Optimization and scaling of patient-derived brain organoids uncovers deep phenotypes of disease

Abstract

Cerebral organoids provide unparalleled access to human brain development in vitro. However, variability induced by current culture methodologies precludes using organoids as robust disease models. To address this, we developed an automated Organoid Culture and Assay (ORCA) system to support longitudinal unbiased phenotyping of organoids at scale across multiple patient lines. We then characterized organoid variability using novel machine learning methods and found that the contribution of donor, clone, and batch is significant and remarkably consistent over gene expression, morphology, and cell-type composition. Next, we performed multi-factorial protocol optimization, producing a directed forebrain protocol compatible with 96-well culture that exhibits low variability while preserving tissue complexity. Finally, we used ORCA to study tuberous sclerosis, a disease with known genetics but poorly representative animal models. For the first time, we report highly reproducible early morphological and molecular signatures of disease in heterozygous TSC+/− forebrain organoids, demonstrating the benefit of a scaled organoid system for phenotype discovery in human disease models.

Hierarchical confounder discovery in the experiment–machine learning cycle

“…Next, we applied RTG analysis to our own multi-phenotypic dataset collected from human iPSC-derived brain organoids (Shah et al 2020). This data consists of three kinds of measurements: gene expression via quantitative polymerase chain reaction (qPCR), morphology via brightfield microscopy, and cell type distribution via single cell RNA sequencing.…”

“…RTG scores when evaluating different aspects of our previously published multimodal dataset (Shah et al2020). For each modality, potential confounders batch, donor, and clone are evaluated, as well as the intersections of batch plus donor and batch plus clone.…”

“…Strikingly, we see that the RTG scores for gene expression, imaging, and cell type distribution (Figures 6a–c) are consistent across modalities (average pairwise Spearman’s rank correlation of 0.924). These results suggest that there is a common biological origin of these phenotypic modalities; thus, the use of cost-efficient, scalable modalities (e.g., imaging) for the characterization of variability may be sufficient without the need to scale-up highly resource intensive techniques such as single cell sequencing (Shah et al 2020). This demonstrates another use for confounder analysis in the experiment–analysis cycle: in addition to informing how to balance future data collection (e.g., more clones), RTG scores can also inform resource allocation toward specific types of assays.…”

“…We furthermore show that linear techniques fail to discover these confounding effects. Next, we compare RTG analyses of a multi-modal dataset from patient-derived iPSC cultures (Shah et al 2020) to interrogate the effects of donor, clone, and batch as well as their interactions. We show that these potential confounders differentially bias the data, but that their relative effects are conserved across three highly disparate modalities of biological measurement: quantitative PCR, brightfield microscopy, and single cell RNA sequencing.…”

Tools and approaches for analyzing the role of mitochondria in health, development and disease using human cerebral organoids

“…Kegeles et al (2020) utilized convolutional neural networks (CNN) to predict early retinal differentiation. Shah et al (2020) trained a CNN to reveal morphological and molecular signatures of disease of heterozygous tuberous sclerosis TSC ± forebrain organoids. Artificial intelligence-based methods could also be recruited for example, for relating mitochondrial structural features with the construct's genetic information.…”