The recurrent, though not exclusive, theme of my research up to this point has been the immune system. Aside from its protective role against infectious agents, and possibly cancer, the immune system is also studied as a paradigm of an adaptive, distributed, complex system. Evolutionary dynamics in the immune system unfolds both on the level of species, as well as on the level of cell populations that take part in one or more immune responses. The immune system currently solves resource allocation problems. What immune receptors are to be encoded by the germline gene fragments? What individual specificity is to handle a particular pathogen? What type of effector mechanisms are to be dispatched in response to the pathogen? What the immune system reacts to is a long-stading question. The fascinating aspect of it is that potential pathogens, as well as self structures have similar biochemical structure, and it is so far unclear what features are being extracted by the immune system from a molecular or cellular entity to be able to tell apart self from non-self.

In my previous work I addressed the question of evolutionary learning in the immune system. The immune receptors that handle a particular pathogen are shaped by
the history of pathogenic encounters of the species, as well as by a miniature evolutionary process that takes place during individual immune responses, namely affinity maturation. My studies show that affinity maturation plays an important role in organisms that live in environments with a large number of pathogens, and suggest that the mechanism of somatic hypermutation appeared very early in phylogeny. I have also been able to demonstrate preferential targeting of this mechanism to the complementarity-determining region of antibodies via codon bias, this finding also arguing for the important role played by somatic hypermutation in generating the immune repertoire.

The process of somatic hypermutation naturally leads to the question of how the mechanisms that generate diversity come about and evolve. This aspect of evolutionary dynamics is less studied than the selection and sampling phenomena, although it is by no means less important. It has long been recognized that diversity-generating mechanisms are recruited when the population finds itself in stressful circumstances, and there has been a long debate of whether mutator phenotypes are being induced or are just a consequence of limited repair capacity of stressed populations. I have studied the targeting and spectrum of germline mutations in humans, and found that there are significant biases. Some of these have been known already, for example the high propensity of CG dinucleotides to mutate once the cytosine has been methylated. Why then the methylation status of these dinucleotides acquired functional significance in gene regulation is an interesting question. Other biases  may be more subtle, and may constitute the substrate of evolutionary adaptation. The type of mismatches that the DNA polymerase recognizes, recruiting the repair mechanisms are a case in point.

I see my future work developing along two main directions. Concerning the immune system, I would like to pursue the question of what type of information is encoded in the antibody gene libraries. I have already started to address this question, and the conclusion I have so far is that, if we disregard the problem of self-non-self discrimination, the germline antibodies would be broadly reactive antibodies. There are experimental studies claiming that neonatal antibodies are indeed "sticky". The next step is to incorporate the process of negative selection in the model. The problem of self-non-self discrimination is more elusive, but I am also following it to some extent. Concretely, I will be studying the process of antigen-induced cell death in the near future. The second direction that I would like to pursue has to do with the evolution of diversity-generating mechanisms. It is clear that the sequence spacethat will be explored in evolution is constrained
 

The goal of my research is a quantitative understanding of biological processes. My education began as an undergraduate in Mathematics and Informatics. I then  took up  Medicine,  completed an M.D.,  and followed this with a Ph.D. in Computer Science.  This path itself illustrates that I am fascinated by the way organisms work, and that I would like to understand that in a way that has predictive power. Most of my research concerned various aspect of the immune response. On the way, I came to realize that many of the questions that I came across,  and the techniques that I used to approach  these questions are of much larger scope than that of the immune system. What are the constraints on the evolution of two interacting systems, such as the immune system and a pathogenic agent? What are the rates at which they evolve? Are these rates correlated? What makes a system evolvable? How does a pathogen modify pre-existenting biochemical pathways to develop antibiotic  resistance? Is the structure of the antibody libraries the result of phenotypic selection or is it largely determined by processes that take place at the level of genes? Can we infer something about the pathogenic environment of a species by analyzing the structure of the antibody libraries of that species? How do we extract information from the wealth of sequence data that will be generated by the genome projects? In the projects that I took up so far, I began  to address some of these issues.

During my medical studies, I was attracted to the pre-clinical fields, physiology in particular, as well as bioinformatics. I was involved in a number of projects. I began with a clinical study of the cardiovascular modifications during physical effort.  I then moved to a pharmacological study of the effects of drug associations on the guinea pig trachea, part of a larger project on asthma.  Eventually, as a teaching assistant in physiology, trying to explain to the students the processes of inflammation, infection, and tissue repair, I became interested in the immune system, and in the many gaps in the knowledge we had about these topics. For my thesis, I developed a model of an idiotypic network - the most prevalent approach to immune system modeling at that time - and  analyzed some of its properties.  After graduation, I came to New Mexico, to work on immune system modeling. My first study,on the memory capacity of a model idiotypic network, appeared  in the Bulletin of Mathematical Biology 56:899-921 (1994).

