Bow ties, metabolism and disease Marie Csete 1 and John Doyle 2 1 Emory Anesthiology Research Labs, 1462 Clifton Rd NE, Room 420, Atlanta GA 30322, USA 2 Control and Dynamical Systems, California Institute of Technology, 1200 E. California Blvd, Pasadena CA 91125, USA Highly organized, universal structures underlying bio- logical and technological networks mediate effective trade-offs among efficiency, robustness and evolvability, with predictable fragilities that can be used to under- stand disease pathogenesis. The aims of this article are to describe the features of one common organizational architecture in biology, the bow tie. Large-scale organi- zational frameworks such as the bow tie are necessary starting points for higher-resolution modeling of com- plex biologic processes Systems biology is a multidisciplinary approach for describing, understanding and controlling the properties and dynamics of whole biological networks and systems that combines high-throughput data generation with new computational and mathematical tools [1]. The long-term goal in systems biology – that is, the integration of information from varied sources into models that are in some sense ‘complete’ – will require the connection of data and models at several levels of abstraction and detail [2]. Without an organizational framework, however, the sheer complexity of whole cells and organisms can be overwhelming, and model-based predictions about the nature of rare events of greatest interest (i.e. disease states) will be difficult to extract. Thus, complex systems can be understood only by identifying their organizing principles, theories, design rules and, in particular, proto- cols. By protocols, we mean the rules and interfaces by which modules interact; these protocols are organized within a global framework referred to as the ‘architecture’ [3]. Engineering approaches to understand the organiz- ational principles of biological networks have had both a rich, successful history and a recent revival in interest [4]. Engineering tools have been used to identify important biological motifs and modules [5], as well as their regu- lation by universal principles of robustness [3]. Even in engineering, however, complex networks such as very- large-scale integrated circuits are not modeled simul- taneously at the level of whole-chip or ‘device physics’, but instead are modeled with a hierarchy of schemes of various resolution. Here we address global architectures and protocols that complement the device physics of local circuit motifs [6]. The natural language in which to describe these universal principles is, of course, mathematical. Unfortunately, theory for the type of distributed and asynchronous global control used in biology is relatively new [7]. Nevertheless, from existing concepts it is possible to distill insight into candidate universal architectures that can be tentatively confirmed by comparing biological systems with one another and with technological systems. ‘Bow-tie’ structures and protocols in metabolism Although bacterial metabolism is probably the best- studied biological network, little is known about the detailed kinetics necessary to model quantitatively the vast regulatory feedbacks that control metabolism. Yet much progress has been made by taking stoichiometry, which is known in some detail, as a prerequisite and by simply assuming that the unmodeled control actions are optimal (i.e. they rapidly create controlled equilibrium that are perfectly adapted to environmental conditions and cellular demands) [8–10]. In this article, we consider in what sense the global architecture of metabolism itself, including stoichiometry and regulation, can be thought of as ‘optimal’, and we argue that this remarkable, evolved optimal architecture, together with its protocols, reflects universal organizational principles of complex networks. Bacterial metabolic networks are a striking example of ‘bow-tie’ organization and illustrate the flexibility that such a structure provides. As shown in Figure 1, a myriad of nutrient sources are catabolized, or ‘fan in’, to produce a handful of activated carriers (e.g. ATP, NADH and NADPH) and 12 precursor metabolites (e.g. glucose 6-phosphate, fructose 6-phosphate, phosphoenolpyruvate and pyruvate), which are then synthesized into roughly 70 larger building blocks (e.g. amino acids, nucleotides, fatty acids and sugars). The precursors and carriers can be thought of as two ‘knots’ of separate bow ties that are both fed by catabolism, but whereas the former ‘fan out’ locally to the biosynthesis of universal building blocks, the latter fan out to the whole cell to provide energy, reducing power and small moieties. The building blocks then further fan out into the complex assembly of macromolecules by general-purpose polymerases. Although this description of the bow-tie structure of metabolism is an engineering interpretation of familiar textbook biochemistry, it also has been derived from the computational analysis of genomic data from 65 microorganisms [11]. The transcription and translation (‘trans’) processes also have a bow-tie architecture. A few universal poly- merase modules that make up the ‘knot’ of the trans bow- tie machinery function efficiently with a universal codon usage protocol, facilitating the fan in of a large variety of genes and the fan out of an even larger variety of proteins. Nested together, the bow ties of core metabolism and the trans machinery create a larger ‘metabolism bow tie’ that Corresponding authors: Marie Csete (marie_csete@emoryhealthcare.org), John Doyle (doyle@caltech.edu). Available online 30 July 2004 www.sciencedirect.com 0167-7799/$ - see front matter Q 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.tibtech.2004.07.007 Opinion TRENDS in Biotechnology Vol.22 No.9 September 2004