Few inputs can reprogram biological networks

Abstract

Arising from Y. Liu, J. Slotine & A. Barabási Nature 473, 167–173 (2011)10.1038/nature10011; Liu et al. reply

Liu, Slotine and Barabasi¹ identify subsets U of nodes in complex networks, which are required to exert full control of these networks. Control in this context means that for each possible state S of the network there exist inputs for all nodes in U, which are sufficient to force the network to state S¹. Application of the methodology to gene regulatory networks suggests that roughly 80% of all nodes must be controlled to drive such a network. This seems to contradict recent empirical findings^2,3,4,5,6 in the cellular reprogramming field.

Main

Mammalian cells have been shown to be efficiently reprogrammable towards other cellular phenotypes^2,3,4,5,6. Controllability of such complex transitions in transcriptional networks underlying cellular phenotypes appears to be an intrinsic biological property and is being used for the development of novel disease models and cellular therapeutics in regenerative medicine. However, in contrast to technical or social networks, biological networks show a high degree of intrinsic co-regulation. Co-regulation appears to quench the admissible space of states of gene expression networks to a combinatorial expression space with relatively low dimensionality. Recent publications suggest that this subspace is spanned by only a few (five to fifteen) genome-wide differential expression patterns, which are—surprisingly—sufficient to characterize most observable biological phenotypes^7,8,9. Each of these patterns is given by weighted sums over the expression of relatively extensive gene sets contributing coherently to the respective pattern^7,8,9. Although these empirical studies suggest linear structures underlying the respective subspaces, nonlinear or star-shaped topologies cannot be excluded.

Moreover, genome-wide co-regulation may result in dynamic response features, as observed in chemical networks. Hence, owing to their low dimensionality, full control and reprogramming of biological systems may be achieved by only a few key control factors, seemingly contradicting the more general concept put forward by ref. 1. Recent experimental studies show proof of concept for a wide range of reprogramming of biological systems using overexpression of only one to five transcription factors^2,3,4,5,6,10, which effectively regulate only a subset of downstream genes. We note that the number of nodes (five or fewer genes out of about 30,000) needed to fully control a biological system^2,3,4,5,6,10 is much less than the estimate of 80% of all nodes¹.

Thus, for the special case of biological systems, it may be sufficient to weaken the standard definition of controllability of networks—control of large sets of nodes as suggested by ref. 1—to mean the controllability of restricted, biologically admissible network states.