate manner.
- Multi-animal tracking
- Image analyses
- C. elegans
Animal behavior is rich and complex . It spans a tremendously wide range of phenotypes such as sleep, mating, food search, and fighting. In its early days, this fascinating field was dominated by classical field studies involving visual inspection and hand-written documentation. The discipline has considerably evolved and nowadays is known as “computational ethology” . Technology advances are now replacing the laborious manual work and scientists employ sophisticated computational approaches to generate accurate quantitative understanding of the detail-rich complex behaviors.
In the last few decades, significant advances in genetic- and neural-related techniques have been achieved, and animal model systems have become popular to study complex behaviors under lab-controlled environments. These animal models range from simple invertebrates, such as worms and flies, to mammalian models, including mice and monkeys .
Of particular interest is the small roundworm C. elegans . Among the many advantages that this nematode offers are its short generation time, the easy and inexpensive handling, and the availability of its fully-reconstructed neural system consisting of only 302 neurons . C. elegans animals also show remarkably complex behaviors, including mating , roaming and foraging , lethargus , and taxis towards various stimuli such as preferred temperature , magnetic fields , and food [10–12]. Over the years, large genetic screens have identified a vast set of mutations in genes that modulate these behavioral outputs, enabling a mechanistic insight into these complex behaviors [13, 14].
In addition, a growing interest is directed towards quantitative measures of aging and proteotoxicity-associated physiological decline. C. elegans is an appealing model organism to study these processes as many human neurodegenerative diseases that result from aggregation of proteotoxic proteins can be recapitulated in the worms’ muscular and nervous systems . These diseases are often associated with deteriorating locomotion abilities, culminating eventually in complete paralysis. It therefore becomes an important, yet extremely challenging, task to quantify minute locomotive changes in a high-throughput manner.
To study such complex behaviors, numerous tracking systems that image freely-moving animals have been developed. These systems extract various fine locomotion features, such as animal posture, undulation properties, speed, and attraction or repulsion towards or away from stimulants [16–23]. To support high-throughput studies, some of these systems were designed to track multiple worms at a time. This sort of tracking is particularly challenging as accurate extraction of genuine worm entities from ‘noisy’ background is difficult. Moreover, frequent animal collisions preclude extraction of long individual tracks.
Here, we present a novel Multi-Animal Tracker (MAT), an end-to-end, user-friendly solution for imaging, tracking, and analyzing complex behaviors. At the heart of the tracker is a machine learning algorithm; thus, the tracker makes no prior assumptions regarding the size or the shape of the tracked animals and hence is compatible in studying virtually any animal of interest. We also included built-in functions that allow non-programmers to analyze and visualize the immense data. We demonstrate its high-compatibility tracking various animals, and its powerful abilities studying complex behaviors in C. elegans animals in an accurate and high-throughput fashion.
In the following sections, we demonstrate the vast uses of the novel MAT to studying complex behaviors. We begin by describing the packages included in the software suite and then provide experimental evidence, including novel gained insights, for the extensive usage of this MAT in studying chemotaxis and sensory integration, as well as aging- and proteotoxicity-associated locomotion decline.
The new MAT provides an end-to-end solution for tracking animals – from video acquisition and track identification to advanced analyses
The ‘AnimalsRecorder’ module allows capturing of time-lapse movies in a variety of video formats. This package circumvents the need to rely on commercial software that often interfaces with exclusive cameras and setups. The user can easily modulate various imaging parameters such as recording duration, number of frames, and exposure time.
The main module in the software is ‘AnimalsTracker’. This module uses a machine learning algorithm to identify individual animals and extracts their continuous tracks over long temporal recordings, typically consisting thousands of frames. The Graphical User Interface (GUI) in this module displays several panels that allow scrolling between the different frames, and visually comparing the original image to a filtered one (Fig. 1b). This feature provides a simple and convenient mean to inspect and evaluate how well the tracker extracts the observed entities from their background in each frame. The user can quickly adjust the extraction sensitivity between different frames. Once satisfied with the entities extraction, the user is asked to mark individual animals in order to train the machine learning algorithm to correctly identify them.
The additional module, ‘AnalysesPlugins’, contains several useful built-in analyses functions; some of these functions are basic analysis tools, but others offer advanced toolkits that allow non-specialist programmers to sophisticatedly analyze the complex behavioral paradigms. These tools are further described below.
A machine learning algorithm extracts accurate tracks of multiple animals over long time periods
Additional file 2: Movie S1. The Multi-Animal Tracker can track > 100 worms at a time. (AVI 62803 kb)
Additional file 3: Movie S2. The Multi-Animal Tracker can track multiple flies at a time. (MP4 11655 kb)
Additional file 4: Movie S3. The Multi-Animal Tracker can track multiple zebrafish at a time. (MP4 39896 kb)
Additional file 5: Movie S4. The Multi-Animal Tracker can track multiple mice at a time. The machine learning algorithm, together with the Kalman predictor, filter out possible noise, such as user intervention, during the experiment. (MP4 18744kb)
A major obstacle in maintaining long tracks of individual animals is the frequent collisions between them. To overcome this, we implemented a simple variation of the Kalman filter , which predicts the future position of the colliding animals based on previously observed motility features (e.g., velocity, acceleration; see Methods for implementation). Implementing such a predictor significantly improves detection following animal collision; for example, implementing the predictor reduces the cases where none of the animal trajectories could be resolved following collision to approximately 2% only (whereas as many as 46% of the collision events could not be resolved if not implementing the predictor; Additional file 6: Figure S1b). Moreover, the prediction also significantly extends the tracks’ length as it compensates in cases where animal segmentation fails to identify a genuine animal (Fig. 2c, Additional file 7: Figure S2). Overall, this allows the tracker to maintain identification of individual animals over long imaging time periods (Additional file 6: Figure S1c, Additional file 7: Figure S2, Additional file 8: Movie S5, Additional file 9: Movie S6, Additional file 10: Movie S7).
