Automated multimodal imaging of Caenorhabditis elegans behavior in multi-well plates

Abstract

Assays of behavior in model organisms play an important role in genetic screens, drug testing, and the elucidation of gene-behavior relationships. We have developed an automated, high-throughput imaging and analysis method for assaying behaviors of the nematode Caenorhabditis elegans. We use high-resolution optical imaging to longitudinally record the behaviors of 96 animals at a time in multi-well plates, and computer vision software to quantify the animals’ locomotor activity, behavioral states, and egg-laying events. To demonstrate the capabilities of our system, we used it to examine the role of serotonin in C. elegans behavior. We found that egg-laying events are preceded by a period of reduced locomotion, and that this decline in movement requires serotonin signaling. In addition, we identified novel roles of serotonin receptors SER-1 and SER-7 in regulating the effects of serotonin on egg laying across roaming, dwelling, and quiescent locomotor states. Our system will be useful for performing genetic or chemical screens for modulators of behavior.

Behavioral assays in model organisms are widely used in genetic and chemical screens. This study presents an imaging system for longitudinally recording behavior of C. elegans in a 96-well plate. The software automatically measures locomotor activity, behavioral states, egg-laying events, and locomotor frequency. The system is tested by analyzing the roles of serotonin signaling on locomotion and egg-laying behaviors. The results revealed roles of serotonin receptors in regulating these behaviors across different behavioral states. This imaging and analysis pipeline will facilitate screens for genetic or pharmacological modulators of behavior.

Introduction

Caenorhabditis elegans behavioral assays are widely used to study biological mechanisms in diverse areas such as aging (Stroustrup et al. 2013; Churgin, Jung, et al. 2017), sleep (Raizen et al. 2008), development (Stern et al. 2017), and neurological disorders (Calahorro and Ruiz-Rubio 2011; Mondal et al. 2016). Assays of worm behavior have traditionally been done by manual assays and visual observations, which are labor intensive, subject to observer error, and low in throughput due to examination of a single animal at a time.

To address these limitations, researchers have developed systems for analyzing worm behaviors via video recording and software analysis (Cronin et al. 2006; Husson et al. 2018). These include methods for imaging worms on standard agar plates (Cronin et al. 2005; Stephens et al. 2008; Swierczek et al. 2011; Yemini et al. 2013; Luo et al. 2014; Cermak et al. 2020; Ji et al. 2021), in microfluidic devices (Chalasani et al. 2007; Hulme et al. 2007; Chung et al. 2008; Doitsidou et al. 2008; Albrecht and Bargmann 2011; Ji et al. 2023), and in multi-well substrates (Churgin, Jung, et al. 2017; McCloskey et al. 2017; Chen et al. 2020; Fouad et al. 2021; Harrington et al. 2022; Fryer et al. 2024).

Our laboratory has been conducting behavioral screens for the effect of genetic mutations and/or pharmacological compounds on various aspects of C. elegans behavior, including locomotion, egg laying, and behavioral states such as quiescence, dwelling, and roaming (Churgin, Jung, et al. 2017; Churgin, McCloskey, et al. 2017; McCloskey et al. 2017; Fouad et al. 2018; Ji et al. 2021; Ji et al. 2023). Toward this end, we sought to develop an efficient method for screening a large number of conditions for multiple behaviors simultaneously.

Here, we report a method for multimodal behavioral recording and analysis of C. elegans using 96-well plates. We developed a computer vision software toolbox to automate analysis of several behavioral metrics including locomotion activity, behavioral states, locomotor frequency, and egg laying. This automated approach is capable of continuously recording behaviors from 96 individual animals simultaneously for several hours under controlled conditions. We demonstrate the capabilities of this system by analyzing the roles of serotonin signaling in locomotor and egg-laying behaviors. Our findings underscore the extensive capabilities of longitudinal imaging systems for large-scale phenotypic assays.

