86 - Computational Anatomy for Fetal Phenotype Discovery in Lethal and Sub-viable Knock-out Mice

Poster Board Number: 86
There are separate poster presentation times for odd and even posters.
Odd poster #s – first hour
Even poster #s – second hour

Abstract Body : Phenotypic investigations of genetically modified animals typically focus on gross morphological phenotypes and overt structural defects, as they are easy to identify under the microscope in traditional screens. Progress in the areas of 3D imaging, data-sharing, and open-access analytical resources are creating new opportunities for phenotype discovery using computational methods, which we call computational anatomy (CA). CA allows more in-depth analysis and detection of more subtle phenotypes that otherwise may go unnoticed. The International Mouse Phenotyping Consortium (IMPC) and Knock-Out Mouse Phenotyping Program (KOMP2) aim to characterize 20,000 genes in the mouse genome. A subset of these knockouts, those with lethal and sub-viable mutations, are micro-CT scanned within a standardized framework and the data have been publicly shared (www.mousephenotype.org). Most of these lines are bred under experimentally identical conditions and were collected at either E15.5 or E18.5 and imaged using the same protocol (Dickinson et al., 2016, doi:10.1038/nature19356). We wanted to use this large dataset to test the feasibility of developing fast, high-throughput computational approaches for phenotype discovery. We hypothesized that if the phenotypic disruption caused by the knocked out gene is consistent, samples belonging to the same line should cluster more closely with each other in morphospace. If there are no systematic effects, samples belonging to the same line would be more dispersed or cluster with wildtypes. To this end, we have selected 114 E15.5 fetuses that represent wildtype (WT; N=20) and 14 knockout (KO) lines (N=94) and downloaded their diceCT scans from IMPC data portal. We used the Advance symmetric Normalization Tools package in R (ANTsR) to do a multi-step image registration. First, fetuses were rigidly aligned to the published anatomical atlas to create a common coordinate system and then were registered deformably with the atlas. The output of deformable registration is a warp field that maps voxels from the subject to the atlas. Principal component analysis (PCA) of total warp fields (a combination of both registration steps) belonging to 114 samples allowed us to ordinate each sample within a morphospace. Our current results showed that fetuses belonging to the same genotype cluster together in morphospace and are readily distinguishable from wildtype fetuses. Within the morphospace, we observe a distinct cluster of KO lines (Pcsk5, Psph, Rab34, Tmem132a) that display morphological characteristics of earlier developmental stages, suggesting a possible global developmental delay relative to the WT and other KO lines. In two of these lineages, Tmem132a and Rab34, growth retardation or abnormal embryonic size at E15.5 was previously unreported. Our preliminary results indicate that CA in KO mouse lines is a feasible approach for phenotypic discovery. In the future, we plan to scale our analysis to include to more KO lines. We will also refine it with region-specific analysis to detect and visualize localized differences in specific organs and body regions.