Additional file 8: Movie S5. An exemplary movie extracting tracks of swimming zebrafish that does not implement neither the machine learning (ML) algorithms nor the Kalman predictor. Without using these features, trajectory extraction is poor despite the fact that animals are easily detected due to a high contrast with the background. (MP4 39899 kb)
Additional file 9: Movie S6. An exemplary movie extracting tracks of swimming zebrafish analyzed by implementing ML algorithms only (the Kalman predictor is not used in these analyses). While segmentation and trajectory extraction is improved when compared to analyses that use neither of them, the results still fall short if compared to analyses that use both (the ML and Kalman predictor). (MP4 33474 kb)
Additional file 10: Movie S7. An exemplary movie extracting tracks of swimming zebrafish analyzed by implementing both ML algorithms and the Kalman predictor. Using both features significantly improves segmentation and trajectory extraction. (MP4 35674 kb)
Clearly, collision events become more probable as animal density increases, thereby limiting the time of efficient tracking. In our experimental setups (circular, 50 mm diameter field of view), we found a negative linear correlation (ϱ = −0.7, P < 0.001) between the number of worms on the experimental plate and the average duration of tracking (Fig. 2d). In addition, the implemented Kalman-type predictor aids ignoring potential interference during the experiment. For example, the software discounts transient user interventions, such as manual handling of mice during the course of the experiment, and proceeds by focusing on mice tracks only (Additional file 5: Movie S4).
The new tracker is highly suitable for studying large-scale dynamics of chemotaxis behavior
To demonstrate the many advantages and the high suitability of the MAT to study complex behaviors, we focused on C. elegans nematodes. One of the fascinating complex behaviors that these worms exhibit is chemotaxis, in which animals navigate up a chemical gradient towards the chemical source. Very often, chemotaxis assays are quantified by an attraction index that reflects the end position of the animal, disregarding its intricate trajectories throughout the course of the chemotactic behavior . High-throughput analyses of such trajectories with their relation to the chemical source are imperative to underpinning the mechanisms by which animals navigate through gradients. Our tracking system is particularly suitable for studying complex chemotactic behaviors – it extracts long behavioral tracks of dozens of animals at a time, and positions these trajectories relative to the stimulus source that is set as a reference point.
Presenting the rich trajectory data for multiple worms may become cumbersome given the enormous amount of data obtained by the tracking of multiple worms in a spatiotemporal manner over long time periods. To overcome this, we developed the Attraction Field (AF) view, which captures many of the intricate parameters extracted throughout the course of the experiment in a single representation (Fig. 3d, f). For this purpose, we binned the entire field of view into squares (for our resolution we used 35 × 25 squares), and for each square we plotted both the average direction of the worms (arrows) and the overall occupancy over time (color coded). This AF representation provides a quantitative spatial measure of the chemotactic response; for example, higher chemoattractant levels yield more direct trajectories towards the chemoattractant from wider regions in the experimental field and higher worm occupancy around the target region compared to lower levels of the chemoattractant (Fig. 3d–f). When combining the AF views with the ROI kinetics representation (Fig. 3f, g), these data culminate in a rich quantitative spatiotemporal representation of the chemotactic response as measured simultaneously for multiple worms over long time periods. The AF and ROI analyses are supplied as built-in functions in our software analysis suite (Additional file 1).
Worms incorporate directional mechanisms to improve chemotaxis efficiency
Surprisingly, the above experiments provided new insights regarding chemotaxis strategies. During chemotaxis, worms use two main strategies, namely (1) a biased random walk (klinokinesis) where worms increase or decrease turning rates depending on whether they crawl down (dc/dt < 0) or up (dc/dt > 0) a chemical gradient, respectively ; or (2) a weathervane movement (also known as klinotaxis ), where worms moving perpendicular to the gradient reorient to move towards the gradient. Interestingly, Shimomura et al.  showed that worms’ klinokinesis is not entirely composed of uniform reorientation events, and worms that are off-course just before the pirouette tend to compensate with a larger change in their direction following the pirouette such that they are better oriented towards the source.
To understand the functional significance of this reorientation strategy we simulated chemotaxis of worms adopting different criteria for choosing directional changes following a pirouette (sampled from our experimental data, see Methods for details). Interestingly, chemotaxis performance of worms simulated according to the strategy detailed above was superior over other possible strategies as the worms reached the target significantly faster (by approximately 50%, P < 10–5, Wilcoxon rank-sum test, Fig. 5c). These findings provide a novel understanding of animal chemotaxis strategies, and further underscore the significant roles of pirouettes in mediating efficient chemotaxis.