Materials and methods

C. elegans strains and maintenance

C. elegans strains were cultivated on Escherichia coli strain OP50 at 20°C using standard conditions (Sulston and Hodgkin 1988). The strains used in this work were obtained from the NIH Caenorhabditis Genetics Center. Mutant strains were the following: MT1082egl-1(n487), DA1814ser-1(ok345), DA2100ser-7(tm1325), DA2109ser-7(tm1325); ser-1(ok345). All experiments were performed with 1-day-old adult animals in a food-free environment. Worm populations were synchronized by timed egg lays according to standard methods (Gandhi et al. 1980).

Multi-well plate assays

To study the effect of various chemicals on egg-laying and locomotion behaviors, we used 96-well microtiter assay plates (Corning 3795, round bottom), with each well containing the indicated number of worms immersed in 60 µL of NGM buffer dissolved with the compound of interest. The liquid NGM buffer consists of the same ingredients as solid NGM but without agar, peptone, or cholesterol (Stiernagle 2006).

Image data acquisition and processing

The assay plate was placed on a platform within a custom-built imaging system. Dark-field illumination was generated by an LED light ring (outer diameter 12 inches; Sunpak) above the platform. A video was captured using a CMOS camera (DMK 33UX183, The Imaging Source) with a C-mount lens (Fujinon HF12.5SA-1, focal length 35 mm, f/2.8 or Nikon, focal length 55 mm, f/3.5 F-mount lens with a C-mount adapter) for 20 s at 5 fps using IC Capture software (The Imaging Source). A whole course of behavioral assay was performed by capturing a sequence of the 20 s videos, taken every 2 min for 5 h of recording time. The field of view of the system was set to display all 96 wells of the plate, which yields an 18 µm pixel resolution.

Video imaging data from the assay plate were analyzed using custom-written MATLAB and Python software. The behavioral outcomes from the analysis include locomotion activity, locomotor frequency, and egg-laying behavior. Locomotion activity is defined as the number of pixels whose intensity changes between subsequent image frames due to worms’ movements (Raizen et al. 2008; Churgin, Jung, et al. 2017). Locomotor frequency is calculated by our automated image analysis software capable of rapidly estimating rhythmic behaviors (Ji et al. 2024). Egg-laying behavior is quantified by the number of eggs released into the environment. Eggs are detected by a trained mask-regional convolutional neural network (Mask-RCNN) (He et al. 2017).

Measuring locomotory behavior

We calculated locomotor activity using previously described methods (Raizen et al. 2008; Churgin, McCloskey, et al. 2017). Pairs of temporally adjacent video images were subtracted to generate difference images, which were then divided by the average pixel intensity, yielding normalized maps of pixel value intensity change. A Gaussian smoothing filter with a standard deviation of 1 pixel was applied to decrease image noise. A binary threshold was applied to the filtered image to determine the presence or absence of movement at each pixel location. The sum of all pixels exhibiting movement was calculated as the locomotor activity.

To calculate locomotor frequency we used our custom algorithm Imaginera (image invariants ensemble for rhythmicity analysis) (Ji et al. 2024). For each video frame, our algorithm estimates the pixel area of a worm ($A_{\mathrm{worm}}$) in each well. When animals approached the edge of wells, they were occasionally occluded such that the animals could not be easily observed, decreasing $A_{\mathrm{worm}}$. To mitigate the influence of this occlusion, we censored any frames during which $A_{\mathrm{worm}}$ was smaller than a preset empirical threshold when calculating locomotor activity or frequency. For all the image data used in this study, the number of censored images within individual videos was less than 0.2% on average and less than 5% at maximum.

Measuring egg-laying behavior

To evaluate egg-laying behavior, we counted the eggs that have been released into each well in 20 s videos. Each video was sampled by 5 images evenly spaced across the 20 s interval, with adjacent images separated by 4 s. Through manual counting, the events where eggs were occluded by moving worms were counted to yield a frequency less than 0.3%.

Following previously described methods (Li et al. 2023), we deployed a Mask-RCNN to detect and identify eggs in each image. The Mask-RCNN model was trained, evaluated, and deployed by PyTorch GPU (v2.0.0 + cu1.1.8) on a system equipped with a 13th Intel(R) Core(TM) i7-13700K processor and an NVIDIA GeForce RTX 4080 GPU.

