Session T13: Physics of biological computation across scales

How an immune cell repertoire physically computes

Computing is central to biological function at all scales, from proteins and cells to tissues and organisms. Our immune system encompasses all organizational scales to detect and respond quickly to invading pathogens. While detection is mediated by interaction between molecules, elicited responses rely on a dynamic repertoire of a variety of cells throughout the body, featuring a remarkable ability to adapt on the fly to unexpected challenges. Despite increasing knowledge of the parts, many basic questions about organization principles remain open. How much computation can be achieved on the receptor level for sensing and discrimination? Can active processes inside individual cells influence their collective evolution? What controls adaptability of an immune cell repertoire – both the limits and potential? Guided by in vivo observations, we build a theoretical framework that maps molecular recognition to clonal fitness via active force usage by the cell during signal acquisition. Our results suggest a physical origin for the apparent 'ineffectiveness' of clonal selection in vivo, revealing a multi-faceted role of active forces in limiting response potency while enabling phenotypic plasticity. This framework thus uncovers computational principles – at and across different scales – that balance the depth of response to current infection and breadth of coverage against future variants, within a finite repertoire.   

Author not Attending

From bacteria to humans, adaptive immune systems provide learned memories of past infections. Despite their vast biological differences, adaptive immunity shares features from microbes to vertebrates such as emergent immune diversity, long-term coexistence of hosts and pathogens, and fitness pressures from evolving pathogens and adapting hosts, yet there is no conceptual model that addresses all of these together. To address these questions, we propose and solve a simple phenomenological model of CRISPR-based adaptive immunity in microbes. We show that in coexisting phage and bacteria populations, immune diversity in both populations emerges spontaneously and in tandem, that bacteria track phage evolution with a context-dependent lag, and that high levels of diversity are paradoxically linked to low overall CRISPR immunity. We define average immunity, an important summary parameter predicted by our model, and use it to perform synthetic time-shift analyses on available experimental data to reveal different modalities of coevolution. Finally, immune cross-reactivity in our model leads to qualitatively different states of evolutionary dynamics, including an influenza-like traveling wave regime that resembles a similar state in models of vertebrate adaptive immunity. Our results show that CRISPR immunity provides a tractable model, both theoretically and experimentally, to understand general features of adaptive immunity.   

Optimization and historical contingency in protein sequences

Protein sequences are shaped by functional optimization on the one hand and by evolutionary history, i.e. phylogeny, on the other hand. A multiple sequence alignment of homologous proteins contains sequences which evolved from the same ancestral sequence and have similar structure and function. In such an alignment, correlations in amino-acid usage at different sites can arise from structural and functional constraints due to coevolution, but also from historical contingency.

Correlations arising from phylogeny often confound coevolution signal from functional or structural optimization, impairing the inference of structural contacts from sequences. However, inferred Potts models are more robust than local statistics to these effects, which may explain their success [1]. Dedicated corrections can further increase this robustness [2]. Moreover, phylogenetic correlations can in fact provide useful information for some inference tasks, especially to infer interaction partners from sequences among the paralogs of two protein families. In this case, signal from phylogeny and signal from constraints combine constructively [3], and explicitly exploiting both further improves inference performance [4].

Protein language models have recently been applied to sequence data, greatly advancing structure, function and mutational effect prediction. Language models trained on multiple sequence alignments capture coevolution and structural contacts, but also phylogenetic relationships [5]. They are able to disentangle signal from structural constraints and from phylogeny more efficiently than Potts models [5], and they have promising generative properties [6].   

Structure, function and dynamics of the C. elegans nervous system

The nematode C. elegans’ small nervous system of 302 neurons and its completely mapped anatomical wiring, or connectome, make it a powerful model for exploring biological neural networks. Yet even with this granular anatomical description, it remains challenging to accurately predict brain wide neural dynamics or detailed circuit function because it is unknown for many neural connections how signals propagate from one neuron to the next. To fill this gap, we created a comprehensive neural response map of the C. elegans head at cellular resolution by measuring network calcium activity in response to single-neuron optogenetic perturbations for more than 10,000 neuron-pairs. We captured the sign (excitatory or inhibitory), strength, temporal properties, and the causal direction of signal propagation between neurons. We find that signal propagation in the brain differs from what anatomy predicts. Moreover, simulations constrained by our functional measurements better agree with spontaneous dynamics than do simulations constrained by anatomy. We find that extrasynaptic signaling, including via neuropeptides, is one mechanism underlying this divergence of structure and function.  Because neuropeptide machinery is widely expressed in many nervous systems, our findings may have broader implications for structure-function relations in other organisms.

    

Understanding cellular decisions in latent space

Theoretical biology has historically introduced many physics-inspired metaphors (such as the Waddington landscape), but in the age of molecular biology and network complexity, it is not clear if and how such ideas can be used in a predictive way. In this talk, I will show how multicellular signalling dynamics can sometimes be best represented and modelled in a low dimensional « latent space », offering an alternative modeling approach to standard gene networks. For gap genes during fly development, an auto-encoder model is used to derive a 2D model, allowing for intuitive interpretation of both dynamics and positional information. In the context of immune recognition by T cells, we developed a robotic platform combined to machine learning to exhibit a universal dynamics for cytokines in response to antigens. This model fully explains patterns of immune activations, adversarial interactions in immunotherapy, and can be re-connected a posteriori to molecular interactions. The successful application of a similar concept to two very different contexts suggests this approach can be broadly applied to many biological dynamics.