Worms integrate environmental cues and enhance chemotaxis towards richer environments
Integrating information extracted from the environment is an important ability shared by many animals [28, 29]. In particular, during food search, animals often face confusing cues and attending to the most reliable of them is likely to improve their chance of quickly reaching food sources . For example, multiple cues originating from a source may be rendered as a more reliable food source than a source emitting a single cue.
We next asked which features in the chemotaxis behavior underlie this enhanced attraction. For this, we analyzed the directness of the worms towards the different stimuli by quantifying the projection of their steps on the vector that connects the worm and the chemoattractant (see Methods). We found that worms presented with a mixture of cues move more directly towards the source than worms presented with each cue alone (P < 0.006, Fig. 6c). Together, these data demonstrate that worms can integrate environmental information and attend to more than one cue at a time. Furthermore, this integration is translated to behavioral outputs manifested primarily by the ability to better reorient the trajectory following a pirouette (Additional file 12: Figure S4).
Quantitative analyses of worm motility during aging and neurodegenerative-linked proteotoxicity progression
Finally, we demonstrate the powerful advantages of using this novel MAT in studying fine locomotion deficits associated with aging and neurodegenerative-related diseases. During these processes, animal locomotion deteriorates, culminating eventually in complete paralysis [31, 32]. It therefore becomes extremely challenging to quantify minute locomotion changes during these critical periods. Moreover, due to large locomotion variability within worm populations it is critical to collect large data sets in order to obtain reliable and reproducible results . The MAT system overcomes these hurdles owing to its high-throughput and accurate tracking.
We next employed our tracker to study aging-associated motility impairment following external manipulations such as those used in large-scale RNAi or chemical screens. In C. elegans, knocking down the sole insulin/IGF-1 signaling (IIS) receptor, daf-2, leads to long-lived, stress-resistant animals that maintain robust proteostasis [34–36]. Indeed, worms treated with daf-2 RNAi showed a significantly slowed deterioration in motility starting from day 6 and on (Fig. 7c). Similarly, we utilized our MAT system to quantitatively measure proteotoxicity-associated locomotion decline. For this, we tracked motility of worms expressing the aggregative peptide composed of 35 glutamine repeats (polyQ35) in their body wall muscles, which leads to motility impairment in a time-dependent manner . As velocity declined with age, animals treated with daf-2 RNAi exhibited a significantly slower decline, indicating the protective role of inhibiting IIS in proteotoxicity (Fig. 7d). In particular, the MAT system reliably detected minute motility differences (fractions of mm/sec) that ultimately reflect significant differential aging and proteotoxicity paces. Together, these results demonstrate that our MAT system is highly suitable for studying proteotoxicity- and age-related maladies in large worm populations, enabling fast, high-throughput, and accurate screens of mutant animals or RNAi libraries.
Herein, we present a novel MAT that allows the tracking of multiple animals simultaneously over long time periods. Importantly, this user-friendly system offers an end-to-end solution – from acquisition and recording movies, through tracking algorithms and advanced built-in analyses functions, to displaying the rich data in comprehensive and informative plots.
A major advantage of our system over existing multi-worm trackers is that we implemented a machine learning algorithm coupled with a Kalman-type predictor . These features turn the tracker into a multi-purpose system compatible with tracking various animals differing in size, appearance, velocities, and other activity parameters. Indeed, we show that the software can faithfully track nematodes, flies, zebrafish, and mice (Fig. 1, Additional file 2: Movie S1, Additional file 3: Movie S2, Additional file 4: Movie S3, Additional file 5: Movie S4). Moreover, the tracker is indifferent to the acquisition systems (e.g., a wide variety of optical setups can be used), and it does not require adjusting parameters for particular environments or experimental settings (e.g., resolution, brightness, hue, etc.).
Combining machine learning algorithms together with the predictor also supports accurate detection and segmentation of animal entities, specifically so when background environments are ‘noisy’. For example, long temporal experiments are challenging as the environment is dynamic and often changes over the course of the experiment (e.g., water condensation on the plate lid). Our system reaches approximately 90% detection accuracy (precision and recall) when tracking around 100 worms simultaneously (Fig. 2 and Additional file 7: Figure S2). Furthermore, high-throughput studies that track multiple animals at a time pose an additional challenge – the greater the number of animals tracked, the higher the chances that they will collide, therefore precluding the possibility to track individuals over long time periods. However, implementing the predictor in our system significantly aids in resolving such collision events (Fig. 2c and Additional file 7: Figure S2).
Machine learning approaches had been previously applied to tracking single worms (for example, see [38, 39]). Our approach, however, is slightly different and is particular suitable for tracking multiple animals in long-temporal experiments in which noise accumulates over time. We allow the user to first set a fixed thresholding parameter (as part of the training process), and then the user chooses segmented entities corresponding to genuine animals. Based on the features of these entities, a classifier is learned and is subsequently used to classify other segmented entities throughout the movie. Indeed, using several clicks on animal entities, the user ‘trains’ the algorithm to tell apart genuine animal entities from possible background noise, thereby significantly reducing erroneously detected animals. Importantly, this learning step typically takes 1 minute, after which the algorithm automatically extracts animal entities from thousands of frames. If the same settings are used for subsequent experiments then the same training set can be used for their analyses. Together, the machine learning algorithm and the predictor, the broad possible usage, and the ability to simultaneously track multiple animals over long time periods make this MAT into a unique and powerful system for studying animal behavior.