For the model training, 1,653 cropped single-well images (containing eggs and worms) from 18 videos in all 96 distinct well positions were manually annotated and saved as training datasets using a Python annotation package. We augmented the training set using operations including flipping, intensity adjustment, and Gaussian blurring, yielding a total of 6,612 annotated images in the dataset. The model was trained for 25 epochs, during which the training set was randomly divided into training and testing sets with an 80/20 split ratio (5,290 images for training, 1,322 images for testing). We evaluated the Mask-RCNN model using characterization metrics, including precision (a measure of false positive rate), recall (a measure of false negative rate), and average precision (a measure of overall accuracy). Specifically, the RCNN model yielded 0.85 in precision, 0.81 in recall, and 0.75 in average precision; these metrics were similar to those of previous RCNN models used for C. elegans egg detection (Bates et al. 2022).

With the trained Mask-RCNN model, the egg number in an image was determined by $N_{\mathrm{RCNN}}=A_{\mathrm{RCNN}}/a_0$, where $A_{\mathrm{CNN}}$ represents the pixel number of egg objects inferred by the model, and $a_0$ is the pixel number of a single egg, estimated by $a_0=\sum A_{\mathrm{manual}}{N}_{\mathrm{manual}}/\sum N_{\mathrm{manual}}^2$. Here, $\sum A_{\mathrm{manual}}{N}_{\mathrm{manual}}$ is the sum of the products of egg pixel number and egg number manually counted from individual images in the training set, and $\sum N_{\mathrm{manual}}^2$ is the sum of squares of egg numbers from the same set.

Mask R-CNN model characterization

We evaluated how well our model performed by calculating precision and recall as well as average precision (AP). Precision is the proportion of true positive predictions (correctly detected objects) among all positive predictions (correctly detected objects plus false positives). Recall is the proportion of true positive predictions among all actual positive instances (true positives plus false negatives). Average precision was calculated by computing the area under the Precision-Recall curve through the following steps: (1) for each confidence threshold, calculate precision and recall. (2) Interpolate the precision values by taking the maximum precision for any recall value greater than the current recall value. (3) Compute the area under the interpolated Precision-Recall curve using numerical integration.

To determine true positives (TP), false positives (FP), and false negatives (FN), we used the Intersection-over-Union (IoU) threshold, which measures the overlap of the ground-truth bounding box and the prediction bounding box (Bates et al. 2022). An egg detection is considered a TP if its IoU with the corresponding ground-truth bounding box is greater than or equal to the IoU threshold. If the IoU is below the threshold, it is considered an FP. An FN is a ground-truth bounding box that does not have a corresponding detection. An IoU threshold of 0.3 and a confidence score threshold of 0.01 were used for our egg-detection model.

Statistical analysis

Differences in locomotion activity, active state fraction, locomotor frequency, and egg-laying rate (Figs. 3, i–l, and 5, c–f) between various conditions were determined using the 2-tailed Student's t-test. Comparisons of serotonin resistance between different strains were performed using the 2-way ANOVA test (Fig. 3, i and j). Differences in midpoint egg-laying time distributions (Fig. 3m) were assessed using a Wilcoxon rank sum test. All experiments were carried out in 3 independent replicates with the identity of animal subjects and experimental conditions blinded to experimenters.

Fig. 3.

Automated large-scale tracking of locomotory and egg-laying behaviors. a–c) Heat maps of activity a), locomotor frequency b), and egg-laying rate c) for wild-type and mutant individuals (n = 32 for each strain group). Each row in the heatmap shows the behavioral dynamics of a single animal. d–f) Same as a–c), but for animals in an exogenous serotonin environment. g) Normalized activity over time for an individual. The data are normalized by the mean value of the 95th to 100th percentile of all activity values observed during the experiment. Dashed line indicates the threshold value of normalized activity above which locomotion bouts are labeled as active states. h) Cumulative number of eggs released by 2 individuals. Asterisk indicates midpoint egg-laying time (t₅₀), the time taken to reach one-half of the total egg number accumulated at t = 5 h. i) Locomotion activity of wild-type and mutant individuals. j) Fraction of time during active state of wild-type and mutant individuals. k) Mean locomotor frequency during active state of wild-type and mutant individuals. l) Egg-laying rate of wild-type and mutant animals. m) Midpoint egg-laying time (t₅₀) of wild-type and mutant individuals. i–m) Each point represents an animal from groups in the indicated serotonin conditions, error bars represent population mean ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001, Student's t-test. i and j) Several indicated dosed/undosed pairs of strains were compared to further investigate the interaction between the genetic and chemical factors. #P < 0.05, ###P < 0.001, 2-way ANOVA.

