High-throughput hyperdimensional vertebrate phenotyping

Abstract

Most gene mutations and biologically active molecules cause complex responses in animals that cannot be predicted by cell culture models. Yet animal studies remain too slow and their analyses are often limited to only a few readouts. Here we demonstrate high-throughput optical projection tomography with micrometre resolution and hyperdimensional screening of entire vertebrates in tens of seconds using a simple fluidic system. Hundreds of independent morphological features and complex phenotypes are automatically captured in three dimensions with unprecedented speed and detail in semitransparent zebrafish larvae. By clustering quantitative phenotypic signatures, we can detect and classify even subtle alterations in many biological processes simultaneously. We term our approach hyperdimensional in vivo phenotyping. To illustrate the power of hyperdimensional in vivo phenotyping, we have analysed the effects of several classes of teratogens on cartilage formation using 200 independent morphological measurements, and identified similarities and differences that correlate well with their known mechanisms of actions in mammals.

Figure 1. High-throughput optical projection tomography

(a) Tomographic reconstruction of the craniofacial cartilage after image processing. Insets show details of the hypobranchial bone. The resolution of the 3D reconstruction is sufficient to distinguish bones with dimensions of ~10 µm and gaps of ~5 µm between bones. (b) Schematic representation of the OPT platform. A fluidic system composed of a pump, valves, and tubing is used to load larvae from multiwell plates into an glass capillary located within the FOV of a macrozoom microscope. The capillary is held at two ends by stepper motors with hollow shafts that sit on a position stage. Several dynamic image artifact correction steps are performed prior to reconstruction. (c) Non-uniformity of illumination due to the capillary walls is corrected. (d) The light path through the capillary is simulated and is fed to the reconstruction algorithm to correct for the refraction of light by the capillary. (e) Small movements of the larva along the capillary axis and radial vibrations of the capillary are detected and corrected. The box-plots quantify the wobbling of the capillary (top plot) and the motion of the larva along the capillary axis (bottom plot) prior to correction. (f) Identification of center of rotation (COR) for reconstruction. Reconstructing a cross-section using different COR estimates allows us to identify the true rotation axis with micron precision. The bottom panels show reconstructions using two different CORs. The reconstruction on the left corresponds to a COR estimated solely from the boundaries of the capillary sidewalls. The reconstruction on the right has sharper boundaries and lower entropy, and is therefore selected as the best estimate for true COR. Graphs show 1D intensity profiles across the dashed lines. Entropy of each image is shown on the bottom right corner. (g) Each step of image artifact correction produces increasingly accurate tomographic reconstructions. Images show 3D reconstructions from 320 images of a 5 dpf zebrafish larvae stained with Alcian blue. Insets show zoomed views of the most posterior region of the basihyal cartilage (red).

Figure 2. Quantification of complex skeletal phenotypes induced by teratogenic chemicals

(a) Three-dimensional reconstructions allow visualization of complex phenotypes. Images show representative tomographic renderings of fully segmented craniofacial skeletons from 5 dpf larvae treated with three different teratogens: retinoid acid, cyclopamine, and flusilazole. Representative reconstructions are shown for all chemicals at low, medium, and high doses (relative to the calculated LC₅₀) from left to right. (b) Quantitative analysis of the ethmoidal plate (orange), trabeculae (pink), and Meckel's cartilate (blue) in larvae treated with increasing concentrations of cyclopamine. Phenotypes are concentration-dependent and not all bones are altered in the same manner or to the same degree. (c) Quantitative analysis of the trabeculae in larvae treated with increasing concentrations of all-trans retinoic acid (atRA; blue), cyclopamine (orange), and flusilazole (red). (d) Ratiometric analysis of the relationship between the ethmoidal plate and the ceratohyal cartilage in larvae treated with increasing concentrations of all-trans retinoic acid (atRA; blue), cyclopamine (orange), and flusilazole (red). Taken together, panels (b) through (d) show that individual metrics reveal distinct phenotypic changes in response to different chemicals and exhibit distinct dependencies on chemical concentrations. (e) Representative reconstructions of the samples used to generate the data points in (b–d). Bones colors match corresponding data points. For reconstructions of (c) and (d) green bones indicate untreated (WT) controls (n=4–6 for each treatment condition, n=53 for WT, ± s.e.m).

Figure 3. Hyperdimensional in vivo phenotyping

(a) Schematic representation of the analysis process. Following 3D reconstruction (left) Individual bones are segmented and separated by bone type (center). Morphological image analysis is then used to compute metrics for each bone segment (e.g. volume, length, angles, surface area, and location). These values are stored as a hyperdimensional phenotypic signature for each larva (right). (b) Phenotypic signature array (shown log 10 scale of p-values). Each row represents individual larvae treated with a compound. Measurements are color-coded to indicate a reduction (red) or an increase (green) with respect to untreated controls. Phenotype metrics and compound treated larvae are ordered by hierarchical clustering based on correlation. (c) Dendrogram computed on linkage by correlation of the mean standard score for chemicals applied beginning at 9 hpf. Teratogens known to cause developmental bone abnormalities are grouped based on similarities in their hyperdimensional phenotype signature. Retinol and the retinoids (red) and the triazole derivatives (green) cluster together in groups that are distinct from cyclopamine (a steroidal alkaloid).