Various other machine-vision algorithms that support high-throughput tracking of many individual animals simultaneously are also available (i.e., in flies ). Similarly to our tracker, these systems automatically track large groups of unmarked animals. Furthermore, machine learning methods were developed to support automatic behavioral classifiers based on animal trajectories alone, which subsequently infer higher order behavioral outputs such as social and individual behavior . While our tracking system, together with its built-in analytic functions, is primarily geared to studying complex spatial navigation, it can also be used to study other complex behaviors. It will therefore be interesting to combine such machine learning post-acquisition methods to extend the repertoire of complex behaviors to be studied.
Interestingly, our initial experience with the MAT system already yielded novel insights into the intricate chemotaxis behavior. In a seminal study, Shimomura et al.  demonstrated the critical role of pirouettes (a bout of successive turns) in chemotaxis, where worms entering a pirouette with an off-course angle tend to exit the pirouette better oriented towards the center of the gradient. Our tracking analyses confirmed these results and added an additional important piece to this fascinating strategy – worms that enter the pirouette on-course oriented towards the target tend to maintain this on-course trajectory following the pirouette. Together, these strategies greatly enhance an animal’s navigation abilities, enabling it to reach a target source significantly faster (as we also corroborated through simulations, Fig. 5c).
In addition, we used our MAT system to show that C. elegans animals can integrate environmental cues (Fig. 6). Such integration is an important feature to predict food abundance and quality . Interestingly, we found that this sensory integration is manifested in the directional changes in the pirouette events – animals crawling towards a richer target are more likely to exit the pirouette at an angle that is more directed towards the target (Additional file 12: Figure S4). In our experiment, we used two cues, DA and IAA, sensed primarily by AWA and AWC sensory neurons [42–45]. These two sensory neurons synapse on mutual downstream interneurons (AIA, AIY) that, along with additional inter and command neurons, control worm locomotion [46–48], thus presumably serving the point where sensory information is integrated before dictating locomotion outputs.
We also demonstrated that the MAT system allows the study of aging and proteotoxicity-associated diseases. These processes are characterized by gradual locomotion decline, and quantitative accurate measurements are required to study this slow progression. Our system accurately captures such minute locomotion changes (Fig. 7). Moreover, aging and proteotoxicity effects widely vary from animal to animal, requiring assaying large population cohorts. The novel MAT system is particularly suitable to this end as it allows accurate tracking of animal locomotion in a high-throughput manner, and can therefore serve as a powerful system for extensive laborious screens of mutants or chemical and RNAi libraries.
Herein, we present a novel MAT that provides accurate high-throughput analyses of complex behaviors. Importantly, this user-friendly system is easy to operate and does not require prior programming skills. With a wide range of possible uses and the compatibility of studying various animal models, this MAT will serve as an important system for elucidating novel principles underlying complex behaviors.
The software suite, together with its freely-available code, can be downloaded through github (https://github.com/itskov/MultiAnimalTrackerSuite). The downloaded files also include a detailed user manual (with exemplar analyzed movies) for the installation and use of the different packages in the software.
Machine learning algorithm for the identification of behaving animals
Segmentation and feature selection
To minimize tracking of erroneous objects, we utilized a supervised machine learning approach. Prior to tracking, the GUI prompts the user to mark few samples of genuine animal entities (these will form the training set). The GUI presents the user with a frame from the experiment video next to the same frame filtered with the Laplacian of Gaussian (LoG) filter [49, 50]. That is, the image is convolved with a matrix M LoG . In this filter, the standard deviation of the Gaussian, σ LoG , is used to control the preferred size of the entities to be highlighted in the image. Thus, the user is presented with a scroll bar with which it can control the size of the σ LoG . Basing the filter on the standard deviation of the Gaussian, σ LoG , allows the system to mark and track entities drawn from different imaging settings, and/or different model organisms. The chosen σ LoG are subsequently used to segment the entities that will be then classified as ‘animals’, or else considered as background, and thus discarded.
Next, the tracker extracts features from the entities selected by the user. We designed the tracker to extract the smallest subset of features to enable good performance with a wide range of different model organisms and imaging settings. These features include the area of the entity, and the mean, median, minimum, and maximum of the entity's pixels intensity. While this small set of general features is sufficient to accurately segment animal entities (all the results presented within this paper use these features only), additional features may be included at the users discretion, which will support further flexibility and efficient segmentation under various circumstances. The open source code can be easily accessed and modified as necessary; the identity of the extracted features is found in the ‘extractFeatures’ method, which is in the tracker object.
The algorithm for calling or rejecting animal entities
Since users choose only ‘positive’ entities, and they are not asked to mark entities which are not animals, a traditional machine learning approach (as a linear separator, decision tree, etc.) could not be used. Alternatively, we found the following described method to be superior to regression-based classification methods:
Where T(β) is the value of the Mahalanobis distance, which is larger than (1 – β) × m samples in the training set. Throughout all our experiments, we set β to be 0.96; however, this value can be easily adjusted to control type I and type II errors in the classification.