Fig. 5.

Receptors SER-1 and SER-7 regulate serotonin effects on reproduction in separate locomotory states. a and b) Histogram of normalized activity (stairstep curve, right-side y-axis) and scatter plot showing egg-laying rate vs normalized activity value (dot, left-side y-axis) for wild-type animals in either the absence a) or presence b) of exogenous serotonin. In histograms, the height of the stairstep denotes the relative probability of each of 20 bins of normalized activity, dashed line indicates the threshold value of normalized activity above which locomotion is flagged as active state, shaded areas on the left and right of the dashed line represent proportions of sedentary and active states. In scatter plots, each point represents a 20 s bout of an individual animal (n = 4,832 bouts from 32 animals for each condition). c) (Upper) Total number of eggs released by wild-type animals during sedentary or active states in either the absence or presence of exogenous serotonin. (Lower) Fraction of time of sedentary vs active states in respective serotonin conditions. d) Egg-laying rate of wild-type animals during sedentary or active states in either the absence or presence of exogenous serotonin. e and f) Egg-laying rate of wild-type and mutant individuals during sedentary e) or active f) state in either the absence or presence of exogenous serotonin. c–f) Each point represents an animal, and error bars represent population mean ± SEM. ***P < 0.001.

Results

System design for high-content imaging of multi-well assay platform

Our imaging platform contains a 20 MP CMOS camera, a C-mount lens, a 30 cm diameter LED ring illuminator, and a plate holder (Fig. 1a). One or more worms in liquid NGM media are prepared in each well of a 96-well plate and the plate is imaged under dark-field illumination (Fig. 1, a and b). The platform components are positioned to visualize worms and embryos in each well while minimizing stray light from well edges (Fig. 1b).

Fig. 1.

System design and behavioral quantification. a) Schematic of the imaging system and image of a 96-well assay plate containing 1 worm per well. b) Detail of a well containing a single 1-day adult and multiple eggs. c) Activity is calculated as the sum of the number of pixels whose pixel value intensities change above a threshold. The change in the pixel value intensity is calculated by the pixel-by-pixel difference of temporally adjacent images. d) Activity over time of 2 individual wild-type animals. e) Locomotor frequency over time of 2 wild-type animals. f) Egg detection via Mask R-CNN from the original video frame (left). The egg objects are annotated with masks inferred by the model (right). g) Number of eggs laid for 2 wild-type animals.

Computer vision automates multimodal behavioral measurements

We developed image analysis software for quantifying animal locomotory and egg-laying behaviors within each well. Using a combination of computer vision algorithms and convolutional neural network (CNN), the software analyzes images to measure behavioral parameters, including locomotion activity, locomotor frequency, and egg-laying rate (see Materials and Methods).

Locomotion activity is computed through pixel-by-pixel subtraction of consecutive video frames, with the activity defined by the number of pixels exhibiting intensity changes between subsequent frames (Fig. 1, c and d) (Raizen et al. 2008). To measure locomotor frequency, we used an algorithm based on analysis of image invariants (Fig. 1e) (Ji et al. 2024). The image analysis software also employs a Mask-RCNN to automatically annotate and enumerate worm embryos in each well (Fig. 1, f and g) (He et al. 2017; Li et al. 2023).

Evaluation of automated measurements

We tested the accuracy of our software in analyzing the locomotor frequency and egg-laying events by comparing the software's measurements with those obtained through human observation.

