Anecdotes, data and regulatory modules

Beginning in the late 1980s, Eric Davidson's group at Cal Tech developed a modularity hypothesis of developmental gene regulation, showing that in an expanding number of cases, particular aspects of development were governed by compact ‘modules’ of transcription factor binding sites (TFBSs), and that these modules were separable, complex and interconnected. Davidson made no attempt to further generalize the hypothesis, but others took up the idea, transported it out of development and extended it to a general rule of clustering. Despite such misbegotten origins, the ‘extended’ modularity hypothesis—that TFBSs in general tend to come in compact clusters—has been highly productive, yet it has never been challenged with a large, diverse and unbiased dataset to see how universal it actually is. The aim of the present paper is to do so. Applying human–mouse–rat phylogenetic footprinting to neighbourhoods of a diverse set of TFBSs, including both developmental and non-developmental signals, we find that the extended hypothesis holds in at least 93.5% of cases. Based on this particular sample, we found a mean module length of 609 nucleotides containing, on an average, 24.5 presumptive regulatory signals of length greater than 5 and averaging 8.5 nucleotides each.