TL;DR
This paper demonstrates that the fitness and origin time of a growing population can be reliably inferred from its lineage tree, even when key biological details like mutation rates or generation times are unknown or vary. This robustness makes lineage tree analysis a powerful tool for understanding disease progression and evolution.
What Problem Does This Paper Address?
Inferring the "fitness" (growth rate) and origin time of a population is crucial in many biological contexts, such as understanding cancer progression or tracking viral evolution. Traditionally, this inference relies on detailed knowledge of the underlying biological processes, including mutation rates, generation times, and offspring distributions. However, obtaining this precise information is often difficult or impossible. This paper addresses the challenge of inferring fitness and origin time from lineage trees (which represent the ancestry of sampled individuals) without requiring this detailed biological knowledge, showing that these inferences are surprisingly robust to variations in model assumptions.
Methods
The authors employ a combination of theoretical modeling and computational simulations. They utilize the Wright-Fisher model, a standard framework for population genetics, and extend it to incorporate selection. A key aspect of their approach is analyzing the "coalescent statistics" of lineage trees, which describe how lineages merge backward in time. They mathematically demonstrate that these statistics are invariant to changes in several parameters, including the distribution of offspring per individual and the precise timing of generations. They also use diffusion approximations (Fokker-Planck equations) to model population dynamics and derive analytical results. Finally, they validate their theoretical findings using Approximate Bayesian Computation (ABC) on simulated data.
Datasets
Not discussed in the paper. The study relies on simulated data generated according to various models and parameter settings.
Key Insights and Ideas
- Fitness is robustly inferable: The core finding is that population fitness (growth rate per unit time) can be accurately inferred from lineage trees regardless of the specific biological details of reproduction (e.g., number of offspring, generation timing) or the mutation rate.
- Time of onset is also inferable: Similarly, the time at which a population (or subclone) originated can be estimated with high accuracy, with an error of at most one generation.
- Lineage trees are "universal" for fitness: The structure of lineage trees, when scaled appropriately, can look identical for vastly different biological systems (e.g., cancer cells vs. viruses) as long as their fitness is the same. This suggests a universal signature of fitness in genealogical data.
- Multiple competing clones are manageable: The framework extends to scenarios with multiple competing subclones, allowing for the inference of their individual fitnesses and emergence times.
Key Results
The paper demonstrates that:
- Fitness can be inferred without knowing the mutation rate per generation or the exact number of generations.
- The time of onset of a mutation can be inferred with an average error of at most one generation.
- Coalescent statistics of lineage trees are invariant to changes in the distribution of offspring and fluctuations in generation timing.
- Simulations show that lineage trees from different biological systems (cancer vs. virus) are indistinguishable if their fitness values are the same, and distinguishable if their fitness values differ.
- Approximate Bayesian Computation (ABC) successfully infers fitness and age of onset from simulated trees even when the inference model's assumptions about biological details are significantly different from the simulation parameters.
Limitations and Future Work
The authors acknowledge several limitations:
- The model assumes a clonal population or multiple expanding clones where stochastic behavior within each clone is identical. This assumption might break down if generation timing or variance in offspring number changes rapidly within a clone.
- The framework assumes that selectively advantageous mutations do not arise too rapidly, preventing a clone from establishing and escaping extinction before new mutations emerge.
- The mathematical derivations often rely on approximations (e.g., large population size, small selection coefficient) which might introduce deviations in the early phases of clonal expansion.
- The study assumes a constant mutation rate for converting branch lengths to time, or at least that this rate is known as a function of time.
- The current model does not explicitly account for "dormant cancer cells" which might exhibit power-law distributions in generation timing.
Future work could explore these limitations, such as incorporating more complex distributions of generation timing or investigating scenarios with very rapid introduction of new mutations. The authors also suggest that their findings have implications for understanding the impact of incorrect assumptions on inferred parameters.
Summary updated