The automated frequency results demonstrated good agreement with manual data with an overall root mean squared error of 0.097 Hz and an R-squared value of 0.964 (Ji et al. 2024). We show that compared to previous established computer-vision approaches (e.g. segmentation-based method), our technique is more robust to low image quality and background noise (Ji et al. 2024). We found that locomotion activity is also highly correlated with locomotor frequency by comparing it with the ground-truth data (Fig. 2a).

Fig. 2.

Evaluation of automated behavioral measurements. a) Locomotion activity compared with the frequencies calculated by manual observation of recorded videos. The line indicates a linear fit with zero offset. b) Egg numbers inferred by the Mask-RCNN model compared with those counted manually. The line shows equality between predicted and ground truth numbers.

To assess the accuracy of the egg-detector RCNN model, we applied it to identify and count eggs across all 96 wells of the plate. The collected well images contained 1-day-old adult wild-type worms alongside the eggs they laid, captured at 2 and 5 h after animals were transferred into the assay plate. We found that the egg counts inferred by the model were in good agreement with those counted manually (Fig. 2b).

Together, the automated methods of our system accurately measure locomotion activity, locomotor frequency, and embryo numbers through longitudinal imaging.

Automated tracking of locomotion, behavioral states, and egg laying

C. elegans locomotory and egg-laying behaviors are modulated by serotonergic, dopaminergic, and peptidergic signaling (Brenner 1974; McFarland 1977; Zhen and Samuel 2015). The C. elegans nervous system uses serotonin (5-HT) to modulate both its locomotory and egg-laying behaviors, and external serotonin has been found to inhibit locomotion and stimulate egg-laying activity (Horvitz et al. 1982; Trent et al. 1983; Schafer and Kenyon 1995; McCloskey et al. 2017).

To evaluate our system's reliability and capability for large-scale assays, we tracked the locomotion and egg-laying behaviors of wild-type and mutant C. elegans individuals under controlled serotonin conditions (see Materials and Methods and Fig. 3, a–f). Animals were individually housed in wells of an assay plate filled with (food-free) liquid NGM and defined serotonin concentrations (5-HT−: 0 mM; 5-HT+: 6.5 mM). In each experiment, we monitored up to 96 animals for 5 h, which yielded 151 video clips with 5 fps and 20 s duration.

Both wild-type and mutant animals under serotonin-free conditions showed relatively active locomotion (Fig. 3, a and b) and low egg-laying rates (Fig. 3c). In contrast, the administration of exogenous serotonin induced diverse behavioral dynamics across the tested strains (Fig. 3, d–f).

We defined parameters for quantifying each animal's locomotory states (Flavell et al. 2013). We calculated the mean of each animal's higher 5% activity values across the experiment and used this value to define a normalized activity. These data allowed us to identify sedentary and active bouts of locomotion using an empirical threshold (Fig. 3g) (McCloskey et al. 2017). We focused on assessing the average locomotion activity, the fraction of time spent in the active state, and the average locomotor frequency during the active state.

Wild-type and mutant animal populations exhibited diverse locomotion activity levels and active locomotory states across genotypes and different levels of exogenous serotonin (Fig. 3, i and j). Both wild-type and mutant animals generally exhibited high activity levels and spent the majority of time (over 78%) in the active state under the serotonin-free condition (Fig. 3, i and j). With the addition of exogenous serotonin, activity levels and active state fractions decreased in wild-type animals (reduced by 39 and 29% on average, respectively) and in egl-1 mutants (which lack HSN serotonergic neurons that innervate egg-laying muscles; reduced by 68 and 60%, respectively) (Fig. 3, i and j). In contrast to the wild type, ser-1 and ser-7 mutants (deficient in G protein-coupled metabotropic receptors SER-1 and SER-7) both exhibited resistance to the effects of serotonin on locomotion activity (reduced by 25 and 15% for the respective mutants) and active state fractions (reduced by 24 and 9% for the respective mutants), while ser-1;ser-7 double mutants showed enhanced resistance to exogenous serotonin (2% reduction in locomotion activity, 6% reduction in active state fraction) (Fig. 3, i and j). This result suggests that the SER-1 and SER-7 receptors play a role in mediating inhibitory effects of external serotonin on locomotor behavior.

