The LDSB investigates the organization and activities of developmental regulatory networks using formation of the Drosophila embryonic heart and body wall muscles as a model system. To achieve this objective, we combine contemporary genome-wide experimental and computational approaches with classical genetics and embryology to generate mechanistic hypotheses that we then test at single cell resolution in the intact organism. The cells comprising the Drosophila heart can be subdivided into two general populations, the cardial cells (CCs) which express muscle genes and are contractile, and the pericardial cells (PCs) which perform nephrocytic functions. To uncover the sets of enhancers and both the shared and uniquely discriminating sequence features which characterize the cell-specific activities of the regulatory elements that function in these two cardiac cell types, we modified a machine learning approach that we previously applied to identify somatic myoblast subtypes. This strategy enabled us to computationally classify cell-type specific cardiac enhancers based on the integration of multiple independent datasets that are relevant to such cell-restricted regulatory element activities, thereby defining the sequence features of enhancers that are critical for cell-specific functions. The experiments leading to these conclusions about the structure of cell-specific cardiogenic transcriptional regulatory networks involved the use of discriminative training methods to uncover the chromatin, TF binding and additional sequence features of enhancers that drive gene expression in individual cardiac cells. In particular, we used such a machine learning approach with TF motifs as well as with both chromatin immunoprecipitation (ChIP) data for a core set of cardiogenic TFs and for associated histone modifications to classify Drosophila cell-type-specific cardiac enhancer activities. This approach enabled us to generate classifier models that can be experimentally tested in transgenic reporter assays to validate their predicted cardiac cell subtype cis-regulatory activities. Furthermore, comparing the predicted enhancers with an expression atlas of cardiac genes uncovered gene clusters whose members have transcriptional profiles and biological functions that are limited to individual cardiac cell subtypes. In addition, the cell-specific enhancer models derived from these studies revealed chromatin, TF binding and other sequence features that distinguish enhancer activities in particular subsets of heart cells. Collectively, these results show that computational modeling combined with empirical testing provides a powerful platform to uncover the enhancers, the known TF motifs that are present in these enhancers, additional sequence features that may correspond to as yet uncharacterized TF binding sites, and gene expression profiles that govern individual cardiac cell fates. These research strategies and their corresponding findings differ from similar approaches used in other developmental systems since they focus on discovering not only the general features of the regulatory networks that are responsible for the specification and differentiation of shared organ-specific properties, but they also reveal the unique characteristics of those networks that define cellular subtypes within a single organ. For example, the developing CCs and PCs of the Drosophila heart have both overlapping gene expression patterns and associated cardiac properties as well as features that distinguish these two cell types from each other, the latter being most evident as differentiation progresses. At an even more refined level, the identities of individual CCs and PCs can be divided into subtypes based on differences in morphology, function, position within the heart and gene expression profiles. Our studies delineate the enhancer signatures that enable the subtle distinctions in these cardiac cell phenotypic identities to be generated during heart development, and in vivo testing of many of these predicted sequence features confirmed their importance for proper cardiac enhancer function. Taken together, this work has uncovered a much larger number of TFs that participate in the complex combinatorial regulation that is necessary for normal heart formation than had previously been appreciated. As a specific example of this regulatory mechanism, the present project uncovered a set of enhancers that are active in all PCs and/or in all CCs, as well as other enhancers whose functions are restricted to smaller subsets of these two cell types. Moreover, in a statistically significant large number of cases, the novel enhancers predicted by our machine learning models are associated with known heart genes present in the cardiac gene expression atlas that we independently generated. Our computational studies that included genome-wide histone marks present in the early developing heart also uncovered particular chromatin features that are enriched in enhancers having restricted heart cell activities, a finding that may contribute to how different cardiac cell subtypes emerge as heart development proceeds. These results were similar to those obtained for particular TFs that distinguish enhancer activities in different subsets of cardiac cells at various stages of cardiogenesis. Lastly, hierarchical clustering revealed previously unknown sequence features that discriminate enhancer activities in individual cardiac cells, findings that were empirically verified for a set of unique sequence features using cis mutagenesis in transgenic reporter assays. Collectively, these results establish multiple mechanisms that likely contribute in a combinatorial manner to the cell type-specific restriction of cardiac enhancer activities. The enhancer predictions from the individual cell-specific classifications also enabled us to predict in a more refined manner the expression patterns of known cardiac genes. Combining this information with gene ontology analysis then permitted us to infer the functions of individual heart cells. For example, genes associated with enhancers predicted to be active in all heart cells were associated with developmental, signaling and transcriptional functions, implying that these genes play a critical role upstream in the regulatory network that specifies the overall cardiac lineage. In contrast, genes predicted to be expressed in all CCs were enriched for myogenic functions including cell adhesion and the actin cytoskeleton, consistent with the known contractility of CCs, whereas genes associated with all PC enhancers aligned with renal system development, as expected for the nephrocytic activity of PCs. More specialized functions of individual cardiac cell subtypes also emerged from gene ontology analysis of the gene expression findings derived from our enhancer activity data. For example, one subset of PCs was enriched for chemotaxis and locomotion functions, suggesting that these cells are responsive to migratory cues, an important mechanism involved in heart tube formation during cardiogenesis. Taken together, these results confirm that modeling of cell type-specific enhancer activities can be used both to validate and to infer previously uncharacterized functions of individual cardiac cells. In summary, the present project documents the utility of computational modeling combined with empirical testing to uncover the enhancers, TF motifs, particular histone marks, and gene expression patterns that characterize individual cardiac cell fates. This work also provides a framework for conducting similar analyses in additional cell types and model organisms.
