Structure of developmental gene regulatory networks from the perspective of cell fate-determining genes

ABSTRACT

The core of gene regulatory networks (GRNs) is formed by transcription factors (TF) and cis–regulatory modules (CRMs) present in their downstream genes. GRNs have a modular structure in which complex circuitries link TFs to CRMs to generate specific transcriptional outputs.¹ Of particular interest are those GRNs including cell fate-determining genes, as they constitute developmental switches which activity is necessary and sufficient to promote particular cellular fates. Most of the genetic analysis of developmental processes deals with the composition and structure of GRNs acting upstream of cell fate-determining genes, as they are best suited for genetic analysis and molecular deconstruction. More recently, the application of a variety of in vivo, computational and genome-wide approaches is allowing the identification and functional analysis of GRNs acting downstream of cell fate-determining genes. In this review we discuss several examples of GRNs acting upstream and downstream of cell fate-determining genes, including other TFs which activity pervade across both regulatory networks.

Arguably, the essence of multicellular development is the ordered generation of spatio-temporal patterns of gene expression.¹ This principle is most clearly manifested in processes that we can group under the definition of “pattern formation,” a developmental operation that distributes cellular fates within fields of cells with the same initial identity or developmental potential. Pattern formation consists in a progressive acquisition of differences in gene expression, culminating in the ON/OFF transcriptional state of “cell fate-determining genes.” Genetic approaches have been instrumental and successful identifying cell fate-determining genes and their upstream regulators, to a large extent because these genes usually display complementary loss- and gain-of-expression phenotypes that can be unequivocally associated with the commitment of cells toward a particular fate. In contrast, the reconstruction of cell differentiation cascades operating downstream of cell fate-determining genes is less amenable to genetic analysis, perhaps because the contribution of individual components is small and pleiotropic, and therefore difficult to asses phenotypically. In this paper we aim to describe the similarities and differences between GRNs acting upstream and downstream of cell fate-determinant genes. We will explain the characteristics of these GRNs by describing several examples taken from Drosophila developmental genetics. We will argue that the logic structures of upstream and downstream GRNs do not need to be radically different, as they are both based on combinatorial regulatory operations. The difficulty in dissecting GRNs acting downstream of cell fate-determining genes most likely is due to the multiplicity of molecular functions encoded by their components, and to their small or redundant phenotypic contributions.

Cell fate-determining genes directly regulate the commitment and differentiation of a given cell or group of cells, and they are key components of developmental programs. These genes encode TFs, and likely they operate by regulating an undetermined number of cell differentiation genes. In general, cell fate-determining genes have complex regulatory regions with a modular organization that integrate inputs from a large number of transcriptional regulators (Figs. 1A-B). In addition, cell fate-determining TFs act in combination with other transcriptional regulators that convey further variations to their response. A gene set that illustrates these characteristics is the Drosophila achaete-scute complex (AS-C), which is formed by 4 transcription units (achaete, scute, lethal of scute and asense) spanning 40 kb of DNA containing a complex array of cis-regulatory modules (CRMs) (Fig. 1C). The four genes of the AS-C encode TFs of the basic helix-loop-helix family, and their activities are required for the commitment of epithelial cells as neuronal precursors in the ectoderm, and also for the formation of muscle-founder cells in the mesoderm.² In essence, the AS-C constitutes a molecular device that: 1) integrates and responds to a complex landscape of TFs, which expression is restricted to particular spatial domains in the tissue of interest (Fig. 1B), and 2) promotes the commitment of the cells where its expression is retained toward a neuronal (ectoderm) or muscle (mesoderm) cell fate.^2,3 In this manner, upstream events to the AS-C are exclusively regulatory, and they can be best understood in terms of GRNs composed by transcription factors impinging on the AS-C CRMs (Fig. 1A). Genetic analysis was capable of dissecting the AS-C upstream GRNs, mostly because mutations in the relevant components (the AS-C regulatory TFs, the AS-C coding regions and their CRMs) result in changes in the spatial pattern of differentiated cells that could be correlated to topological changes in the expression the AS-C genes (Figs. 1D, E-E”’).

Figure 1.

