Click
here to close Hello! We notice that
you are using Internet Explorer, which is not supported by Echinobase
and may cause the site to display incorrectly. We suggest using a
current version of Chrome,
FireFox,
or Safari.
???displayArticle.abstract???
A defining feature of sea urchins is their extreme fecundity. Urchins produce millions of transparent, synchronously developing embryos, ideal for spatial and temporal analysis of development. This biological feature has been effectively utilized for ensemble measurement of biochemical changes. However, it has been underutilized in imaging studies, where single embryo measurements are used. Here we present an example of how stable genetics and high content imaging, along with machine learning-based image analysis, can be used to exploit the fecundity and synchrony of sea urchins in imaging-based drug screens. Building upon our recently created sea urchin ABCB1 knockout line, we developed a high-throughput assay to probe the role of this drug transporter in embryos. We used high content imaging to compare accumulation and toxicity of canonical substrates and inhibitors of the transporter, including fluorescent molecules and antimitotic cancer drugs, in homozygous knockout and wildtype embryos. To measure responses from the resulting image data, we used a nested convolutional neural network, which rapidly classified embryos according to fluorescence or cell division. This approach identified sea urchin embryos with 99.8% accuracy and determined two-cell and aberrant embryos with 96.3% and 89.1% accuracy, respectively. The results revealed that ABCB1 knockout embryos accumulated the transporter substrate calcein 3.09 times faster than wildtypes. Similarly, knockouts were 4.71 and 3.07 times more sensitive to the mitotic poisons vinblastine and taxol. This study paves the way for large scale pharmacological screens in the sea urchin embryo.
FIGURE 2. Phenotypes of F3 embryo and larval ABCB1 knockouts and wildtypes. (a) Two-cell knockout embryos accumulate more of the ABCB1 substrates CAM, BVER, and CROAM than wildtypes. Scale bar is 100 μm. (b) Six day old knockout larvae appear normal. Scale bar is 250 μm.
FIGURE 3. Schematic of the high-content imaging pipeline used in this study. Fertilized embryos (a) are transferred to a 96 well plate (b), and dosed using the Dragonfly robotic liquid handler (c). The Dragonfly dispenses variable volumes of stocks, giving the desired exposure concentrations directly in the plate (d). After an incubation period, treated samples are transferred to the ImageXpress HT. ai for image acquisition (e). Image datasets are then processed using convolutional neural networks (f). This produces data on each embryo (g).
FIGURE 4. ABCB1-knockout embryos accumulate more CAM than wildtypes. (a) Increasing fluorescence in embryos incubated in 250 nM CAM. (b) Change in fluorescence at 1 h, normalized to wildtype. Student’s t-test, ***indicates p value of <0.05.
FIGURE 5. Effect of adding an ABCB1 inhibitor on CAM accumulation in knockout and control embryos. Comparisons significantly different via 2-way ANOVA (p < 0.05) are indicated with a different letter grouping.
FIGURE 6. High content, machine learning based assay of the cell cycle effects of vinblastine on knockout embryos versus wildtypes. (a) dose–response curve of vinblastine for inducing aberrations in cell division. Points are means ± SEM of three experiments for wildtype and five experiments for knockouts. (b) Representative micrographs illustrating dose-dependent cell division aberrations. Scale bar is 100 μm.
FIGURE 7. High content, machine learning based assay of the cell cycle effects of taxol on knockout embryos versus wildtypes. (a) dose–response curve of taxol for inducing aberrations in cell division. Points are means ± SEM of three experiments for wildtype and five experiments for knockouts. (b) Representative micrographs illustrating dose-dependent cell division aberrations. Scale bar is 100 μm.
FIGURE 8. Examples of embryo phenotypes scored by the machine learning model. Annotations made by humans (a, b) are comparable to those achieved through our machine learning approach (a’, b’). In all images, the number associated with each embryo indicates the number of cell divisions that embryo is scored for having. Note the high level of agreement between the approaches, with occasional miscategorization by machine learning. Embryos annotated with a number only are considered normal for that cell stage, and a number appended with “ab” are considered abnormal.
