Session D40: Focus Session: Systems Biology and Biochemical Networks I

Optimal Information Processing in Biochemical Networks

A variety of experimental results over the past decades provide examples of near-optimal information processing in biological networks, including in biochemical and transcriptional regulatory networks. Computing information-theoretic quantities requires first choosing or computing the joint probability distribution describing multiple nodes in such a network --- for example, representing the probability distribution of finding an integer copy number of each of two interacting reactants or gene products while respecting the `intrinsic' small copy number noise constraining information transmission at the scale of the cell. I'll given an overview of some recent analytic and numerical work facilitating calculation of such joint distributions and the associated information, which in turn makes possible numerical optimization of information flow in models of noisy regulatory and biochemical networks. Illustrating cases include quantification of form-function relations, ideal design of regulatory cascades, and response to oscillatory driving.   

Adaptation at the output of the chemotaxis signaling pathway

The chemotaxis signaling pathway allows bacterial cells to sense and respond to changes in concentrations of chemical attractants or repellents. In E. coli, the machinery required for bacterial chemotaxis includes two large membrane-embedded multiprotein complexes, one that processes input (receptor clusters) and the other that generates output (flagellar motors). These complexes are coupled by diffusion of a small phosphorylated cytoplasmic protein, CheY-P, which binds to the flagellar motors, increasing the probability that they spin clockwise. Receptor output (the steady-state concentration of CheY-P) varies from cell to cell. However, the motor is ultrasensitive, with a narrow [CheY-P] operating range. How might the receptor output and motor input be matched? By combining various techniques such as FRET, single-motor TIRF, and single-motor bead assay, we showed that the motor shifts its operating range to match the receptor output by changing its composition. The number of FliM subunits in the C-ring increases in response to a decrement in the concentration of CheY-P, increasing motor sensitivity. Such adaptive remodeling is likely to be a common feature in the operation of many molecular machines.   

The energy cost of accurate adaptation in networks with incoherent type-1 feed-forward loop

The incoherent type-1 feed forward loop (I1-FFL) is a common regulatory motif in many biochemical networks, some of which are responsible for accurate sensory adaptation. In this work, we analyze the sensitivity and adaptation function of the I1-FFL type enzymatic reaction networks. We show that detailed balance is broken in I1-FFL and continuous energy dissipation is needed to improve both the sensitivity and the adaptation accuracy of the network. Our study revealed a relation between the performance improvement and the energy dissipation rate. We find that this energy-assisted improvement is bounded (limited) by intrinsic properties of the molecular reaction system. The performance-energy relation in I1-FFL is similar to the recently obtained Energy-Speed-Accuracy relation in networks with negative-feedback-loop, another key motif for accurate sensory adaptation.   

Noise places constraints on eukaryotic gradient sensing and chemotaxis

Chemotaxis is characterized by the directional cell movement following external chemical gradients. It plays a crucial role in a variety of biological processes including neuronal development, wound healing and cancer metastasis. Ultimately, the accuracy of gradient sensing is limited by the fluctuations of signaling components, e.g. the stochastic receptor occupancy on cell surface. We use concepts and techniques from statistical physics, estimation theory, and information theory to quantify the stochastic and nonlinear information processing in eukaryotic chemotaxis. We mainly address the following questions: (1) What are the physical limits of eukaryotic spatial gradient sensing? (2) How to characterize the movements of chemotactic cells? (3) How much gradient information can be reliably gained by a chemotactic cell? By answering those questions, we expect to derive new insights for general biological signal processing systems.   

Process-driven inference of biological network structure: feasibility, minimality, and multiplicity

For a given dynamic process, identifying the putative interaction networks to achieve it is the inference problem. In this talk, we address the computational complexity of inference problem in the context of Boolean networks under dominant inhibition condition. The first is a proof that the feasibility problem (is there a network that explains the dynamics?) can be solved in polynomial-time. Second, while the minimality problem (what is the smallest network that explains the dynamics?) is shown to be NP-hard, a simple polynomial-time heuristic is shown to produce near-minimal solutions, as demonstrated by simulation. Third, the theoretical framework also leads to a fast polynomial-time heuristic to estimate the number of network solutions with reasonable accuracy. We will apply these approaches to two simplified Boolean network models for the cell cycle process of budding yeast (Li 2004) and fission yeast (Davidich 2008). Our results demonstrate that each of these networks contains a giant backbone motif spanning all the network nodes that provides the desired main functionality, while the remaining edges in the network form smaller motifs whose role is to confer stability properties rather than provide function. Moreover, we show that the bioprocesses of these two cell cycle models differ considerably from a typically generated process and are intrinsically cascade-like.   