(A) Schematic representation of the GRN regulating the ac and sc genes in the dorsocentral enhancer (DC; yellow box). The regulatory interactions mediated by the TFs Pnr and Ush and the Dpp and Wg signaling pathways are represented by green arrows (activation) and red bars (repression) (data modified from reference ²⁷). (B) Thoracic region of a third instar wing disc showing the expression of pnr (blue), iro (red) and sal (green) and the position of the anterior notopleural (ANP), DC and scutelar (SC) proneural clusters, indicated by blue, yellow and red dots, respectively. (C) Representation of the ac-sc gene complex showing the transcription units (ac, sc, l(sc) and ase; small black rectangles), the enhancers operating during the development of the DC, ANP and SC clusters (yellow, blue and red boxes, respectively) and other additional enhancers (gray boxes). (D) Adult Drosophila notum. (E-E”’) Top: schematic representation of left thorax showing the position of macrochaetae (large dots) and microchaetae (small dots). The DC, ANP and SC macrochaetae are represented by yellow, blue and red dots respectively. Bottom: modifications in the ac-s gene complex including loss of coding regions (E’), loss of DC enhancer (E”) and ectopic expression (E”’).

Another example taken from the patterning of the Drosophila larval cuticle illustrates a similar principle: complex GRNs converge on the CRM of cell fate-determining genes and convey positional information to sets of cells distributed in a precise spatial pattern (Figs. 2A-C). The ventral cuticle of the Drosophila larva is decorated with a segmental pattern of cells differentiating either smooth cuticle or cuticular projections named hairs or denticles.⁴ This distinction depends on Shaven-baby (Svb), a TF which activity is necessary and sufficient to promote denticle formation cell-autonomously.⁵ The segmental expression of svb is the culmination of a complex transcriptional cascade initiated during oogenesis with the expression of “cardinal” genes that progressively subdivide the developing embryo in smaller domains of TF coding genes expression (gap and pair rule) to finally set parasegmental boundaries of complementary Wingless (Wg) and Hedgehog (Hh) signaling (Fig. 2A).⁴ The parasegmental boundary, and the regulation of signaling range and polarity of the Wg and Hh ligands, is subsequently used to set the expression of ligands of the EGFR and Notch signaling pathways. Localized, and segmentally repeated, ligand expression subdivides each segment in signaling domains where the TFs activated by each pathway contribute to the regulation of the denticle-fate promoting gene svb (Fig. 2A). In this process, a transcriptional cascade divides the length of the ventral embryonic ectoderm into a periodic pattern of signaling domains that finally set the expression of cell fate-determining genes with segmental expression.⁴

Figure 2.

(A) Schematic representation of the blastoderm segmentation cascade. The first 3 rectangles illustrate the gradients of maternal genes (upper rectangle in red and blue), the expression of the Gap genes in broad domains (middle rectangle in 4 colors) and the expression of one pair-rule gene in 7 stripes (green). The left bottom rectangles represent from top to bottom parasegmental boundaries of wingless (wg) and hedgehog (Hh) expression, the subsequent expression of Rhomboid (Rho) and Serrate (Ser), and the segmental expression of svb. All drawings are modified from reference 4. The Hox genes domains of expression are represented to the right. (B) Schematic representation of different CRMs regulated by Svb and examples of the molecular function of Svb downstream genes promoting the specific behaviors of denticle-forming cells (data taken and modified from reference 8). (C) Transmission microscope view of a larval segment showing denticles (green stripe, svb domain) and smooth cuticle (orange stripe, wg domain). (D) Sand-clock or diabolo representation of regulatory cascades acting upstream (blue) and downstream (green) of cell fate determining genes (AS-C and sbv, red dot), and leading from patterning information to cell responses. The multilevel action of other transcriptional regulators, including the Sal and Hox proteins, is represented in parallel as “micromanagers” (gray arrows).