We calculated the average locomotor frequency of each animal during its active state. The average active locomotor frequency of wild-type and all tested mutant strains showed little or no change in response to exogenous serotonin (Fig. 3k). Therefore, unlike the active state fraction, the active locomotor frequency remained largely constant, independent of exogenous serotonin. This experiment suggests that external serotonin on locomotion primarily affects initiation of activity rather than alteration of an established locomotor rhythm.

To characterize the egg-laying dynamics of each animal, we calculated the average egg-laying rate and the midpoint egg-laying time $t_{50}$ , the time taken to reach half of the total number of eggs laid during the experiment (Fig. 3h). Under serotonin-free conditions, wild-type animals averaged an egg-laying rate of 1.2 eggs/h (Fig. 3l). The population's $t_{50}$ values exhibited a biphasic distribution, with the majority displaying early midpoint egg-laying times (Fig. 3m). Compared to the wild type, the mutant animals tested had similar egg-laying rates but displayed different distributions in $t_{50}$ (Fig. 3, l and m).

In the presence of exogenous serotonin, wild-type and egl-1 mutant animals showed significant increases in egg-laying rates and decreases in $t_{50}$ values (Fig. 3, l and m). The ser-1 and ser-7 mutants had a moderate increase in egg-laying rates in response to exogenous serotonin, while ser-1;ser-7 double mutants maintained low egg-laying rates (Fig. 3l). The values of $t_{50}$ in the wild-type and egl-1 mutant animals were suppressed with the addition of exogenous serotonin, which was not seen in the ser-1, ser-7, and ser-1;ser-7 double mutants (Fig. 3m). This result indicates that SER-1 and SER-7 receptors may be necessary for normal serotonin-mediated egg laying involving both egg-laying rates and temporal patterns.

In summary, our behavioral findings qualitatively agree with previously reported results for each strain, encompassing behavioral parameters such as locomotory state, locomotor frequency, and egg-laying rate, as well as the effects of exogenous serotonin (Horvitz et al. 1982; Trent et al. 1983; Hardaker et al. 2001; Carnell et al. 2005; Hobson et al. 2006; Churgin, McCloskey, et al. 2017; Stern et al. 2017; Chen et al. 2020). These experiments demonstrate our system's reliability, scalability, and efficiency in concurrently and longitudinally assessing multi-metric behavioral dynamics.

Serotonin is required for preparatory movement decline during egg-laying

One advantage of our system is that it is capable of capturing locomotor and egg-laying behaviors simultaneously. We examined the temporal correlation between locomotion and egg-laying by extracting the locomotion pattern before and after each egg-laying event throughout the recording period (Fig. 4, a and b). The average locomotor activity of wild-type animals was found to decrease approximately 10 min prior to egg-laying events, with the minimum activity reached roughly at the time the egg(s) were laid and subsequently recovering to a baseline value within about 10 min (Fig. 4c). Observation of the video clips revealed that the activity of individual animals during egg-laying events was consistently lower than during adjacent periods during which no eggs were laid (individual traces in Fig. 4c). Notably, nearly every egg-laying event was preceded by a decline in activity, although the magnitude and timing of the maximal activity drop varied substantially both within individual animals and among different animals (Fig. 4c).

Fig. 4.

Serotonin is required for preparatory movement decline during egg laying. a and b) Locomotion activity and egg-laying dynamics for 2 individual wild-type animals in either the absence a) or presence b) of exogenous serotonin. Curves represent activity trajectories. Dots on time axis indicate egg-laying events scored by automated method, and dot size is used to visualize the egg-laying rate. c and d) Average normalized activity change around egg-laying events (dashed line, aligned at t = 0) of wild-type animals (n = 32) in either the absence c) or presence d) of exogenous serotonin. The solid line with shaded area indicates population mean ± SEM. Dashed curves represent 20 randomly selected individual trajectories for each group. e and f) Same as c and d), but for tph-1 mutant animals (n = 32). g and h) Same as c and d), but for egl-1 mutant animals (n = 32).

