Session K37: Evolutionary Dynamics II

Physical insights into the evolutionary dynamics of viruses

Viruses can replicate and mutate with remarkable speed. When we encounter a virus, whether through natural infection or vaccination, our immune systems develop specialized cells and antibodies to help clear the infection and protect us against future exposure to similar pathogens. However, viruses like HIV, influenza, and SARS-CoV-2 can still evolve to escape immune control. This is one of the main reasons why the immune system can't clear HIV infection, and why vaccines against viruses like influenza and SARS-CoV-2 need constant updates. In this talk, I'll tell you about our recent efforts to use ideas from physics to understand the evolutionary dynamics of pathogens like these. Building predictive models of viral evolution could help us to design better, more effective vaccines and treatments against infectious disease.   

Quantifying microbial fitness under conceptual uncertainty

Few concepts are as central to ecology and evolution as the notion of relative fitness, and yet the quantification of relative fitness in experiments is effectively a matter of choice. For the same data, alternative measures of relative fitness are available and continue to be used side-by-side. Are they all equivalent? What reasons can we give an experimentalist to choose one over the other? Here we develop such an argument, based on a single first principle: relative fitness measures must have the function to predict strain frequencies in the upcoming time window. From this axiom, we characterize the set of possible fitness metrics (it is large!) and settle two questions for high-throughput fitness measurements with barcoded mutant libraries: 1. In batch culture growth cycles, the relative fitness per-cycle provides the best all-around choice while the relative fitness per-generation is restricted to special-use cases and disagrees in the ranking of mutant genotypes. 2. In batch cultures with an entire mutant library, the mutant fitness is amplified compared to pairwise competitions due to higher-order interactions. We calculate this error for a typical trait distribution in Yeast knockouts and provide quantitative rules for the ideal inoculum frequency in barcoding experiments. Altogether, our axiomatic approach recovers all important fitness metrics used in the past and demonstrates that the same information content can be captured in almost arbitrary form, but some metrics are more practical than others.   

Correlations between mutations under recurrent genetic hitchhiking

Sweeping beneficial mutations can dramatically perturb the variation at neighboring sites in the genome ("genetic hitchhiking"). While the effects of isolated selective sweeps have been studied in detail, the cumulative effect of recurrent hitchhiking remains poorly characterized in long recombining genomes. Previous theoretical work has focused on predicting the patterns of genetic diversity at a single neutral site, which can be hard to distinguish from the effects of other population genetic processes. Since sites that hitchhike together share similar frequency dynamics until they are unlinked by recombination, the correlation between sites contains richer information about the evolutionary forces at play. Yet, incorporating strong selection and recombination into the conventional framework of coalescent theory has been a long-standing challenge. Here, we will describe a forward-in-time approach that stratifies the dynamics of variants by their present-day frequencies, which provides new opportunities to predict the multi-site effects of recurrent hitchhiking.   

Strong environmental memory revealed by experimental evolution in static and fluctuating environments

Evolution in a static environment often proceeds via large-effect beneficial mutations that may become maladaptive in other environments. Conversely, natural settings require populations to endure environmental fluctuations. A sensible assumption is that the fitness of a lineage in a fluctuating environment is the time-average of its fitness over the sequence of static conditions it encounters. However, transitions between conditions may pose new challenges, which could cause deviations from the time-average. To test this, we tracked hundreds of thousands of barcoded yeast lineages evolving in static and fluctuating conditions and subsequently isolated 900 mutants for pooled fitness assays in 15 environments. We find that fitness in fluctuating environments indeed often deviates from the expectation based on static components, leading to fitness non-additivity. Moreover, closer examination reveals that fitness in one component of a fluctuating environment is often strongly influenced by the previous component. We show that this environmental memory is especially common for mutants with high variance in fitness across tested environments. We employ a simple mathematical model and whole-genome sequencing to propose mechanisms underlying this effect, including lag time evolution and sensing mutations. Our results demonstrate that environmental fluctuations have large impacts on fitness and suggest that variance in static environments can explain these impacts.   

Detectability of epistasis from temporal genetic data

Epistasis refers to a phenomenon wherein the occurrence of two or more mutations in the same genetic sequence has a different effect than expected from the effects of the individual mutations alone. Epistasis is common in nature and plays an important role in shaping evolution and the speed of adaptation.

In principle, the ability to detect epistasis in data is affected by genetic linkage, the chance cooccurrence of mutations on the same genetic background, and genetic recombination, the exchanges of genetic material between individuals, which acts to dilute linkage. At present, few studies have explored how epistasis can be detected, under what conditions, and which types of epistasis are detectable. Here, we inferred epistasis from simulated temporal genetic data with different types of epistasis and recombination rates. We tested multiple inference methods, including one that efficiently distinguishes linkage and epistasis. We found that different inference approaches work better in different evolutionary regimes, and we derived underlying scaling factors that constrain their accuracy.   