These two examples illustrate that cell fate-determining genes constitute key nodes converting pattern information into cell differentiation programs, a hierarchical position that has been referred to as “the beginning of the end.”⁶ What happens now downstream of cell fate-determining genes? In principle the role of these TFs is to engage in novel GRNs that must include as components batteries of cell differentiation genes (Fig. 2B). The logic structure of GRNs acting upstream and downstream of cell-fate determining genes does not need to be radically different. However, the identities of downstream components and their interrelationships are less well characterized. We would like to argue that this is so because the molecular functions of cell differentiation genes have an additional aspect that is more elusive to genetic analysis. Thus, cell differentiation genes encode proteins which functions translate gene expression into the modulation of cellular behaviors, such as cell growth, division and cytoskeleton organization, resulting in changes in cell shape, viability, patterns of division and so on. Because these behaviors are universal to all cells, it is expected that perturbations in these genes would result in less tractable phenotypes that might be more challenging to relate to the function of the corresponding proteins. In addition, the requirement of each individual cell differentiation gene might be small, difficulting the analysis of their contributions to the differentiation process. Genetic analysis is more efficient identifying genes involved in “commitments” (cell fate-determining genes) than in “modulation” of general cellular properties.

The case of Svb illustrates that cell differentiation GRNs are also susceptible to identification and molecular analysis, by using a clever combination of computational, transcriptomic and in vivo approaches.^7,8 Several interesting findings related to the GRNs operating downstream of Svb are worth mentioning. For example, Svb only regulates a subset of the genes expressed in the cells forming the denticles, indicating the existence of additional GRNs acting independently in these cells. Also, Svb target enhancers are built from different combinations of cis-regulatory motifs, showing that co-expressed genes mediating the same terminal cell differentiation can have diverse cis-regulatory architectures (Fig. 2B). The knowledge of Svb response elements helped in sorting functional binding sites from the total of about 6000 genomic sites bound by this TF.⁸ This is a still controverted aspect of many TF, for which chromatin immunoprecipitation (ChIP) experiments indicate a generalized binding to the DNA of unknown functional relevance.^9,10 Interestingly, global gene expression analysis in a svb loss of function background showed that the reduction in mRNA levels of validated Svb targets is often small. Widespread binding and relatively weak effects on the expression of a large fraction of target genes might be a general characteristic of cell fate-determining TFs that adds difficulty to the identification of their downstream genes by conventional genome-wide techniques. The identification of a large subset of Svb target genes is very relevant, because allows expanding the inventory of molecular functions involved in a cell differentiation program. In this particular case, Svb target genes encode proteins regulating F-actin organization, cuticle formation and pigmentation, extracellular matrix reorganization, enzymes involved in oxidation-reduction, proteolysis, cell trafficking and sugar binding, among others (Fig. 2B).⁸ Furthermore, the functional analysis of a subset of Svb target genes showed that the inactivation of each individual target cause specific defects in the morphology of denticles, but that only the simultaneous elimination of several targets leads to a failure in denticle formation.⁷ This observation implies that the collective action of many effectors is required to promote the changes in cell shape and cuticular differentiation associated with the formation of denticles. In essence, the genetic and molecular analysis of Svb downstream GRNs allows breaking the black box of differentiation: the group of cellular transformations and underlying biological operations occurring in differentiating cells under the regulation of cell fate-determining genes.

Not all genes involved in developmental processes fit in rigid hierarchies of cell fate-determining upstream or downstream components. A singular case is the Hox genes, a phylogenetically conserved family of gene complexes encoding transcription factors with homeodomain DNA binding motives.^11,12 Hox genes are expressed in broad domains during development (Fig. 2A), and they impose morphological differences to homologous structures that otherwise share the same developmental plan. Because of the systemic effects of mutations in Hox genes in serially homologous structures, such as Drosophila segments or appendages arising from different segments, it was argued that they sit on top of regulatory hierarchies, acting as “selectors” of developmental programs.¹³ The description of Hox genes as “selectors” was inferred from a phenotypic point of view, but the molecular analysis of their regulatory functions reveals a more complex scenario. Thus, Hox proteins present widespread binding across the genome, and their putative targets include genes that show clear-cut differences of expression comparing homologous structures,¹⁴ but another, more numerous, set of targets displaying only quantitative differences in expression levels.^14-16 In this way, Hox proteins can “regulate” ON/OFF transcriptional states, and also “modulate” quantitative differences in the expression levels of different target genes. Likely, the poor binding specificity of Hox proteins, in combination with their ability to interact with other regulators, confers them the versatility to influence the transcription of a large variety of targets, and consequently, the capacity to act at many levels of regulatory hierarchies. In this sense, it has been argued that molecularly Hox genes can be more accurately described as “micromanagers” more than “master regulatory” proteins.¹⁷