Since serotonin functions in circuits that control C. elegans locomotion and egg laying (Horvitz et al. 1982; Trent et al. 1983; Sawin et al. 2000; Flavell et al. 2013; Stern et al. 2017), we examined the effects of serotonin and the neurons regulating egg-laying behavior on the control of locomotion. We asked how exogenous serotonin might modulate the observed correlation between locomotion and egg-laying. In the presence of exogenous serotonin, the decline in movement around egg-laying events was more pronounced. Furthermore, the timing of the activity decline and subsequent recovery occurred approximately 4 min earlier than without exogenous serotonin (Fig. 4d).

Next, we explored the effects of endogenous serotonin signaling on the coordination between egg-laying and locomotion. We analyzed mutant animals deficient in serotonin due to a mutation in the biosynthetic enzyme tryptophan hydroxylase (TPH-1). While the overall locomotion of tph-1 mutants appeared normal (Yemini et al. 2013), these mutants did not exhibit a decline in movement around egg-laying events (Fig. 4e). The introduction of exogenous serotonin restored this coordination deficit (Fig. 4f). Together, these results show that serotonin signaling is required for the observed decline in movement during egg-laying events.

We further investigated the effects of neurons that regulate egg-laying behavior on the coordination with locomotion, focusing on the role of HSNs, a pair of serotonergic neurons crucial for a normal egg-laying process. We observed the behavior of egl-1 mutants, which specifically lack HSNs. While egl-1 mutants did exhibit a decline in movement around egg-laying events, the timing of their minimum activity was delayed, occurring consistently after egg-laying (Fig. 4g). Administering exogenous serotonin to egl-1 mutants altered the timing of the movement decline (Fig. 4h). These results suggest that HSNs may be bypassed with the administration of exogenous serotonin for restoring the coordination between egg laying and locomotion.

The metabotropic serotonin receptors SER-1 and SER-7 regulate serotonin effects on reproduction in separate locomotory states

Our analyses of C. elegans locomotion activity and behavioral states showed that animals occasionally alternated between sedentary and active states (Fig. 3, a, d, and g). This alternation between states was influenced by exogenous serotonin and seemed to happen at a frequency similar to that of egg-laying events (Fig. 4, a and b). We asked whether there is a correlation between C. elegans egg-laying behavior and behavioral states and whether one is influenced or modulated by the other.

We discerned sedentary and active periods from the locomotion activity trajectories of individual animals and registered these periods with their respective egg-laying rates (Fig. 5, a and b). We noted that in serotonin-free environments, wild-type animals spent the majority of their time in the active state (Fig. 5a) and laid more eggs during these active periods (Fig. 5c). Conversely, in environments with exogenous serotonin, animals spent nearly equal amounts of time on sedentary and active states (Fig. 5b), with more eggs being laid during the sedentary states (Fig. 5c).

To compare the egg-laying activity between different locomotory states, we calculated the egg-laying rate by dividing the total egg count during each state with the corresponding state duration. The data revealed that, irrespective of exogenous serotonin, wild-type animals consistently exhibited higher egg-laying rates during sedentary locomotor state (Fig. 5d).

The observed dual role of serotonin in regulating both locomotory states and egg-laying behavior (as shown in Fig. 5, a and b) prompted a closer examination of serotonin signaling in modulating the observed dependence of egg-laying behavior on locomotory states (Figs. 4c and 5d). We revisited the behavioral recordings of HSN-defective mutants (egl-1) and serotonin receptor mutants (ser-1, ser-7, and ser-1;ser-7), with a particular focus on the effects of exogenous serotonin on the egg-laying rate during distinct locomotory states. By registering the egg-laying rate over time for each mutant individual and correlating it with the respective locomotory states, we observed that both wild-type and egl-1 mutant animals experienced a significant increase in egg-laying rate during both sedentary and active states when exposed to exogenous serotonin (Fig. 5, e and f). Interestingly, the ser-1 and ser-7 mutants resisted the effects of exogenous serotonin on their egg-laying rates, but each during one of the 2 states (Fig. 5, e and f). Meanwhile, the ser-1;ser-7 double mutants resisted the egg-laying effects of exogenous serotonin during both states (Fig. 5, e and f).

