Shaping individual diversity for collective success

Traditionally, the behavior of human-engineered systems has been easier to predict than that of biological systems. However, the growing scale and complexity of engineered systems such as the internet, the energy grid, and financial markets are straining our ability to control and predict their behavior. Engineers are increasingly facing difficulties similar to those tackled by biologists and doctors.

Engineering successes have often relied on precisely predicting the behavior of individual components, so that when the components are assembled in a particular design, we can accurately predict the behavior of the full system. This approach, however, breaks down for complex systems in which individual components do not behave predictably and consistently. In large systems such as communication networks, financial markets, and the human body, similar components often exhibit different responses to identical inputs: e.g. different responses to load imbalances by substations in an energy grid, different reactions to the same financial information, and different responses to the same drug by individual cells in an organism. A grand challenge in complexity science of interest to both biologists and engineers is to determine how to study, understand, and eventually optimize distributed systems composed of parts that exhibit significant behavioral variability.

In biology, cells rely on networks of biomolecules to detect, interpret, and respond to information about their environment. Cellular behaviors therefore arise from the dynamic responses of biomolecular networks. These dynamics are in turn determined by the architecture of the networks (i.e. who binds who) and by a set of dynamical parameters (e.g. gain, relaxation time, etc) that are nonlinear functions of the binding constants and relative abundances of the network components. Although selective pressures shape the genetic information that encodes for the specificity and expression levels of network components, the relative abundances of these components can be highly variable from cell-tocell due to biological noise. Sources of noise include: fluctuations in the chemical reactions underlying component synthesis and degradation; progression through the cell cycle or life cycle; local availability of metabolites; and fluctuations in environmental signals. Variability in the abundance of network components can tune a particular biological function differently for each individual cell, resulting in a distribution of network behaviors, or "tunings," even throughout a genetically homogeneous population of clones.

If a single optimum response exists for all situations, then behavioral variability would prevent a significant fraction of the population from performing optimally. On the other hand, if environmental conditions fluctuate in space and time, then population heterogeneity might confer robustness. Isogenic populations might use a distribution of tunings advantageously to create a diversified portfolio of specialized subpopulations while still maintaining the same network architecture. Matching the distribution of capabilities within a population to the variations of a particular environment may be crucial for survival or competitive fitness. For example, for a bacterium like Escherichia coli, environmental parameters range from the high temperatures, low oxygen tensions, low pH, and high flow rates of mammalian guts to mostly opposite properties when living freely outside of mammals.

Our current understanding of the relationship between phenotypic heterogeneity and robustness in isogenic populations comes primarily from the study of cellular systems responsible for binary decisions, in which biological noise is used to control the percentage of cells in each state. For more complex behaviors it remains unclear how phenotypic variability is controlled and what functional role it may play.