Inter-gastruloid heterogeneity revealed by single cell transcriptomics time course: implications for organoid based perturbation studies

Recent advances in organoid and genome editing technologies are allowing for perturbation experiments at an unprecedented scale. However, before doing such experiments it is important to understand the gene expression profile in each of the organoid's cells, as well as how much heterogeneity there is between individual organoids. Here we characterise an organoid model of mouse gastrulation called gastruloids using single cell RNA-sequencing of individual organoids at half-day intervals between day 3 and day 5 of differentiation (roughly corresponding to E6.5-E8.75 in vivo). Our study reveals multiple differentiation trajectories that have hitherto not been characterised in gastruloids. Intriguingly, we observe that individual gastruloids displayed a strong bias towards producing either mesodermal (largely somitic) or ectodermal (specifically neural) cell types. This bifurcation is already seen at the earliest sampled time point, and is characterised by increased activity of WNT-associated pathways in mesodermally-biased gastruloids as compared to neurally-biased gastruloids. Notably, at day 5, mesodermal gastruloids show an increase in the proportion of neural cells, while neural gastruloids do not produce notably more mesodermal cells. This is in line with previous studies on how the balance between these cell types is regulated. We demonstrate using in silico simulations that without proper understanding of the inter-organoid heterogeneity, perturbation experiments have either very high false positive or negative rates, depending on the statistical model used. Thus in future studies, modelling of inter-organoid heterogeneity will be crucial when designing organoid-based perturbation studies. Overall design: MULTI-seq (10X + individual gastruloid barcoding) for gastruloids between day 3 and day 5 of differentiation at half day intervals. A total of 4 experiments were performed. In the first experiment, day 3 and day 4 gastruloids were sequenced using 10X without MULTI-seq. The first MULTI-seq experiment involved 24 day 5 gastruloids spread over 3 10X lanes. The second MULTI-seq experiment involved 3 pools of day 3 and day 3.5 gastruloids (“exp2A_d3_d3.5”, “exp2C_d3_d3.5”, “exp2D_d3_d3.5”; note pools B and D were not sent for 10X sequencing), as well as 3 pools of day 4 and day 4.5 gastruloids (“exp3A_d4_d4.5”, “exp3B_d4_d4.5”, “exp3C_d4_d4.5”). Unfortunately, the day 3 and day 3.5 gastruloid data quality was too poor and these data were excluded in downstream processinng. For the day 4 and day 4.5 gastruloids, out of the 3 pools (pool A, B, and C), only pools B and C were sequenced, and from these only lanes (samples) 2 and 3 from pool B, and lanes 1 and 2 from pool C. The third experiment involved 1 pool of 24 day 3 gastruloids, spread across 2 10X lanes (“exp4_d3”), as well as one pool of 16 day 3.5 and 16 day 4 gastruloids, spread across 4 10X lanes (“exp5_d3.5_d4”), and finally 16 day 4.5 and 16 day 5 gastruloids, spread across 3 10X lanes (“exp6_d4.5_d5”). The MULTI-seq barcodes for this third experiment were additionally resequenced at higher depth and using 16 PCR amplification cycles during library preparation due to low yield.