In summary, our findings suggest that C. elegans preferentially lay eggs during the sedentary state, with serotonin receptors SER-1 and SER-7 distributing the effects of serotonin on egg-laying behavior across distinct locomotory states.

Discussion

Our automated image recording and analysis pipeline, set within the conventional 96-well plates, yielded substantial benefits in throughput, preservation of animal individuality, multi-metric behavioral output, and manageable and controlled external conditions. Our setup facilitates the individual evaluation of parameters involving locomotion activity, behavioral states, and egg-laying behavior, revealing diverse characteristics at both individual and population levels. Applications enabled by our method included pharmacological screens (Kaletta and Hengartner 2006), genetic screens related to diverse neural functions (Bargmann and Marder 2013), and studies of complex behavioral patterns and states (Haspel et al. 2023).

Our system is scalable, platform-independent, and built from off-the-shelf parts. It is compatible with other standard or customized multi-array setups used for large-scale behavioral assays. Its modular analysis algorithm and customizable hardware configuration should allow for the integration of complex image-analysis routines specific to various screening tasks. For example, our system can be adapted for optogenetics-based screens by adding a high-intensity light source (Churgin, Jung, et al. 2017). The system can be suitable to taxis-based behavioral assays by using a customized plate with elongated lanes and controlled stimulus sources (e.g. chemical cues for chemotaxis) (Fryer et al. 2024). Because of the system's moderate dimension, it can fit in an intermediate sized thermal incubator and can potentially be used for temperature-dependent behavioral analyses. It also presents a versatile tool adaptable to studying other small organisms like Drosophila melanogaster and its larvae.

We observed that the algorithm for egg counting tends to slightly underestimate at higher egg numbers (Fig. 2b). We speculate that this may be due to the tendency for C. elegans egg laying to occur in short bursts, forming clumps of multiple eggs (Schafer 2006). In such a clump, eggs are likely to partially occlude each other, leading to an underestimate of egg numbers. To address this issue in the future, we may use an image dataset with a wider range of egg numbers to train a neural network model so that it detects individual eggs with higher accuracy in both sparse and crowded scenarios.

In our assays, the wild-type animals exhibit a decline in movement within a few minutes of egg-laying events. This finding diverges from those of previous analyses in which locomotion was observed to increase around egg-laying events (Hardaker et al. 2001; McCloskey et al. 2017). This discrepancy may stem from variations in experimental food conditions; while our studies monitored animals in food-free environments, prior experiments involved animals on food, which potentially influenced their basal behavioral and physiological states (Baugh and Hu 2020). Future experiments could explore how different levels of food availability influence the coordination and adaptations of these behaviors.

One potential concern in any multi-well imaging setup is uneven variability of conditions across the plate, i.e. positional effects (Roselle et al. 2016). These include nonuniformity in temperature, dehydration, and illumination across a plate, temporal effects due to different loading times, and the perspective effect, which causes wells near the edge of the field of view to appear slightly tilted compared to wells near the center. Common strategies for mitigating positional effects include avoiding using outer wells, complete or block-wise randomization of plate layout, and proper equipment calibration (Roselle et al. 2016). The perspective effect can be reduced by increasing the distance between the camera and plate and using a longer focal length lens such that the plate continues to match the size of the field of view; we have conducted experiments with lens focal lengths ranging from 35 to 75 mm. Another potential approach to reducing the perspective effect is to use a telecentric lens, which has a magnification independent of an object's distance from the lens.

Our multi-array and multi-metric behavioral tracking automation will facilitate rapid execution of large, complex compound or genetic screens, making it possible to detect subtle phenotypes that are usually difficult to identify with single-metric behavioral tracking prevalent in existing systems.

Data availability

Strains are available upon request. A full parts list with installation instructions, analysis software (freely available under a GNU General Public License), and test data are available at https://github.com/cfangyen/multiwell.

Funding

This work was funded by the National Institutes of Health (R01DA056358).