Seminars

Lymphocyte recirculation between blood, lymphoid organs and lymph is crucial for antigen scanning and effector functions of lymphocytes. Although a lot was already known about lymphocyte entry into the lymph nodes (LNs) from the blood, the anatomical location and mechanism of lymphocyte exit from the LNs was unclear. By 2-photon intravital imaging we identified cortical LYVE-1+ sinuses as the sites of lymphocyte exit and suggested a multistep model of T cell exit from the LNs. Previous work indicated that intrinsic expression of S1P1 receptor by T cells and a gradient of its ligand, S1P, are required for T cell egress. Quantitative characterization of T cell migration in relation to the cortical sinuses demonstrated that T cells migrate to the sinuses and probe their surface in an S1P1-independent manner; however, commitment to transmigration into the sinus requires expression of S1P1 on T cells. After transmigration into cortical sinuses T cells are retained and passively carried by flow towards “portals” in the subcapsular space proximal to the medullary region and efferent lymphatics. Based on the quantitative 2-photon imaging data and confocal 3D reconstruction of High Endothelial Venules (HEVs) and LYVE-1+ sinuses we built a quantitative random walk model of T cell migration and exit from the LNs. The model suggests that T cells go through a few rounds of entry into the LYVE-1+ sinuses before they can reach the efferent lymphatics and egress from the LNs.

The mammalian immune response is a striking example of the coordinated operation of many cell types. Intercellular communication is mediated by signaling molecules that form concentration gradients, which requires cells to respond to a wide range of signal intensities. We used high-throughput microfluidic cell culture, quantitative gene expression analysis and mathematical modeling to investigate how mammalian cells respond to different concentrations of the signaling molecule TNF-α, and modulate gene expression via the transcription factor NF-κB. We measured NF-κB activity in thousands of live cells under TNF-α doses covering four orders of magnitude. In contrast to population studies (i.e. Western Blot's), NF-κB activation is found to be a switch-like process at the single cell level with fewer cells responding at lower doses. The activated cells up-regulate early genes independent of the TNF-α concentration, while only high dose stimulation results in the expression of late-term genes. Cells also encode a subtle set of analog parameters such as the NF-κB peak intensity, response time and number of oscillations to modulate the outcome. Using our comprehensive data, we developed a mathematical model that reproduces both the digital and analog dynamics as well as the gene expression profiles at all measured conditions, constituting a broadly applicable model of NF-κB signaling. In addition to their biological significance, our results highlight the value of high-throughput quantitative measurements at the single-cell level in understanding how biological systems operate.

Nonlinear models with large numbers of difficult-to-measure parameters appear commonly in many fields of science, particularly in biology. Here I focus on modeling the dynamics of biochemical reaction networks and the spread of genetic variation among human populations. In both cases, estimating parameters and uncertainties is a major obstacle to developing useful models. I demonstrate that nonlinear models, particular of biochemical networks, exhibit a universal "sloppy" pattern of sensitivity to parameter variation; different directions in parameter space vary by orders of magnitude in their constraint. Consequently, predictions may be usefully constrained even when the available data only very poorly constrain individual parameter values. I also present a powerful method for inferring models of demographic history from population genetic data. With this method, we have developed the most complex statistically well-characterized model of human genetic history to date.

Modern computational neuroimaging studies employ multidisciplinary approaches to address challenging biomedical problems using multimodal data, e.g., imaging, phenotype and genotype. We are developing, validating and disseminating diverse computational resources and distributed web-services for processing, analyzing and visualizing large amounts of heterogeneous data. This talk will discuss specific neuroimaging challenges, mathematical modeling and statistical techniques, computational infrastructure, and research findings based on brain development and aging data. Examples will include classification of spatially unaligned fMRI data, brain tissue classification, computational brain atlasing, spectral and PDE based shape representations, Bayesian network modeling, and statistical analyses of biological manifolds. We will also discuss interoperability of disparate computational resources and graphical design, management and execution of complex neuroimaging workflows (data analysis protocols).