Communication channels between membrane bound proteins

Much of what might be called biological computation takes place on the plasma membrane, a 2D liquid composed of a diverse soup of lipids and embedded proteins. Motivated by the recent discovery that these membranes seem to be tuned close to a 2D liquid-liquid critical point, we set out to understand the different channels through which membrane bound proteins can communicate. Diffusing proteins can carry out reactions like phosphorylation when they come in contact with each other. Near criticality, proteins can also exert long-ranged critical Casimir forces on one another by coupling to the local composition order parameter. By modulating the growth and breakdown of the rigid cytoskeleton, they can direct forces on even more distant regions. In addition, proteins can control the release and production of second messengers that diffuse either through the bulk, or in the plane of the membrane itself. By making simple models for these processes we bound functional measures for them as communication channels. These include information theoretic measures of bandwidth, as well as physical measures of energetic efficiency and speed. Our results will likely shed light on the functional role of clustering and other collective behaviors often seen in experiments.   

Maximum entropy principle for predicting response to multiple-drug exposure in bacteria and human cancer cells

Drugs are commonly used in combinations larger than two for treating infectious disease. However, it is generally impossible to infer the net effect of a multi-drug combination on cell growth directly from the effects of individual drugs. We combined experiments with maximum entropy methods to develop a mechanism-independent framework for calculating the response of both bacteria and human cancer cells to a large variety of drug combinations comprised of anti-microbial or anti-cancer drugs. We experimentally show that the cellular responses to drug pairs are sufficient to infer the effects of larger drug combinations in gram negative bacteria, \textit{Escherichia coli}, gram positive bacteria, \textit{Staphylococcus aureus}, and also human breast cancer and melanoma cell lines. Remarkably, the accurate predictions of this framework suggest that the multi-drug response obeys statistical rather than chemical laws for combinations larger than two. Consequently, these findings offer a new strategy for the rational design of therapies using large drug combinations.   

Robust regulation of oscillatory Min-protein patterns

Robust spatial patterning was crucial just from the beginning of cellular evolution, and is key to the development of multicellular organisms. In E. Coli, the oscillatory pole-to-pole dynamics of MinCDE proteins functionality prevent improper cell divisions apart from midcell. Min-oscillations are characterized by the remarkable robustness with which spatial patterns dynamically adapt to variations of cell geometry. Moreover, adaption, and therefore proper cell division, is independent of temperature. These observations raise fundamental questions about the underlying core mechanisms, and about the role of spatial cues. With a conceptually novel and universal approach to cellular geometries, we introduce a robust model based on experimental data, consistently explaining the mechanisms underlying pole-to-pole, striped and circular patterns, as well as the observed temperature-dependence. Contrary to prior conjectures, the model predicts that MinD and cardiolipin domains are not colocalized. The key mechanisms are transient sequestration of MinE, and highly canalized transfer of MinD between polar zones. MinD channeling enhances midcell localization and facilitates stripe formation, revealing the potential optimization process from which robust Min-oscillations originally arose.   

Mathematical Analysis of Biomolecular Network Reveals Connections Between Diseases

Connections between cancer and metabolic diseases may consist in the complex network of interactions among a common set of biomolecules. By applying singularity and bifurcation analysis, the phenotypes constrained by the AKT signaling pathway are identified and mapped onto the parameter space, which include cancer and certain metabolic diseases. By considering physiologic properties (sensitivity, robustness and adaptivity) the AKT pathway must possess in order to efficiently sense growth factors and nutrients, the region of normal responses is located. The analysis illuminates the parameter space and reveals system-level mechanisms in regulating biological functions (cell growth, survival, proliferation and metabolism) and how their deregulation may lead to the development of diseases. The analytical expressions summarize the synergistic interactions among many molecules, which provides valuable insights into therapeutic interventions.   

Modeling the mammalian circadian clock

In biology, important processes often depend on a temporal schedule. The 24-hour periodicity of solar illumination caused by the earth's rotation has consequences for environmental factors such as temperature and humidity as well as ecological factors such as the presence of food, predators, or potential mates. As a result, many organisms have evolved to develop a circadian clock that allows them to anticipate these environmental changes in the absence of direct temporal cues. In recent years, extensive efforts have been made to deconstruct the biological clockwork from various organisms, develop mathematical models of circadian function, and construct synthetic analogues to test our understanding. My present work has two major foci. First, we have used regulatory principles revealed by recent experimental work to construct a model of the core genetic oscillator of the mammalian circadian system that captures key system-level behaviors. Second, we are exploring the possibility of a post-translational phosphorylation-based oscillator that is coupled to the core oscillator, conferring enhanced robustness and stability on the complete system. A simple model of this post-translational oscillator reveals key design constraints that must be satisfied by any such oscillator.