Automated behavioural analysis reveals the basic behavioural repertoire of the urochordate Ciona intestinalis

Quantitative analysis of animal behaviour in model organisms is becoming an increasingly essential approach for tackling the great challenge of understanding how activity in the brain gives rise to behaviour. Here we used automated image-based tracking to extract behavioural features from an organism of great importance in understanding the evolution of chordates, the free-swimming larval form of the tunicate Ciona intestinalis, which has a compact and fully mapped nervous system composed of only 231 neurons. We analysed hundreds of videos of larvae and we extracted basic geometric and physical descriptors of larval behaviour. Importantly, we used machine learning methods to create an objective ontology of behaviours for C. intestinalis larvae. We identified eleven behavioural modes using agglomerative clustering. Using our pipeline for quantitative behavioural analysis, we demonstrate that C. intestinalis larvae exhibit sensory arousal and thigmotaxis. Notably, the anxiotropic drug modafinil modulates thigmotactic behaviour. Furthermore, we tested the robustness of the larval behavioural repertoire by comparing different rearing conditions, ages and group sizes. This study shows that C. intestinalis larval behaviour can be broken down to a set of stereotyped behaviours that are used to different extents in a context-dependent manner.


Supplemental information on used methods
For every analysed video the position of the centre of the arena is determined with a Hough Circle Transform algorithm in OpenCV in Python. For every trace the [x,y]-positions are corrected so that [0,0] was at the centre of the arena. All positions were then multiplied by the factor of 11.56µm/pixel for the setup the recording originates from. From these positions distances, speeds and subsequently all other parameters are derived. We excluded animals that were completely immobile and hence indistinguishable from dead from further analysis by excluding all traces where the maximal displacement from the starting position was less than one body-length (comprising approximately 10% of all examined traces). Similarly, traces where the animal was tracked for less than 2,000 frames were considered unrepresentative and excluded from further analysis.
Activity coefficient (AC) is defined as the fraction of time an animal spent locomoting actively. Filtered speed values of 200 µm/s and above were considered as active. In Supplemental Figure 1 we present the distribution of speed values in our wild-type animals. Speeds higher than the node in the distribution at ca. 200 µm/s in practical terms represent active swimming as well as movement of the animals' centre-point due to tail flicks and twitching, but exclude moments of complete immobility or very slow drifting due to inertia.
Local path complexity was calculated using a method presented by Roberts et al 30 . In summary, this method uses embedding matrices for positions in a specific time window, over which the local path complexity is calculated in bits of entropy. At each time point we consider a matrix of (x,y) positions over 30 frames and by subtracting the mean values for x and y we are focusing on the variation of position around the mean in the selected time window. We perform singular value decomposition of this embedded matrix and calculate entropy of the normalised eigenvalues. Minimal complexity values calculated by this method correspond to the most invariable or 'predictable' trajectory with a straight-line trajectory resulting in the lowest value and paths with variation in speed and direction of movement resulting in high values. It is noteworthy that this measure is independent of the absolute values for mean position, speed and orientation in the time window, so that local path complexity represents the trace's variability at a given period regardless of the mean speed of movement.
Newly hatched chorionated or dechorionated animals were fixed in 4%PFA for one hour at room temperature. Then animals were washed in 1x PBS three times for 15 minutes each time. Animals were permeabilized with Proteinase K (4µg/ml for 30 minutes at 37°C. Proteinase K activity was stopped with Glycine (2mg/ml) for 5 minutes. The animals were washed three times in 1x PBS. Each wash lasted 5 minutes. We postfixed with 4%PFA for 1 hour at room temperature. Subsequently animals were incubated in BODIPY 493/503 (Thermo Fisher D3922) at a concentration of 1mg/ml dissolved in DMSO. The incubation lasted for two hours. Animals were then washed in 1x PBS three times for 15 minutes each time. A final 1 hour long incubation in 1x PBS with 1/10,000 DAPI was performed to stain cell nuclei. The PBS/DAPI solution was replaced with a DABCO/Glycerol mix and animals were mounted on slides for confocal imaging.

BODIPY Image analysis
Confocal stacks (40x) of dechorionated and wild type animals treated with BODIPY 493/503 were obtained on a Leica SP5 confocal microscope using equal settings for all acquisitions. Using FIJI/ImageJ scripts all acquired stacks were analyzed by splitting out the channel containing the BODIPY signal and calculating statistics on the volumes found by the built-in "3D objects counter" (parameters: min.=10, max.=1000. threshold.=125, exclude_objects_on_edges). This data was subsequently analyzed and visualized using python (numpy, pandas, matplotlib and scipy.stats). For each measured animal the total volume of all present signal was calculated. Data was transformed using a Box-Cox transformation, tested for normality with a Shapiro-Wilk test and compared using an independent t-test.

Oil Red O staining
Stock solution of Oil Red O (Sigma O0625) was prepared as follows : We added 0.5g of Oil Red O to 100ml of isopropanol and shaked for several hours. To generate a working solution, we 40ml of stock solution and diluted this to 60% using H2O, shaked for >1hr and filtered using a o,22µm filter. Chorionated and dechorionated larvae were harvested and fixed in 4%PFA for 1 hour at room temperature. Then they were washed 3 times in 1xPBS (15 minutes each wash). The animals were then placed in 60% isopropanol. After one hour the isopropanol was replaced with Oil Red O and animals were shaken for 24 hours. The animals were then washed 3 times in 1x PBS/0,01% Triton over a 2 hours period, prior to be mounted on a slide for microscopic analysis.

Supplemental Figure 7. Clustering methods
Variance explained in the dataset as a function of number of clusters is presented as the logarithm of the sum of distances to cluster centroids and its second differential. We present the data for the agglomerative clustering used in the paper (a) and k-means clustering (b).
Supplemental Figure 9 Median distance from the arena centre for the different experiments: As an additional measure of thigmotaxis, we used the median value for distance from the centre of the arena. The plots presented here use the same raw data as the main figures they refer to: (a) Acclimatization experiments (  (Fig. 4). (d) Temperature experiment (Fig. 6) p=0.068229, (e) dechorionation experiment (Fig. 7) p=0.022593. p values stated using Mann-Whitney U test.

Supplemental Video 1
Behavioural mode 1 Examples video of an inactive animal. The coloured dots represent the animal's position during 50 frames based on which the current behavioural mode was calculated. In green are the 25 positions succeeding the current timepoint and in red the 25 preceding it.

Supplemental Video 2
Behavioural modes 2 and 3 Examples video of an animal twitching, exhibiting behavioural modes 2 and 3. The coloured dots represent the animal's position during 50 frames based on which the current behavioural mode was calculated. In green are the 25 positions succeeding the current time-point and in red the 25 preceding it.

Supplemental Video 3
Behavioural modes 3, 4 and 8 Example video of an animal twitching, exhibiting behavioural modes 3, 4 and 8. The coloured dots represent the animal's position during 50 frames based on which the current behavioural mode was calculated. In green are the 25 positions succeeding the current time-point and in red the 25 preceding it.

Supplemental Video 4
Behavioural modes 1, 3, 4, 7, 8 and 9 Example video of an animal performing several different behavioural modes including the behavioural modes 7 and 9. The coloured dots represent the animal's position during 50 frames based on which the current behavioural mode was calculated. In green are the 25 positions succeeding the current time-point and in red the 25 preceding it.

Supplemental Video 5
Behavioural modes 5, 7, 9, 10 and 11 Example video of an animal performing several different behavioural modes. The coloured dots represent the animal's position during 50 frames based on which the current behavioural mode was calculated. In green are the 25 positions succeeding the current time-point and in red the 25 preceding it.