THE ORIGINS OF ROBUSTNESS1

The endoparasitic wasp Copidosoma floridanum has a bizarre ontogeny. The tiny female wasp deposits a single egg into the egg of a moth host. The moth egg hatches, and develops into a larva that grows and develops until pupation. As expected, infected larvae produce wasps and not moths, but unexpectedly it is not a single wasp, but more than a thousand adult wasps that appear from a moth larva that was originally infected by a single wasp egg. This phenomenon is due to polyembryony, the division of the embryo into a number of clonal replicates that each develop as a separate individual. It is easy to understand why polyembryony is a good strategy for a tiny parasite infecting a much larger host. The mother wasp is too small to produce a large number of eggs, and the polyembryony provides a clever way around this constraint, which allows her to use all the resources provided by the moth larva. In fact, polyembryony has evolved multiple times among parasitic wasps (Grbic 2003). Although the evolutionary biologist may see this as yet another clever strategy forged by natural selection, the developmental biologist may see reasons for surprise. After all, the textbook version of insect development describes a complex series of orchestrated events that sets up the morphogenetic pattern of the insect embryo. And the general notion, captured in von Baer's laws, is that early development is highly conservative and acts as a constraint on later diversification. In Drosophila, as well as in wasps related to Copidosoma, the embryo develops from a large yolky egg, where maternal prepatterning is essential in initializing the body axis, and where a sequence of expression of gene sets takes this as an input to generate the familiar compartmentalization of the embryo. The early parts of this sequence even take place in a syncytium before cellularization. The patterning can thus depend on diffusion and does not involve cell-cell communication. All this must change in Copidosoma development. The egg lacks yolk, cell division starts immediately, and maternal positional effects can not specify the axis of development. Still, the adult wasps emerge with the same stereotyped body as other insects, and their adult morphology is closely homologous to other wasps that do not share their unusual development. To Andreas Wagner, the author of Robustness and Evolvability in Living Systems, this is but one of many illustrations of the striking robustness of biological systems. Despite huge differences in early development, the developmental process somehow manages to produce the same outcome. Another telling insight into the robustness of insect development comes from the work of von Dassow et al. (2000), who showed that one step of the gene cascade that patterns the embryo, the segment-polarity genes, in fact constitute a module, the function of which, producing sharply striped patterns of gene expression, is strikingly robust to variation in its parameters and input conditions. Perhaps such robust modularity is what allows development to produce the same outcome regardless of starting point. I do not want to give the impression that Wagner's discussion is limited to development, which constitutes a minor part of his book. His thesis is that robustness is a ubiquitous property of living systems in general, and he attempts to document this on many levels of organization. He starts out with the genetic alphabet, and argues that the G-C and A-T base pairs are more robust to replication error than are other possible Watson-Crick base pairs. He follows with the genetic code, which he argues is robust to point mutations. It is not just that the code is redundant; when mutations cause changes in amino acids, it is usually to other amino acids that are similar in their physio-chemical properties. Wagner shows that, within the distribution of possible codes, the actual code is far into the tail of extreme robustness towards chemical alteration. He then works his way through RNA secondary structure, which he argues is robust to nucleotide changes, and protein structure, which he argues is robust to point mutations and recombination. In the next set of chapters, he goes through gene expression, which is robust to changes in regulatory elements; metabolic flux, which is robust to changes in enzyme activity; metabolic networks, which are robust to changes in, and even elimination of, individual reactions; and gene-regulatory networks, which are robust to a variety of changes in interaction strengths, gene copy number, input from upstream genes, etc. He then moves on to the organismal level, where he argues that developmental pathways are robust to both genetic and environmental variation, and finally that the adult body plan is robust to variation in development. There is, however, little or no discussion of physiological and behavioral sources of robustness.