The natural language processing (NLP) applications which ultimately affect people's daily lives are high level, semantically-oriented ones: question answering, machine translation, machine reading, speech interfaces for robots and machines, and others that we haven't even thought of yet. Humans are very good at these types of tasks, in part because they naturally employ holistic language processing. They effortlessly keep track of many layers of low-level information, while simultaneously integrating in long distance information from elsewhere in the conversation or document. In contrast, much NLP research focuses on lower-level tasks, like parsing, named entity recognition, and part-of-speech tagging. Moreover, for the sake of efficiency, researchers modeling these phenomena make extremely strong independence assumptions, which completely decouple these tasks, and only look at local context when making decisions. This talk will cover multiple aspects of holistic language processing, and describe systems for joint parsing and named entity recognition; named entity recognition which incorporates long-distance information; and multi-task learning over multiple domains, and over multiple datasets with varying amounts of annotated information. These systems are designed to produce analyses which are more consistent, of higher quality, and generally more useful for doing the kinds of tasks that non-researchers actually care about.

Laboratory experiments with microorganisms offer unique opportunities to study evolution in action. We used next-generation DNA sequencing data to reconstruct the dynamics of genome evolution from the 40,000-generation frozen "fossil record" of an Escherichia coli population. Surprisingly, the rate at which beneficial mutations substituted in this population was relatively constant at first, despite a dramatically decelerating rate of adaptation. In contrast, the neutral substitution rate and amount of genetic diversity in the population were highly variable over time and increased dramatically after a mutator phenotype evolved. Additional experiments that "replay the tape" of early evolution in this population show that a specific lineage is able to reproducibly overtake competitors of higher fitness because it maintains a greater potential for further adaptation.

The bacterial chromosome is condensed into a compact DNA-protein complex called the nucleoid. It is the nucleoid, not naked DNA, which is the substrate for all genetic processes from gene expression and DNA repair to chromosome replication. The physical structure of chromosomes has functional consequences: It affects gene regulation from the simplest prokaryotes to multicellular organisms. Nucleoid organization also plays a poorly understood yet central role in chromosome segregation. In spite of its biological significance, little is known about the mechanisms that physically organize the chromosome on a cellular-scale. In this talk I will describe work that combines biophysics with more traditional genetics and cell biology techniques to probe the mechanisms of nucleoid organization. In spite of the common assumption that the interphase chromosome is well-modeled by an unstructured polymer, measurements of the locus positions reveal that the E.coli chromosome is precisely organized into a filament with a linear order. The vast majority of genetic loci are positioned in the cell with a precision of 10% of the cell length, with the exception of loci close to the replication terminus. The measured dependence of the precision of inter-locus distance on genomic distance singles out intra-nucleoid interactions as the mechanism responsible for chromosome organization. From the magnitude of this variance, we infer the existence of an as-yet uncharacterized higher-order DNA organization in bacteria. We demonstrate that both the stochastic and average structure of the nucleoid is captured by a fluctuating elastic filament model. The analysis of mutant strains reveals that the poorly structured terminus region plays a central but unexpected role in the organization of the entire nucleoid filament.

Abstract: Damage to the lining of a blood vessel triggers the intertwined processes of platelet aggregation and coagulation that result in the formation of a thrombus (clot) at the injury site. The thrombus itself is made up of platelets adherent to the vessel and to one another, and of a fibrin protein gel surrounding and between the platelets. An enzyme thrombin is critical to both platelet deposition and to fibrin gelation and is produced by a complex network of reactions on the vascular surface, in the blood plasma, and on the surfaces of platelets. This process happens under flow and, in turn, can strongly influence the flow. I will present work addressing two problems related to these processes:
1) How do platelet deposition and coagulation up through thrombin production interact under flow?
2) How can the rate at which thrombin produces fibrin momoners affect the ultimate branching structure of the fibrin gel?