Built to Run, Not Fail

See allHide authors and affiliations

Science  16 Mar 2007:
Vol. 315, Issue 5818, pp. 1510-1511
DOI: 10.1126/science.1140979

On first encounter, gene regulatory networks for development often seem so complicated as to defy intuitive understanding. But the overall maze of gene interactions that they represent is actually composed of subcircuits that perform separate functions. The subcircuits are often of elegant and sometimes counterintuitive design, even more so, the ways they are combined in the overall network. As the underlying subcircuit structure is clarified, we see that gene regulatory networks in fact provide a direct and simply organized bridge from the phenomena of development to the detailed genomic programs that encode it. Among the most fascinating aspects of gene regulatory networks are their design principles, for these are often interestingly different from what would seem the “simplest” solution. Gene regulatory networks for development are the direct product of evolution, and the character of their design both illuminates evolution and is illuminated by it.

Each of the specific biological functions which together make up a developmental process is programmed by a specific subcircuit of the network. In other words, large gene regulatory networks have a modular structure: They are composed of different subcircuits that work together to accomplish whole “pieces” of development, such as specification of dorsoventral pattern in the fly embryo or of the endomesoderm territories of the sea urchin embryo. Overall, such gene regulatory networks involve scores of genes [>50 in these cases (1)] organized into many subcircuits, where a single subcircuit controls a specific developmental task. These tasks include specifying regulatory states of a group of cells (i.e., determining which regulatory genes they will express); mounting molecular signals that induce new regulatory states in recipient cells; coordinating the expression of genes that control cell differentiation; stabilizing newly established regulatory states; defining tissues and setting their boundaries; and interpreting prior regulatory instructions. There is a plethora of regulatory jobs different from one another—such as the development of embryos, or of stem cells, or of adult body parts—that all require different kinds of subcircuits. The subcircuit components of gene regulatory networks have evolved independently of one another, and at different rates (2), and are assembled in different contexts in related organisms. Both in their functional organization and in the separate evolutionary histories of their subcircuits, gene regulatory networks are modular in construction.

The individual subcircuits each consist of a few regulatory genes, including their genomic cis-regulatory information processors, which respond in a combinatorial and conditional manner to the transcription factors encoded by other genes of the same module. In considering structure-function causality in gene regulatory network subcircuits, the architecture of the module tells it all. The architecture is the design of the causal linkages between genes of the subcircuit. This is a hard-wired feature because it is constructed by the inherited cisregulatory control sequences of these genes. The biological function depends on the architecture. For example, positive cross-regulatory interactions among a set of genes that encode transcription factors can stabilize the particular regulatory state generated by these genes. As another example, it is the particular set of genes regulated by a given gene that is turned on in response to an inductive signal that determines what the developmental effect of the signal will be.

There are two essential consequences of this concept of a modular network architecture and subcircuit design. First, subcircuit architectures are as varied as the biological jobs they do. Thus, although subcircuits are indeed the modular functional components of developmental gene regulatory networks, they are to be distinguished from simpler “building blocks” or “motifs” that are used for many diverse developmental functions (e.g., feedforward or feedback elements, per se). For instance, feedforward motifs are to be found in every conceivable context in diverse gene regulatory networks (3), whereas the individually designed subcircuits here considered are specific to the type of biological job they do. Second, subcircuit architectures are built from diverse classes of transcription factor, and by and large, a given type of factor is not dedicated to any given type of subcircuit. In terms of logic outputs, circuits that transduce signals and distribute their outputs may operate very similarly, whatever the nature of the signaling system or the identity of the immediate early response factor. The same is true of cross-regulatory subcircuits. It is the genomic architecture of the subcircuit, and not the nature of the factors the genes encode or the families they belong to, that uniquely indicates subcircuit function [multiple examples of subcircuits from diverse developmental systems can be found in (3)].

As we have come to understand developmental gene regulatory networks, there arises an impression of “overlayered” circuit design—or more precisely, deployment of multiple subcircuits—that in different ways support the same end result. In development, the major regulatory task is to specify spatial domains of gene expression. Typically, multiple, distinct kinds of subcircuits are brought to bear in a given spatial specification process, all of which function to ensure the outcome once a unique set of regulatory genes is activated in a given spatial domain of an embryo.

Same design, different actors.

Common subcircuit architecture (center) in diverse instances of the exclusion effect. Specific organisms, and cell types or domains are shown, each specified by unique set of specification inputs. Horizontal lines represent cis-regulatory apparatus of the indicated gene. Genes shown in orange are transcriptional repressors, which are directly activated in the specification process (black arrow). Genes shown in magenta are major drivers of indicated alternative (excluded) cell fates.


First, the new regulatory state is locked down, by deployment of positive feedback or other cross-regulatory relationships among the regulatory genes. The lockdown is dynamic because it requires continuous transcription of the cross-regulated genes, but it acts to stabilize the regulatory state.

Second, the cells within the newly defined spatial domain are linked together by intercellular signaling, by use of subcircuits that make continuation of the regulatory state dependent on reception of the signal. These subcircuits use another kind of intercellular feedback in that the gene encoding the signal responds to its own signal transduction system. Thus, all cells of the domain both receive and emit that same signal—a “community effect.”

In addition to this, the same signal transduction system that promotes the regulatory state within the domain acts as an obligate repressor of genes that respond to it outside the domain. Signal transduction systems often have the Janus-like quality that they act positively in cells receiving the signal but otherwise behave as repressors (4, 5).

Finally, the specification apparatus very frequently also includes transcriptional repressors, which, within the specified spatial domain, target key regulatory genes whose expression is required for alternative regulatory states that could have been available to these cells. This is a so-called “exclusion effect,” and numerous examples can be found across species (see the figure). In each developmental case, the identity of the specific transcription factor that executes the repression is distinct, as are the specifically excluded target transcription factors. The design is the same, the biochemical actors diverse.

So it is not enough in an embryo just to arrange to turn on the right regulatory genes in the right place. These genes must also be dynamically locked on; the regulatory state of cells in a given spatial domain must further be made dependent on signaling among them all; the expression of these same regulatory genes must be specifically forbidden anywhere else; and then, on top of all that, specific alternative states must be excluded. These components are of course interlinked, and experimental tests of whether this is a “necessary” design are not simple. In the sea urchin embryo, where all of the above are to be found, disarming any one of these subcircuits produces some abnormality in expression.

We may interpret this as we like—as overengineering; or as design deluxe, replete with bells and whistles; or as the expected result of an evolutionary process in which individual regulatory modules have been added in and overlain at different times, so that some are more ancient and others more new (1). However, once integrated into the regulatory system, they are there to stay, barring evolutionary redirection. But the generality of this quality of developmental gene regulatory networks is emerging as a fact of life—it is what we see in modern animals. The consequences of evolutionary history determine the shape of the control apparatus that determines life processes. Perhaps in current system design we are seeing something of the grim pressures that modern lineages survived in past evolutionary bottlenecks—of the absolute necessity for lineage survival of genomic regulatory systems built to run and not to fail.

References and Notes

  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
View Abstract

Navigate This Article