Noise robustness and metabolic load determine the principles of central dogma regulation

Protein expression levels optimize cell fitness: Too low of an expression level of essential proteins slows growth by compromising the function of essential processes, whereas the overexpression of proteins slows growth by increasing the metabolic load. This trade-off naively predicts that the cell maximizes its fitness by a Goldilocks principle in which cells express just enough protein for function; however, this strategy neglects the significance of the inherent stochasticity of the gene expression process which leads to significant cell-to-cell variation in protein numbers. How does the cell ensure robust growth in the face of the inherent stochasticity of these central dogma processes? To explore the consequences of noise in the expression of hundreds of essential genes, we build a minimal model where the trade-off between metabolic cost and growth robustness can be analyzed analytically. The model predicts that growth-rate maximization leads to a highly-asymmetric cost-benefit analysis which drives the optimal protein expression levels far above what is required on average, with low-expression essential proteins expressed at more than 10-fold what is required for growth in the typical cell. This robustness mechanism naturally explains the surprisingly strong buffering effect observed in essential protein depletion experiments and leads to a second qualitative prediction: there is a lower floor on the transcription level of essential genes corresponding to one message per cell cycle. We show that nearly all essential, but not non-essential, genes obey this limit in three evolutionarily-divergent model organisms: Escherichia coli, yeast, and human. The model also pre- dicts that the optimal translation efficiency is roughly proportional to message abundance, predicting the observed relation between proteome fraction and message abundance in eukaryotic cells. This optimal translation efficiency predicts a non-canonical scaling of gene expression noise with protein abundance, which we show is observed in yeast. Together, these results reveal that noise robustness and metabolic load determine the global regulatory principles that govern central dogma function.   

The distribution of the time to the most recent common genetic ancestor of a pair of genomes

For every gene in a diploid genome, the maternal and paternal copies share a common ancestor at some point in the past. Because of recombination, each gene potentially has its own ancestral history and time to the most recent common ancestor (TMRCA). Basic evolutionary theory predicts that the marginal distribution of the TMRCA for each gene is exponential with a mean proportional to the population size, but the joint distribution across genes is unknown. We use asymptotic analysis and simulations to find simple expressions for the distribution of the minimum TMRCA across the genome. We find that the minimum TMRCA is tightly peaked around a value that is proportional to the square root of the population size. For humans, this means that your most closely related pair of genes have a TMRCA that is more than two orders of magnitude smaller than a typical pair.   

Impact of crowding on the diversity of expanding populations

Growing cell populations become densely packed as cells proliferate and fill space. Crowding prevents spatial mixing of individuals, significantly altering the evolutionary outcome from established results for well-mixed populations. Despite the fundamental differences between spatial and well-mixed populations, little is known about the impact of crowding on genetic diversity. With microbial colonies on plates, we show that the allele frequency spectrum is characterized by a power law for low frequencies. Using cell-based simulations and microfluidic experiments, we identify the origin of this distribution in the volume-exclusion interactions within the crowded cellular environment, enabling us to extend these findings to a broad range of dense populations. This study highlights the importance of cellular crowding for the emergence of rare genetic variants.   

Predicting influenza evolution.

Seasonal influenza virus evolves to evade immune recognition, necessitating regular vaccine updates, a process that requires optimization by forecasting influenza evolution. Phylodynamic methods have successfully identified strains similar to dominant strains of the virus in upcoming influenza seasons. However, their capacity to predict allele frequency dynamics or fixation of mutations is very limited. This may result from considering virus evolution via phylogeny, simplifying the actual distribution of sequences in the population over time. Existing approaches focusing on single mutations do not account for the different genetic backgrounds, an essential feature of evolving populations. These simplifications undermine methods' predictability. To address this gap, we developed a physics-inspired method that accounts for the full evolutionary history and genetic background to estimate the fitness advantage of influenza mutations and predict strains that are most likely to be dominant in the future. We applied our approach to estimate the selective effects of mutations, investigate the nature of strongly selected mutations, and search for signals of vaccine-driven selection. Quantitative predictions from our research could facilitate informed vaccine update decisions and produce new data that could be integrated into other forecasting tools. We expect our methods could be extended to study other pathogens with similar modes of evolution.   

A hike through the Protein Evolutionary Landscape

Masked language models can predict a protein's residue preference at a specific position using the complete sequence context. We propose a scheme to calculate the masked residue profile for the entire sequence in a single forward pass with unmasked residue embeddings. This allows us to efficiently calculate the pseudo-perplexity of a sequence, a measure of the model's uncertainty in assigning residues. Naturally occurring proteins typically have a low pseudo-perplexity, and in designed proteins, a low pseudo-perplexity can be used as a scalar proxy for function. We use this as a rough estimate for the fitness of a protein, and explore protein morphospace by generating high fitness paths between homologous proteins. We achieve this by using a directed evolution based approach where we iteratively select low pseudo-perplexity mutants that are most proximal to the target sequence. Navigation towards the target is achieved by using an alignment procedure powered by language model embeddings, and proposals for the mutations are drawn from the masked residue profile of the sequence. We cross-validate the reliability of the interpolated paths by folding the sequences using alphafold, and consistently achieve high pLDDT scores.   

Emergence of Meaning in the Sensor Evolution of Anteorganisms

Semantic Information Theory is a recent framework that codifies meaning as the efficacy with which a living system maintain themselves far-from-equilibrium, viability being the most primitive form of meaning. It has been succesfully applied to agent-based foraging models, identifying how morphology influences the quantity of information about the environment needed to maintain viability. Anteorganisms are proposed autopoeitic systems that bridge the gap between non-life and proto-organisms, with simple dynamics resembling complex viability-based behavior requiring neither agency nor sensors. In this work we apply the framework of Semantic Information Theory to a model of an anteorganism whose behavior is chemotactic, a sensorless forager. We allow for an initial sensisitivity to environmental degrees of freedom correlated to the presence of "food," and use a competetive genetic algorithm, to drive the evolution of a refined sensor. Keeping track of the information flow from the environment to the agent using transfer entropy, we quantify the number of bits that are meaningful and how that amount changes as the morphology of the sensor evolves. Our results elucidate how evolutionary processes drive the emergence of meaning, and how the growtth of this informational architecture may be the jump in complexity between abiotic processes and proto-organisms.