We used a simple tracking algorithm which matches spatially closest entities between consecutive frames (the only entities considered are entities that were labeled as animals in the previous step). We incorporated a simple variation of the Kalman filter  to predict the position of animals that are obscured by other entities as, for example, in the case of collisions or in ambiguity in matching entities between consecutive frames. Prior to matching an active track to a new entity found in a new frame, the tracker calculates the track's predicted position based on the animal's recent velocities and accelerations. In case the tracker fails to match a single entity to an active track, it will match the track to its predicted position. The tracker will maintain tracking when the animal reappears, or will discard its prediction if no animal entity was found to match the track during the three subsequent frames.
Evaluating tracker accuracy
Precision: out of X entities classified as worms by the software, how many were actual worms? Based on the machine learning algorithm alone, we find this measure to reach nearly 90% accuracy. In practice, the prediction feature adds on top of this to gain over 90% precision.
Recall: out of all worms genuinely found on the plate (based on the ‘gold standard’ set), how many are indeed classified by the tracker as worms? Figure 2b shows that the tracker reaches over 85% accuracy even when using only 50 worms for the training set.
Fscore is the harmonic average of the two parameters (Precision and Recall).
In addition, we assessed the tracker performance when disabling the machine learning and the Kalman-type predictor. We find that enabling these features significantly improves animal segmentation and tracking (Fig. 2c and Additional file 6: Figure S1 and Additional file 7: Figure S2). In particular, these features are important even when analyzing movies that initially seem to be ‘easy-to-analyze’ with high contrast between animals and background (see analyses in Additional file 7: Figure S2 and Additional file 8: Movie S5, Additional file 9: Movie S6, Additional file 10: Movie S7).
Evaluating worm resolution following collisions
To estimate the precision by which our system correctly resolves worms after collision, we have taken a similar approach as described above to evaluate tracker accuracy. We first defined collisions as events in which worms have come to a close proximity (less than 15 pixels which correspond to approximately 0.4 mm in our setup), and manually matched worm identity before and after the collision event. This is to be used as our ‘gold standard’ measure. We then ran our tracker in two modes, with and without the predictor. The results of the accuracy of predicting the correct tracks following collisions are summarized in Additional file 6: Figure S1b. As evident, applying the predictor significantly improved detection of worms after collision as only approximately 2% of the cases were not resolved (Resolved 0), whereas without implementing our predictor, 46% of the collisions were not resolved. Furthermore, in 44% of the collision cases, one worm was correctly resolved (as opposed to 16% without the predictor) and in 55% of the cases both worms were correctly resolved (as opposed to 38% without the predictor).
To get a large synchronized population of young adult animals (N2, WT strain), we bleached gravid worms and plated approximately 1000 eggs on a 90-mm standard NGM plate pre-seeded with 500 μL E. coli OP 50 culture. These worms were assayed 3 days later when they reached young adulthood (YA). Before the experiment, the YA worms were rinsed off the growth plates and washed three times in chemotaxis buffer (1 mM CaCl2, 1 mM MgSO4, 5 mM K3O4P, pH 6.0). Chemotaxis assays were performed on Chemotaxis plates, which include the same ingredients as the chemotaxis buffer with the addition of 2% agar. Importantly, worms were grown at 20 °C and behavioral assays were also performed in a temperature-controlled room at 20 °C.
We marked an equilateral triangle on the plate’s lid (90-mm round plates) with an edge length of 3 cm. We used agar chunks soaked (15 μL) with the chemoattractants of choice and placed them on two of the triangle vertices (Fig. 3a). On the third vertex, we placed a 5-μL Chemotaxis Buffer drop of washed worms (we first estimated worms’ concentration in the pellet following the last wash to plate a desired number of worms). Chemotaxis assays were then imaged using a Photometrics Micropublisher 5 MB camera, using Olympus SZ61 binocular equipped with a 0.5× lens. To acquire movies we used our own in-house imaging software that is freely available with this MAT and which uses MATLAB’s image acquisition toolbox. Movies were acquired at a rate of one frame per second.
Calculating worms’ directness in relation to the chemoattractant target
In the integration experiments we used only projections of the worms that were on their way towards the chemoattractant for the first time (and ignored revisiting worms).
Identification and analyses of pirouettes
To study the role of pirouettes during chemotaxis, we used the same notations as described previously by Shimomura et al. ; we defined bearing (B) as the angle between the worm’s velocity vector and the spatial vector between the worm’s position and the peak of the chemoattractant. We used BBefore and BAfter as the bearing immediately before and immediately after a pirouette event, respectively, and ΔB as BBefore – BAfter.
We defined sharp turns as succeeding movement vectors with an angle of > 100° between them, and used the definition suggested by Shimomura et al.  for a pirouette, which is a bout of sharp turns. Following the observation that run distribution can be described by the sum of two exponents , we chose the minimal size of a pirouette to be Tcrit = 5 sec. Any run shorter than this will be considered to be a component of a pirouette. We defined Bbefore as the average bearing for three consecutive steps prior to the pirouette, and similarly we defined Bafter to be the average bearing of three consecutive steps after the pirouette.
Integration of environmental cues
We first looked for two chemoattractants that attract the worm in the same manner. To do so, we performed an array of chemotaxis experiments with varying chemicals and concentrations and examined motility parameters such as chemotaxis index dynamics, probability for a pirouette, and lengths of runs, etc. We chose two chemoattractants that showed the same effect on the worm’s chemotaxis: 0.75 × 10−5 DA, and 0.5 × 10−4 IAA. We then created a mixture of the two chemicals such that each chemical was diluted twice in the final solution, and performed the chemotaxis assay with it. We compared chemotaxis dynamics, directness, and bearing between the single chemoattractant experiments and the mixture.