A similar example of genes that manage GRNs at multiple levels during development is the spalt family (sal). The Sal proteins are also phylogenetically conserved, and they contain several pairs of Zn fingers that can bind DNA as well as other protein-protein interaction domains.¹⁸ The function of Sal proteins is required in different organisms for a variety of developmental processes including cell fate specification, organogenesis, neural development and the generation of territorial differences within developing organs.¹⁸ The sal genes and proteins display multiple interactions with Hox proteins, acting upstream or downstream of Hox genes in different developmental settings.^19-23 In addition, they can also modulate the regulation of Hox target genes during vertebrate limb formation.²⁴ Apart from this multiplicity of Hox-Sal interactions, the sal genes share several characteristics with Hox genes that likely contribute to their versatility and context specific activities. Thus, Sal proteins bind with poor specificity to the DNA and display thousands of binding sites in ChIP experiments (MM, MFO and JFdC, unpublished results), and they regulate the expression of multiple target genes, acting on different targets as transcriptional activators or repressors.^25,26 Furthermore, and in a similar manner to Hox genes, the Sal proteins act in GRNs regulating pattern formation, and also in GRNs modulating the expression of genes controlling common cellular properties such as adhesion, division and epithelial morphogenesis.²⁵ In this sense, Sal proteins also work as “micromanagers” or “territorial coordinators” conferring cells subtle different developmental characteristics (Fig. 2D).

The study of how different TFs modulate the expression of their targets genes is a necessary step to understand both the logic structure of regulatory GRNs controlling pattern formation and the regulation of cell responses by downstream components. The distinction between cell fate-determining genes and territorial coordinators might be useful from a functional point of view, as they perform genetic operations that have different consequences on cell types or cell territories. In this manner, cell fate-determining genes operate on individual cells to open a precise differentiation program, such as “sensory organ precursor” or “muscle founder cell,” in a given developmental context. In contrast, territorial coordinators operate on cell populations and confer topological information such as “Haltere” (Hox) or “wing central region” (Sal). This distinction, however, does not imply any major qualitative difference from the perspective of the molecular operations they perform or the logic structure of the GRNs in which they participate. Thus, all these genes share the existence of complex and modular CRMs capable of reading previous pre-patterns of transcriptional regulators, and normally they encode TFs with variable specificity toward their target DNA and other co-factors. For these reasons, these proteins are capable of engaging multiple regulatory regions in the DNA, and also interact or behave as cofactors for a variety of transcriptional regulators. Most likely, it is this versatility of interactions what determines the potential to co-opt these transcriptional regulators into pre-existing GRNs, where they contribute to the implementation of context dependent information to generate variations in gene expression. A convenient description of developmental processes upstream and downstream of cell fate-determining genes has the geometry of a diabolo or sand clock, in which one sphere represent GRNs regulating the expression of a cell fate-determining gene and the other sphere the regulatory interactions and components acting downstream to implement cell behaviors (Fig. 2D). In this model, proteins such as the Hox or Sal TFs would act on a large set of both upstream and downstream genes, modulating gene expression across the entire hierarchy (Fig. 2D).

Disclosure of potential conflicts of interest

No potential conflicts of interest were disclosed.

Acknowledgments

We would like to thank Carlos Estella and Cristina M. Ostale for comments to the manuscript and Covadonga F. Hevia and Ernesto Sánchez-Herrero for Figures 1D and 2C, respectively. We would like to dedicate this contribution to the memory of Prof. Eric H. Davidson, a pioneer and most relevant contributor to the concept and analysis of GRNs.

Funding

This work was supported by institutional grants from Fundación Ramón Areces and Banco de Santander to the Centro de Biología Molecular Severo Ochoa and BFU2012–33994 grant to JFdC.