Quantitative Spatial and Temporal Pattern Classification of Embryonic Stem Cell Aggregate Differentiation

Pluripotent embryonic stem cells (ESCs) can differentiate into the three germ lineages, making them an intriguing cell source for regenerative medicine therapies and drug screening. However, in order to precisely control differentiation it is still necessary to investigate the mechanisms governing cell fate decisions. One method of studying differentiation is to use aggregates of ESCs called embryoid bodies (EBs) which are capable of spontaneously differentiating into all three germ layers. Several groups have reported the spontaneous formation of spatially organized, complex tissue structures [1-3] in EB systems highlighting the importance of understanding the inherent emergent behavior capable of generating such complexity. We have recently described the creation of a computational model to simulate the 3D differentiation of EBs [4], but quantitative descriptions of the multicellular pattern formation in biological systems remain lacking.

Others have attempted to quantify spatial patterns using machine learning approaches [5] in combination with digital reconstruction algorithms [6]. However, such approaches are limited to specific systems of interest, or lack the necessary resolution to make claims about single cell spatial pattern dynamics. Here we describe the novel use of spatial networks for capturing single cell spatio-temporal heterogeneity associated with complex biological pattern formation. By utilizing network theory, various classifiers such as node counts, connections counts or novel user defined metrics, can be evaluated to create rich data sets for deriving predictive trends that can be used in conjunction with computational modeling approaches for model validation.

Formation and culture of murine ESCs (D3 line) into EBs was performed as previously described for two initial cell seeding densities (250 and 1000 cells/well) [4]. EB?s were fixed with formalin, counterstained with Hoechst and Phalloidin Alexflour 546 and immunostained for the pluripotency transcription factor Oct4. Stained EB?s were imaged by confocal microscopy and analyzed via Cell Profiler software coupled with a network reconstruction algorithm. Computational modeling was performed as previously described [4].

During EB differentiation, four types of multicellular patterns were observed: random, inside-out, outside-in, and connected (Fig 1A). To classify these patterns quantitatively, various metrics were validated by generating in silico patterns (Fig 1B). Using principal component analysis (PCA), we showed that the patterns could be correctly segregated with these metrics (Fig 1C). Furthermore, classification using machine learning approaches indicated patterns could be correctly classified with >98% accuracy (Fig 1D) suggesting that these metrics were capable of capturing pattern heterogeneity. Next we applied these metrics to our differentiating EB system [4]. PCA depicted a distinct differentiation trajectory that was independent of EB size (Fig 1E), which was further confirmed by pattern classification performed on these trajectories (Fig 1F). Additionally these patterns matched spatial patterns extracted from computational modeling traces as well (Fig 1G).

Subtle, yet distinct, spatial patterns were distinguished in the in silico training set utilizing network metrics which subsequently classified spatial patterns in both computational and experimental systems. These results highlight the power of our approach for quantitatively describing pattern trajectories during morphogenesis in a manner which allows direct comparison with complex computational modeling strategies.

References:

1.    Self-organizing optic cup morphogenesis in three dimensional culture. Eiraku M, Takata N, Ishibashi H, Kawada M, Sakakura E, Okuda S, Sekiguchi K, Adachi T, Sasai Y. Nature. 2011 Apr;472(7341):51-6.

2.    Generation of functional thyroid from embryonic stem cells.  Antonica F, Kasprzyk DF, Opitz R, Iacovine M, Liao XH, Dumitrescu AM, Refetoff S, Peremans K, Manto M, Kyba M, Costagliola S. Nature. 2012 Nov1;491(7422):66-71.

3.    Self-formation of functional adenohypophysis in three-dimensional culture. Suga H, Kadoshima T, Minaguchi M, Ohgushi M, Soen M, Nakano T, Takata N, Wataya T, Muguruma K, Miyoshi H, Yonemura S, Oiso Y, Sasai Y. Nature 2011 Nov 9;480(7375):57-62.

4.    Spatial pattern dynamics of 3D stem cell loss of pluripotency via rules-based computational modeling. White DE, Kinney MA, McDevitt TC, Kemp ML. PLoS Comput Biol. 2013;9(3).

5.       Towards a digital model of zebrafish embryogenesis. Integration of cell tracking and gene expression quantification. Castro-Gonzalez, C., M. A. Luengo-Oroz, et al. Conf Proc IEEE Eng Med Biol Soc 2010: 5520-5523.

6.       Imaging morphogenesis: technological advances and biological insights. Keller, P. J.  Science 340(6137): 1234168.