To study the significance of directional changes following a pirouette in light of our findings, we simulated worm courses towards an attractant using three different reorientation strategies, namely (1) choosing the directional change (ΔB) uniformly between –180° and +180°; (2) uniformly sampling from the directional changes observed in our experiments regardless of the specific angle in which the worm entered the pirouette; and (3) we first divided the directional changes observed in our experiments into two groups – one group contained directional changes made by worms which were initially directed towards the target (–90° < BBefore < +90°), and the other contained directional changes made by off-course worms (+90° < BBefore < 270°). We next chose the directional change based on the angle of the ‘simulated worm’ just prior to the pirouette (e.g., if the worm was initially directed towards the target then the directional change was sampled from the directional change group of the directed worms). Interestingly, worms simulated using the third strategy reached their target significantly faster than if simulated using the first two strategies (Fig. 5c).
For these simulations, we used a simple model considering a minimal number of parameters, namely worm start point, chemoattractant position (target point), and probabilities for a pirouette of directed and undirected worms. The first two parameters, the start and target coordinates, were extracted directly from the chemotaxis experiments. The probability for a pirouette of directed worms was set to 0.03 per second, reflecting two pirouettes per minute, and the probability for a pirouette of undirected worms was set to be five times more probable (0.15). All the angular differences before and after a pirouette were directly drawn from our experimental data based on directed and undirected worm behavior. For this, we extracted from the experimental data all pirouette instances and calculated angular differences before and after a pirouette and constructed a distribution curve. For simulations, we drew angular angle differences based on these distributions. Importantly, the simulation results were not sensitive to small changes in the parameters set.
Aging- and neurodegenerative-associated locomotion decline
To study aging- and neurodegenerative-associated locomotion decline we prepared animals as described for the chemotaxis assays above. Briefly, for aging assays, we employed temperature-sensitive sterile worms (strain CF512) that become sterile when exposed to 25 °C during development. We synchronized animals by bleaching gravid animals and performed the motility assays on days 1, 4, 8, and 12 of adulthood. For the daf-2 RNAi experiments we used empty vector as control, and age-synchronized worm locomotion was measured at days 1, 6, and 12 of adulthood.
To assay locomotion in neurodegenerative-associated disease we used the AM140 strain. These worms express the aggregative peptide composed of 35 repeats of glutamine fused to the yellow fluorescent protein in the body wall muscles . Motility was assayed at days 2, 5, and 8 of adulthood. In all these assays, the repellent 2-nonanone was added to stimulate motility. We started measuring worm locomotion 2 minutes after the addition of 2-nonanone for the duration of 1 minute.
We thank Gil Lefkowitz and Michael Gliksberg for providing movies of swimming zebrafish, Sebastian Kadener and Daniella Apelblat for the fly movies, and Ami Citri and Bogna Ignatowska for the mice movies. We thank Yossi Gruenbaum, Guy Bloch, Sagiv Shifman, Ami Citri, Dana Rubinstein, Rotem Ruach, Aharon Azulay, Aharon Wayne, and the rest of our lab for critical comments.
ERC starting grant (#336803), the American Federation for Aging Research (AFAR), and the Joseph H. and Belle R. Braun funds with which AZ is a Senior Lecturer chair.
Availability of data and material
The software suite together with its freely-available code can be downloaded through github (https://github.com/itskov/MultiAnimalTrackerSuite) as well as through our website (www.zaslaverlab.com). The downloaded files also include a detailed user manual (with exemplar analyzed movies) for installing and using the different packages in the software. Additional file 12: Table S1 provides raw values for Fig. 7a, c, and d.
AZ and EI conceived the study. EI developed and implemented the software with AZ guidance. EI performed all the experiments (guided by AZ) except for the aging-associated motility assays that were performed by AL (guided by EC). AZ and EI wrote the paper with input from EC and AL. All authors read and approved the final manuscript.
The authors declare that they have no competing interests.
Consent for publication
Ethics approval and consent to participate
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- Tinbergen N. On aims and methods of Ethology. Anim Biol. 2005;55(4):297–321.View ArticleGoogle Scholar
- Anderson DJ, Perona P. Toward a science of computational ethology. Neuron. 2014;84(1):18–31.View ArticlePubMedGoogle Scholar
- Brenner S. The genetics of Caenorhabditis elegans. Genetics. 1974;77(1):71–94.PubMedPubMed CentralGoogle Scholar
- White JG, Southgate E, Thomson JN, Brenner S. The structure of the nervous system of the nematode Caenorhabditis elegans. Philos Trans R Soc Lond B Biol Sci. 1986;314(1165):1–340.View ArticlePubMedGoogle Scholar
- Barr MM, Garcia LR. Male mating behavior. WormBook. 2006. The C. elegans Research Community. doi:10.1895/wormbook.1.7.1.
