Session H4: Specificity, Recognition and Coding in Biology

Temporal coding in gene regulation

Cells face the fundamental problem of having to respond to a nearly unlimited set of distinct input signals with a highly limited set of signaling pathways. Accordingly, many signaling networks exhibit a bow-tie topology: multiple distinct signal inputs (e.g. hormone or stress exposure) converge on a single master regulator, often a transcription factor (TF), which then controls the expression of different downstream target genes. This raises the question of how specificity is achieved. We will discuss recent evidence\footnote{Purvis, J.E. and Lahav, G., 2013. Cell, 152(5), pp.945-956.} that signal input specificity is \textit{encoded} through regulation of TF activation dynamics. For example, in budding yeast the master TF Msn2 exhibits short pulses of activity with dose-dependent frequency in response to glucose starvation, but sustained activation with dose-dependent amplitude in response to oxidative stress\footnote{Hao, N. and O'Shea, E.K., 2012. NSMB, 19(1), pp.31-39.}. Combining high-throughput microfluidics with quantitative time-lapse microscopy, we will show that gene promoters exhibit different activation timescales (slow vs. fast) and thresholds (low vs. high), such that four extreme promoter classes exist\footnote{Hansen, A.S. and O'Shea, E.K., 2013. MSB, 9(1), p.704.}. Further, we will show that each promoter class can be preferentially induced by a specific set of Msn2 dynamics such that four distinct gene expression programs can be encoded in the dynamics of a single TF\footnote{Hansen, A.S. and O’Shea, E.K., 2016. Current Biology, 26(7), pp.R269-R271.} and that promoter class correlates with gene function. Together, our results reveal a temporal code where cells \textit{encode} both signal \textit{identity} and \textit{intensity} information in the activation dynamics of TFs and then \textit{decode} this information at the promoter level. This may allow cells to respond specifically to many signal inputs despite having few signaling pathways.   

Intrinsic limits to gene regulation by global crosstalk

Gene activity is mediated by the specificity of binding interactions between special proteins, called transcription factors, and short regulatory sequences on the DNA, where different protein species preferentially bind different DNA targets. Limited interaction specificity may lead to crosstalk: a regulatory state in which a gene is either incorrectly activated due to spurious interactions or remains erroneously inactive. Since each protein can potentially interact with numerous DNA targets, crosstalk is inherently a global problem, yet has previously not been studied as such. We construct a theoretical framework to analyze the effects of global crosstalk on gene regulation, using statistical mechanics. We find that crosstalk in regulatory interactions puts fundamental limits on the reliability of gene regulation that are not easily mitigated by tuning proteins concentrations or by complex regulatory schemes proposed in the literature. Our results suggest that crosstalk imposes a previously unexplored global constraint on the functioning and evolution of regulatory networks, which is qualitatively distinct from the known constraints that act at the level of individual gene regulatory elements.   

Biophysical constraints on the computational capacity of biochemical signaling networks

Biophysics fundamentally constrains the computations that cells can carry out. Here, we derive fundamental bounds on the computational capacity of biochemical signaling networks that utilize post-translational modifications (e.g. phosphorylation). To do so, we combine ideas from the statistical physics of disordered systems and the observation by Tony Pawson and others that the biochemistry underlying protein-protein interaction networks is combinatorial and modular. Our results indicate that the computational capacity of signaling networks is severely limited by the energetics of binding and the need to achieve specificity. We relate our results to one of the theoretical pillars of statistical learning theory, Cover’s theorem, which places bounds on the computational capacity of perceptrons.   

Scaling of Adaptive Immune System Repertoires

The adaptive immune system has evolved a stochastic method called VDJ recombination for the purpose of generating the necessary receptor diversity to identify all foreign pathogens. Recent work characterizing the probability distributions of this VDJ recombination process in mouse and human T-cell repertoires shows a massive difference in the corresponding diversities. The increased diversity of the human repertoire is wholly driven by an increase in the average number of nucleotide insertions in VDJ recombination. In this talk the impact of different insertion profiles is quantified and a model for the scaling of such repertoires with respect to the size of the repertoire is laid out.   

Organization of an optimal adaptive immune system.

The repertoire of lymphocyte receptors in the adaptive immune system protects organisms from a diverse set of pathogens. A well-adapted repertoire should be tuned to the pathogenic environment to reduce the cost of infections. I will discuss a general framework for predicting the optimal repertoire that minimizes the cost of infections contracted from a given distribution of pathogens. The theory predicts that the immune system will have more receptors for rare antigens than expected from the frequency of encounters and individuals exposed to the same infections will have sparse repertoires that are largely different, but nevertheless exploit cross-reactivity to provide the same coverage of antigens. I will show that the optimal repertoires can be reached by dynamics that describes the competitive binding of antigens by receptors, and selective amplification of stimulated receptors.   