I continued to explore constraints on cellular interactions in  immune responses in the context of more realistic models. First, I developed a model of the primary humoral immune response. From the dynamics of the model, we inferred: 1) that T-cell activation and replication has to precede the B-cell response; 2) that once activated, B cells must undergo a number of divisions,  without requiring contact with T cells; 3) that B cell division has to precede their differentiation into plasma cells. This issues were at that time under experimental investigation. The division/differentiation dilemma, for example, has been since clarified, B-cell differentiation requiring indeed a number of cell divisions(see Hodgkin P.D. - J. Exp. Med. 184:277-281, 1996 and Hasbold, J. - Eur.J.Immunol. 28:1040-1051, 1998). Continuing with the dynamics of secondary immune responses,  I analyzed the efficiency of affinity maturation in a model of germinal center reaction which obeys the known architectural constraints of this microenvironment. Germinal centers are a very interesting experimental model, as they harbor miniature evolutionary processes during on-going immune responses. The affinity of the B-cell receptors participating in the immune response increases by as much as 2-3 orders of magnitude, over a period of 1-2 weeks, due to rounds of mutation and antigen-based selection. The nature of the mutation mechanism is unknown. The mechanisms by which the very few high affinity cells that are generated in the process come to dominate the germinal center population withing a few days, are also not understood.We have concluded that part of the efficiency of this process is due to multiple cycles of proliferation, mutation and selection. This hypothesis, which is now gaining more and more acceptance in the immunological community, was proposed by Kepler & Perelson, and I later showed that it is consistent with the  architectural constraints of the germinal centers. The  classical view of the germinal center reaction assumed one-pass, with no feedback from the selected cells to the proliferative compartment. I constructed such a model, and showed that it could not give rise to the numbers of  high affinity cells that one observes in germinal centers. In the context of this model, I was also able to explore more general features of systems in which selection is due to an agent that decays with time.

A second property contributing to the high efficiency of the germinal center reaction is targeted mutation, that is, somatic mutations are preferentially introduced in the antigen-binding regions of the immunoglobulin genes. Together with Prof. Kepler, from North Carolina State University, I developed a resampling-based methodology to analyze the factors that contribute to the higher tendency of these regions, complementarity-determining regions (CDR), to undergo replacement mutations. We developed methods to analyze individual sequences, as well as complete sets of variable region genes, and we showed that codon composition is largely responsible for targeting mutations to the CDRs. We performed a comparative analysis of data sets from various species. We found that the V-region sequences from all these species have codon usage bias that would target somatic hypermutation to the CDRs.  This result argues for somatic hypermutation being discovered early in the phylogeny of vertebrates.  We are currently in the process of analyzing the sequence bias of the  somatic hypermutation mechanism, with the more remote goal of understanding what the nature of this mechanism might be.

Yet another  application on which I have focused is the evolution of immune system libraries in a pathogenic environment. With my advisor at the University of New Mexico, Prof. Stephanie Forrest, I developed a model using genetic algorithms to simulate the evolutionary process. The classical view, introduced by Tonegawa, is that the immune system generates a very large number of random receptors, which collectively cover the "complete" pathogen set. There are problems with this view however, because in certain species (such as sharks), or developmental stages (such as neonatally), the diversity of immune receptors is much more limited than initially believed. We showed  that the scaling of the fitness of the organism (seen as its probability to survive a series of pathogenic challenges) as a function of the size of its immune repertoire is likely logarithmic, or sublogarithmic. Thus we proposed the hypothesis that germline diversity serves as a coarse-graining of the pathogen space, with somatic hypermutation improving the affinity/specificity of the initial antibodies. We investigated the effects of static and evolving pathogen sets, and concluded that the efficiency of the encoding realized by the antibody repertoire depends most crucially on the size of the pathogen set that an organism confronts during  its lifetime. It is, however, unlikely that the immune system tries to recognize "as many molecular shapes as possible". Instead, we argue that the germline-encoded antibodies have the role of rapidly directing the immune response in the regions of the molecular shape space that are crucial for the survival of the organism.

My collaboration with Prof. Kepler is aimed at developing a combination of computational and statistical methods to address interesting biological phenomena. One of the projects on which we have worked stemmed from the lack of appropriate methods for estimating mutation rates in germinal centers. On the way, we were able to design an improved method for estimating mutation rates in bacterial cultures, and to give corrections for the cell cycle time distribution. The currently available method makes a Markovian assumption about cell replication, which we show overestimates the mutation rate in the culture by as much as 30%. I also attempted  an extension of these methods to germinal centers, based on mutation data from passenger genes. Some hypotheses underlying my approach remain to be tested.