- Flavell SW, Pokala N, Macosko EZ, Albrecht DR, Larsch J, Bargmann CI. Serotonin and the neuropeptide PDF initiate and extend opposing behavioral states in C. elegans. Cell. 2013;154(5):1023–35.View ArticlePubMedPubMed CentralGoogle Scholar
- Raizen DM, Zimmerman JE, Maycock MH, Ta UD, You YJ, Sundaram MV, Pack AI. Lethargus is a Caenorhabditis elegans sleep-like state. Nature. 2008;451(7178):569–72.View ArticlePubMedGoogle Scholar
- Mori I. Genetics of chemotaxis and thermotaxis in the nematode Caenorhabditis elegans. Annu Rev Genet. 1999;33:399–422.View ArticlePubMedGoogle Scholar
- Vidal-Gadea A, Ward K, Beron C, Ghorashian N, Gokce S, Russell J, Truong N, Parikh A, Gadea O, Ben-Yakar A, et al. Magnetosensitive neurons mediate geomagnetic orientation in Caenorhabditis elegans. Elife. 2015;4. doi:10.7554/eLife.07493.
- Bargmann CI, Horvitz HR. Chemosensory neurons with overlapping functions direct chemotaxis to multiple chemicals in C. elegans. Neuron. 1991;7(5):729–42.View ArticlePubMedGoogle Scholar
- Bargmann CI. Chemosensation in C. elegans. WormBook. The C. elegans Research Community. 2006. doi:10.1895/wormbook.1.7.1.
- Pierce-Shimomura JT, Morse TM, Lockery SR. The fundamental role of pirouettes in Caenorhabditis elegans chemotaxis. J Neurosci. 1999;19(21):9557–69.PubMedGoogle Scholar
- de Bono M, Maricq AV. Neuronal substrates of complex behaviors in C. elegans. Annu Rev Neurosci. 2005;28:451–501.View ArticlePubMedGoogle Scholar
- Sengupta P, Samuel AD. Caenorhabditis elegans: a model system for systems neuroscience. Curr Opin Neurobiol. 2009;19(6):637–43.View ArticlePubMedPubMed CentralGoogle Scholar
- Carvalhal Marques F, Volovik Y, Cohen E. The roles of cellular and organismal aging in the development of late-onset maladies. Annu Rev Pathol. 2015;10:1–23.View ArticlePubMedGoogle Scholar
- Cronin CJ, Feng Z, Schafer WR. Automated imaging of C. elegans behavior. Methods Mol Biol. 2006;351:241–51.PubMedGoogle Scholar
- Husson SJ, Costa WS, Schmitt C, Gottschalk A. Keeping track of worm trackers. WormBook. 2012. The C. elegans Research Community. doi:10.1895/wormbook.1.7.1.
- Baek JH, Cosman P, Feng Z, Silver J, Schafer WR. Using machine vision to analyze and classify Caenorhabditis elegans behavioral phenotypes quantitatively. J Neurosci Methods. 2002;118(1):9–21.View ArticlePubMedGoogle Scholar
- Yemini E, Jucikas T, Grundy LJ, Brown AE, Schafer WR. A database of Caenorhabditis elegans behavioral phenotypes. Nat Methods. 2013;10(9):877–9.View ArticlePubMedPubMed CentralGoogle Scholar
- Ramot D, Johnson BE, Berry Jr TL, Carnell L, Goodman MB. The Parallel Worm Tracker: a platform for measuring average speed and drug-induced paralysis in nematodes. PLoS One. 2008;3(5):e2208.View ArticlePubMedPubMed CentralGoogle Scholar
- Swierczek NA, Giles AC, Rankin CH, Kerr RA. High-throughput behavioral analysis in C. elegans. Nat Methods. 2011;8(7):592–8.View ArticlePubMedPubMed CentralGoogle Scholar
- Yemini E, Kerr RA, Schafer WR. Tracking movement behavior of multiple worms on food. Cold Spring Harb Protoc. 2011;2011(12):1483–7.PubMedPubMed CentralGoogle Scholar
- Ma DK, Vozdek R, Bhatla N, Horvitz HR. CYSL-1 interacts with the O2-sensing hydroxylase EGL-9 to promote H2S-modulated hypoxia-induced behavioral plasticity in C. elegans. Neuron. 2012;73(5):925–40.View ArticlePubMedPubMed CentralGoogle Scholar
- Salmond D. Target tracking: introduction and Kalman tracking filters. In: IEEE International Seminar Target Tracking: Algorithms and Applications. Stevenage: IET; 2001. p. 1–16.Google Scholar
- Gomez-Marin A, Stephens GJ, Louis M. Active sampling and decision making in Drosophila chemotaxis. Nat Commun. 2011;2:441.View ArticlePubMedPubMed CentralGoogle Scholar
- Pierce-Shimomura JT, Dores M, Lockery SR. Analysis of the effects of turning bias on chemotaxis in C. elegans. J Exp Biol. 2005;208(Pt 24):4727–33.View ArticlePubMedGoogle Scholar
- Iino Y, Yoshida K. Parallel use of two behavioral mechanisms for chemotaxis in Caenorhabditis elegans. J Neurosci. 2009;29(17):5370–80.View ArticlePubMedGoogle Scholar
- Stein BE. The New Handbook of Multisensory Processes. Cambridge: MIT Press; 2012.Google Scholar
- Fetsch CR, DeAngelis GC, Angelaki DE. Bridging the gap between theories of sensory cue integration and the physiology of multisensory neurons. Nat Rev Neurosci. 2013;14(6):429–42.View ArticlePubMedGoogle Scholar
