During the project, we investigated how reverse causal effects influence the reliability with which individual cells make decisions. Using highly multiplexed measurements of signaling responses together with markers of cellular state in human cells, we quantified how signaling networks process information in heterogeneous cellular contexts. We discovered that signaling nodes display adaptive information-processing properties that generate heterogeneous growth factor responses while capturing partially non-redundant information about cellular state. Collectively, these activities form a multimodal percept enabling cells to interpret growth factor concentrations within the context of their internal state and make state-dependent decisions. These findings indicate that heterogeneity and complexity in signaling networks are not merely noise but represent evolved strategies enabling context-aware cellular decision-making in multicellular environments. This conceptual advance was published in Science (Kramer et al., 2022; PMID: 35857483).
A second achievement was the development of computational frameworks to quantify spatial organization within cells across experimental conditions. Because cellular organization is inherently multi-scale and heterogeneous, conventional approaches struggle to capture consistent spatial patterns across cell populations and perturbations. We therefore developed the deep-learning framework CAMPA (Conditional Autoencoder for Multiplexed Pixel Analysis), which learns representations of multiplexed molecular pixel profiles consistent across heterogeneous cell populations and perturbations. Clustering these representations identifies reproducible subcellular landmarks that can be quantitatively compared with respect to size, composition, and spatial organization. Application of CAMPA to multiplexed immunofluorescence data revealed how subcellular organization changes upon perturbations of RNA synthesis, RNA processing, and cell size, and uncovered links between membraneless organelle composition and variability in RNA synthesis rates. The method enables systematic comparisons of spatial cellular organization and has been released as open-source software. This work was published in Nature Methods (Spitzer et al., 2023; PMID: 37248388).
Building on these advances, the project extended the analysis of multi-scale organization to multicellular systems grown in vitro. We investigated retinal organoids derived from human induced pluripotent stem cells that recapitulate aspects of human retinal development. Multiplexed protein maps across a developmental time course were compared with adult human retinal tissue. Computational tools were developed to visualize progenitor and neuronal distributions, extracellular and subcellular organization, and tissue patterning. In parallel, single-cell transcriptomic and chromatin accessibility datasets were generated and used to infer gene regulatory networks. Integrating genomic data with spatially segmented nuclei enabled construction of a multimodal atlas of retinal organoid development, revealing spatial neighborhoods of retinal ganglion cells and pathways associated with their cell death. This work was published in Nature Biotechnology (Wahle et al., 2023; PMID: 37156914). Overall, the project delivered conceptual, technological, and computational advances enabling quantitative study of biological organization across spatial scales.