Inference of selection in the adaptive immune system

The adaptive immune system can recognize many threats by maintaining a large diversity of immune cells with different membrane receptors. This receptor diversity is based on initial random sequence generation, using a recombination mechanism, followed by functional selection stages via interactions with self and foreign peptides. These selection processes shape the initially random receptor ensemble into a functional repertoire that can bind many foreign pathogens. We analyzed high throughput data of human receptor sequences to infer the selection pressures on particular elements of the receptors using maximum likelihood methods. We can quantify the global and site-specific selection pressures and disentangle selection on amino acids from biases in the generated repertoire. We find correlations between generation and initial selection of receptors, and a significant reduction of diversity during selection, suggesting natural evolution of the generating mechanisms.   

The role of disorder in olfactory sensing

Olfactory systems perform the remarkable task of sensing a rich and dynamic space of volatile molecules with a limited set of receptors. To support relevant behaviors, this sensing must provide a faithful embedding of the input space that maps similar odors onto similar receptor representations. We show that disordered sensing---in which a single receptor binds to many odorants, and many receptors bind to any given odorant---is an efficient strategy for faithfully encoding complex odor mixtures with a limited set of receptors. This strategy exploits a key feature of olfactory signals: natural odors are composed of a relatively small combination of all possible monomolecular odorants, and are thus “sparse” in the space of molecules. Importantly, the proposed strategy does not require any fine-tuning to the detailed structure of the olfactory signals. When combined with downstream processing, we show that this strategy supports flexible associations between odor signals and behaviors. Finally, we provide empirical evidence that the olfactory system implements this sensing strategy.   

Speed, Dissipation, and Accuracy in Early T-cell Recognition

In the immune system, T cells can perform self-foreign discrimination with great foreign ligand sensitivity, high decision speed and low energy cost. There is significant evidence T-cells achieve such great performance with a mechanism: kinetic proofreading(KPR). KPR-based mechanisms actively consume energy to increase the specificity of T-cell recognition. An important theoretical question arises: how to understand trade-offs and fundamental limits on accuracy, speed, and dissipation (energy consumption). Recent theoretical work suggests that it is always possible to reduce the the error of KPR-based mechanisms by waiting longer and/or consuming more energy. Surprisingly, we find that this is not the case and that there actually exists an optimal point in the speed-energy-accuracy plane for KPR and its generalizations.   

Environmental adaptation of olfactory receptor distributions

Olfactory sensory neurons (OSNs) in mammals are replaced every few weeks. Each neuron expresses one of hundreds of receptor genes, each with a different binding profile to a wide array of odorants. Experiments show that after replacement, the proportions of OSNs with different receptor types can change. We propose that these changes reflect adaptation of newly-born neurons to the olfactory experience of the animal in order to enhance detection of natural odors. We build a model for olfactory adaptation in which the distribution of receptor types is chosen so that receptor responses form a maximally-accurate representation of odorant concentrations given a fixed total number of OSNs. For small numbers of neurons, the optimal distribution involves a single receptor type. For large numbers of neurons, the distribution becomes almost uniform. In intermediate cases, our model predicts that a variation in olfactory environment should lead to a significant change in the abundances of various receptor types. Such an effect has recently been observed in mice. Our model can be used to predict the change in receptor distribution given a change in olfactory environment, or conversely, to gain insight about the olfactory environment given measured receptor affinities and abundances.   

A competitive binding model predicts the response of mammalian olfactory receptors to mixtures

Most natural odors are complex mixtures of many odorants, but due to the large number of possible mixtures only a small fraction can be studied experimentally. To get a realistic understanding of the olfactory system we need methods to predict responses to complex mixtures from single odorant responses. Focusing on mammalian olfactory receptors (ORs in mouse and human), we propose a simple biophysical model for odor-receptor interactions where only one odor molecule can bind to a receptor at a time. The resulting competition for occupancy of the receptor accounts for the experimentally observed nonlinear mixture responses. We first fit a dose-response relationship to individual odor responses and then use those parameters in a competitive binding model to predict mixture responses. With no additional parameters, the model predicts responses of 15 (of 18 tested) receptors to within $10-30\%$ of the observed values, for mixtures with 2, 3 and 12 odorants chosen from a panel of 30. Extensions of our basic model with odorant interactions lead to additional nonlinearities observed in mixture response like suppression, cooperativity, and overshadowing. Our model provides a systematic framework for characterizing and parameterizing such mixing nonlinearities from mixture response data.   

A model of olfactory associative learning

We propose a mechanism, rooted in the known anatomy and physiology of the vertebrate olfactory system, by which presentations of rewarded and unrewarded odors lead to formation of odor-valence associations between piriform cortex (PC) and anterior olfactory nucleus (AON) which, in concert with neuromodulators release in the bulb, entrains a direct feedback from the AON representation of valence to a group of mitral cells (MCs). The model makes several predictions concerning MC activity during and after associative learning: (a) AON feedback produces synchronous divergent responses in a localized subset of MCs; (b) such divergence propagates to other MCs by lateral inhibition; (c) after learning, MC responses reconverge; (d) recall of the newly formed associations in the PC increases feedback inhibition in the MCs. These predictions have been confirmed in disparate experiments which we now explain in a unified framework. For cortex, our model further predicts that the response divergence developed during learning reshapes odor representations in the PC, with the effects of (a) decorrelating PC representations of odors with different valences, (b) increasing the size and reliability of those representations, and enabling recall correction and redundancy reduction after learning.