- Duistermars BJ, Frye MA. Multisensory integration for odor tracking by flying Drosophila: Behavior, circuits and speculation. Commun Integr Biol. 2010;3(1):60–3.View ArticlePubMedPubMed CentralGoogle Scholar
- Dubnikov T, Cohen E. Proteostasis collapse, inter-tissue communication, and the regulation of aging at the organismal level. Front Genet. 2015;6:80.View ArticlePubMedPubMed CentralGoogle Scholar
- Herndon LA, Schmeissner PJ, Dudaronek JM, Brown PA, Listner KM, Sakano Y, Paupard MC, Hall DH, Driscoll M. Stochastic and genetic factors influence tissue-specific decline in ageing C. elegans. Nature. 2002;419(6909):808–14.View ArticlePubMedGoogle Scholar
- Stroustrup N, Anthony WE, Nash ZM, Gowda V, Gomez A, Lopez-Moyado IF, Apfeld J, Fontana W. The temporal scaling of Caenorhabditis elegans ageing. Nature. 2016;530(7588):103–7.View ArticlePubMedPubMed CentralGoogle Scholar
- Kenyon CJ. The genetics of ageing. Nature. 2010;464(7288):504–12.View ArticlePubMedGoogle Scholar
- Cohen E, Bieschke J, Perciavalle RM, Kelly JW, Dillin A. Opposing activities protect against age-onset proteotoxicity. Science. 2006;313(5793):1604–10.View ArticlePubMedGoogle Scholar
- Lithgow GJ, White TM, Melov S, Johnson TE. Thermotolerance and extended life-span conferred by single-gene mutations and induced by thermal stress. Proc Natl Acad Sci U S A. 1995;92(16):7540–4.View ArticlePubMedPubMed CentralGoogle Scholar
- Morley JF, Brignull HR, Weyers JJ, Morimoto RI. The threshold for polyglutamine-expansion protein aggregation and cellular toxicity is dynamic and influenced by aging in Caenorhabditis elegans. Proc Natl Acad Sci U S A. 2002;99(16):10417–22.View ArticlePubMedPubMed CentralGoogle Scholar
- Sznitman R, Gupta M, Hager GD, Arratia PE, Sznitman J. Multi-environment model estimation for motility analysis of Caenorhabditis elegans. PLoS One. 2010;5(7):e11631.View ArticlePubMedPubMed CentralGoogle Scholar
- Greenblum A, Sznitman R, Fua P, Arratia PE, Sznitman J. Caenorhabditis elegans segmentation using texture-based models for motility phenotyping. IEEE Trans Biomed Eng. 2014;61(8):2278–89.View ArticlePubMedGoogle Scholar
- Branson K, Robie AA, Bender J, Perona P, Dickinson MH. High-throughput ethomics in large groups of Drosophila. Nat Methods. 2009;6(6):451–7.View ArticlePubMedPubMed CentralGoogle Scholar
- Kabra M, Robie AA, Rivera-Alba M, Branson S, Branson K. JAABA: interactive machine learning for automatic annotation of animal behavior. Nat Methods. 2013;10(1):64–7.View ArticlePubMedGoogle Scholar
- Bargmann CI, Hartwieg E, Horvitz HR. Odorant-selective genes and neurons mediate olfaction in C. elegans. Cell. 1993;74(3):515–27.View ArticlePubMedGoogle Scholar
- Chalasani SH, Chronis N, Tsunozaki M, Gray JM, Ramot D, Goodman MB, Bargmann CI. Dissecting a circuit for olfactory behaviour in Caenorhabditis elegans. Nature. 2007;450(7166):63–70.View ArticlePubMedGoogle Scholar
- Sengupta P, Chou JH, Bargmann CI. odr-10 encodes a seven transmembrane domain olfactory receptor required for responses to the odorant diacetyl. Cell. 1996;84(6):899–909.View ArticlePubMedGoogle Scholar
- Zaslaver A, Liani I, Shtangel O, Ginzburg S, Yee L, Sternberg PW. Hierarchical sparse coding in the sensory system of Caenorhabditis elegans. Proc Natl Acad Sci U S A. 2015;112(4):1185–9.View ArticlePubMedPubMed CentralGoogle Scholar
- Gray JM, Hill JJ, Bargmann CI. A circuit for navigation in Caenorhabditis elegans. Proc Natl Acad Sci U S A. 2005;102(9):3184–91.View ArticlePubMedPubMed CentralGoogle Scholar
- Piggott BJ, Liu J, Feng Z, Wescott SA, Xu XZ. The neural circuits and synaptic mechanisms underlying motor initiation in C. elegans. Cell. 2011;147(4):922–33.View ArticlePubMedPubMed CentralGoogle Scholar
- Kocabas A, Shen CH, Guo ZV, Ramanathan S. Controlling interneuron activity in Caenorhabditis elegans to evoke chemotactic behaviour. Nature. 2012;490(7419):273–7.View ArticlePubMedPubMed CentralGoogle Scholar
- Haralick R, Shapiro L. Computer and Robot Vision, vol. 1. Boston: Addison-Wesley Publishing Company; 1992.Google Scholar
- Horn B. Robot Vision. Cambridge: MIT Press; 1986.Google Scholar
- Anderson TW. Introduction to Multivariate Statistical Analysis. Hoboken: Wiley; 1958.Google Scholar