Science 18 May 2001:Vol. 292. no. 5520, pp. 1315 - 1316DOI: 10.1126/science.1060852PERSPECTIVESMOLECULAR BIOLOGY AND EVOLUTION:Can Genes Explain Biological Complexity?Eörs Szathmáry, Ferenc Jordán, Csaba Pál*Although natural selection does not guarantee that organisms will increase in complexity as they evolve, it is apparent that the complexity of certain lineages, such as our own, has increased during evolution. Although we have an intuitive appreciation of biological complexity--often thinking in terms of morphological or behavioral complexity, or the variety of cell types in an organism--the term itself is notoriously hard to define. One could resort to algorithmic complexity, where the number of steps in the shortest possible algorithm that solves a given task has proven to be a convenient measure (1). In this case, complexity could be defined as the number of steps in the developmental program out of which the embryo is "computed." The snag here is that evolution is not an engineer but a tinkerer, so that there is no reason to expect that, for example, elephants have developed according to a minimalist program (2).Is the number of genes in an organism's genome an appropriate measure ofbiological complexity? It has been assumed that eukaryotes have more genes than bacteria, animals have more genes than plants, and vertebrates have more genes than invertebrates (2, 3)--which nicely fits with the traditional notion of a scala naturae. The recent flurry of completed genome sequences, including our own, suggests that this is not necessarily the case (4-6). Rather surprisingly, it turns out that the worm Caenorhabditis elegans has 18,424 genes in its genome, the fruit fly Drosophila melanogaster 13,601, the plant Arabidopsis about 25,498, and humans about 35,000. This suggests that there must be other, more sensible genomic measures of complexity than the mere number of genes.Transcription factors are DNA binding proteins that switch target genes onand off. For all transcription factor families, their members increase in number in the order yeast, nematode, fruit fly, human (7). The diversity of cell types in these organisms also increases in this order (5). This makes sense, given that maintaining the differentiated state of increasingly diverse cell ADVERTISEMENTADVERTISEMENTCurrent Issue Previous IssuesScience ExpressScience Products My ScienceAbout the JournalHome > Science Magazine > 18 May 2001 >Szathmáry et al. , pp. 1315 - 1316types requires the presence of more and more molecular switches (6). In commenting on the human genome sequence, Claverie has suggested that we define biological complexity in terms of the number of transcriptome states (a transcriptome being the complete set of RNA transcripts) that the genome of an organism can achieve (6). Following this line of thought, how, then, can one obtain a measure of true biological complexity?We propose that biological complexity might be better explained byconsidering networks of transcription factors and the genes they regulate, rather than by simply counting the number of genes or the number of interactions among genes. One could borrow indices from other fields that have an older tradition of quantifying networks. For instance, when trying to obtain a measure of ecosystem complexity, ecologists consider not only the number of species but also the types and numbers of interactions among them. For example, the complexity of interactions within a food web can be defined by the connectivity (C): C = 2 L/[N(N - 1)], where the number of actual trophic links (L) is divided by the number of all possible links, with N as the number of species. It would be intriguing to know whether gene-regulation networks in bacteria or eukaryotic cells can also be defined in terms of their connectivity (see the table). A global analysis of transcriptional regulation in the bacterium Escherichia coli reveals that on average each transcription factor regulates three genes, and that each gene is under the control of two transcription factors (8). Certainly the connectivity of gene-regulation networks in eukaryotes is likely to be greater than that in bacteria, but for now we lack a way to measure the magnitude of this difference.GENETIC NETWORKS AND BIOCOMPLEXITYIndex Scale RelevanceNumber of nodes, NGlobal Number of relevant genes in a genetic networkNumber of links, LGlobal Number of gene interactionsConnectivity, C = 2 L/[N(N-1)] Global Realized fraction of possible gene interactionsIn-degree, DinLocal Number of genes affecting a particular geneOut-degree, DoutLocal Number of genes affected by a particular geneDegree, DLocal The number of genes directly interacting with a particular geneAverage degree, DavGlobal Average number of gene interactions per geneTo Advertise Find ProductsADVERTISEMENTFEATURED JOBSFACULTY SEARCH-COMPUTATIONAL GENOMICSCase Western Reserve UniversityCleveland, OHPost doctoral positionCNRSNice Utrecht, FranceVICE CHANCELLOR FOR RESEARCHUniversity of Missouri, ColumbiaColumbia, MOFACULTY POSITIONUniversity of MiamiMiami, FLNeuroscience PositionGeorgia State UniversityAtlanta, GAFaculty Positions, Department of PharmacologyUniversity of MichiganAnn Arbor, MIMEDICAL INFORMATICS FACULTY POSITION AND DIRECTOR OF NORTHWESTERN UNIVERSITY BIOMEDICAL INFORMATICS CENTERNorthwestern UniversityChicago, ILMore jobsHeterogeneity (the standard deviation of degrees)Global Evenness of link distribution among genesClustering coefficient, (the average connectivity of subnetworks containing each nodes's neighbors), CCGlobal Appearance of tightly connected regulatory subnetworksAverage distance, Dav = [dij]/[N(N-1)]Global Number of communication steps between two randomly chosen genesArc connectivity Global Minimal number of gene interactions whose deletion results in a disconnected networkNode connectivity Global Minimal number of genes whose deletion results in a disconnected networkIndices describing interactions within networks. Such indices include those used by ecologists to determine complex interactions within food web networks. These indices can be applied to the measurement of interactions within and between gene-regulation networks (13, 14).There are other indices derived from analyses of food web complexity thatmight be useful for analyzing the connectivity of gene-regulation networks (see the table). For instance, the clustering coefficient could be used to define relatively autonomous groups of developmental genes.