Inference of population demographic history captures differing evolutionary signals based on the number of individuals in the dataset

This study demonstrates that demographic inference results are critically dependent on sample size, as varying the number of individuals can shift inferred evolutionary signals from ancient contractions to recent expansions by altering which historical epoch dominates the coalescent branch lengths.

The Gist

Imagine you are trying to reconstruct the history of a family by looking at a photo album. But instead of photos, you have a collection of tiny genetic clues (DNA) left behind by ancestors. Scientists use these clues to figure out if the family once shrank to just a few people (a bottleneck) and then exploded in size (an expansion).

This paper is about a tricky problem: The size of the photo album changes the story you tell.

The Problem: Too Few or Too Many Photos?

The researchers found that the number of people (individuals) you include in your study acts like a camera lens with a changing focus.

• Small Sample Size (A blurry, zoomed-out lens): If you only look at a few people, the genetic clues make it look like the family had a long, slow decline in the distant past. It's like looking at a forest from far away and only seeing the bare, old trees, missing the new saplings.
• Large Sample Size (A sharp, zoomed-in lens): If you look at hundreds of people, the story flips. The clues suddenly scream, "We just had a baby boom!" It's like walking into that same forest and suddenly seeing thousands of new, young trees that were invisible from afar.

The Experiment: A Tale of Two Epochs

To test this, the scientists created a "fake" family history with two distinct chapters:

1. Chapter 1 (Ancient Times): A terrible famine caused the population to crash (a bottleneck).
2. Chapter 2 (Recent Times): The survivors recovered and the population grew rapidly.

They then ran their detective work on this fake history using different numbers of "witnesses" (sample sizes).

• With few witnesses: The detectives concluded, "This family has been shrinking for a long time." They missed the recent boom.
• With many witnesses: The detectives concluded, "This family is exploding with new life!" They missed the ancient crash.

The "Why": The Branch Length Analogy

Why does the lens change the story? The authors explain this using a family tree analogy.

Imagine the history of a population as a giant, upside-down tree. The roots are the ancient past, and the branches are the present.

• Ancient events (like the famine) affect the thick, main trunk of the tree.
• Recent events (like the boom) affect the tiny, delicate twigs at the very tips.

When you have a small sample, you are mostly grabbing the thick trunk. You see the old, heavy history clearly, but the tiny new twigs are too small to notice.
When you have a large sample, you are grabbing so many branches that you finally catch all those tiny, new twigs. The "recent boom" becomes the loudest signal, drowning out the quiet history of the old trunk.

The Takeaway

The paper teaches us that there is no single "correct" sample size that tells the whole truth.

• If you only look at a few people, you might think a population is dying out.
• If you look at many, you might think it's thriving.

The Solution: Don't just pick one number and stick with it. The best way to understand a population's full history is to look at the data through multiple lenses (different sample sizes). By doing this, you can see both the ancient tragedy (the bottleneck) and the modern triumph (the expansion), piecing together the complete, complex story of the family's past.

Technical Summary

Problem Statement

Accurate reconstruction of population demographic history is a fundamental objective in population genetics, yet it faces significant challenges regarding methodological robustness. A critical issue identified is the sensitivity of inference methods to two variables: the number of individuals (sample size) included in the dataset and the specific demographic scenario assumed during the modeling process. Specifically, the Site-Frequency Spectrum (SFS), a primary summary statistic used to infer demographic history, has been shown to yield qualitatively different results—such as inferring population expansion versus contraction—depending heavily on the sample size. This variability creates uncertainty in interpreting evolutionary signals and suggests that current inference frameworks may be capturing different aspects of history based solely on data volume rather than biological reality.

Methodology

The authors employed a rigorous approach combining simulation and empirical analysis to investigate this phenomenon:

• Datasets: The study utilized two simulated datasets and one empirical dataset. All datasets were characterized by a specific demographic history: an ancient population bottleneck (contraction) followed by a recent population expansion.
• Modeling Framework: The researchers fitted a two-epoch demographic model to the data.
• Variable Manipulation: A key experimental variable was the sample size. The model was fitted across a continuous range of sample sizes to observe how the inferred parameters and evolutionary signals shifted as the number of individuals increased.
• Statistical Analysis: Beyond the SFS, the study analyzed other summary statistics, specifically Tajima's D and the proportion of singletons, to track changes in genetic variation patterns relative to sample size.
• Theoretical Explanation: To explain the observed shifts, the authors analyzed the coalescent branch lengths, specifically determining which demographic epoch contributed the highest mean proportion of branch lengths to the genealogy.

Key Contributions

• Identification of Sample Size Bias: The study formally demonstrates that demographic inference is not static; it is critically dependent on the number of individuals analyzed.
• Mechanistic Explanation: The authors provide a theoretical mechanism for these shifts, linking the change in inferred signals to the dominance of specific coalescent branch lengths. They show that different sample sizes effectively "weight" different time periods in the population's history, causing the epoch with the highest mean proportion of branch lengths to dictate the inferred signal.
• Methodological Recommendation: The paper proposes a novel analytical strategy: rather than relying on a single large dataset, researchers should analyze datasets at multiple sample sizes to uncover complementary aspects of population history that might otherwise be obscured.

Results

• Signal Reversal: When fitting the two-epoch model, the inferred evolutionary signal shifted dramatically based on sample size.
  • Small Sample Sizes: The inference predominantly captured signals of ancient population contraction (the bottleneck).
  • Large Sample Sizes: The inference shifted to capture signals of recent population expansion.
• Consistency in Summary Statistics: This shift was not isolated to the SFS. Other statistics, including Tajima's D and the proportion of singletons, exhibited parallel changes as the sample size increased, confirming that the phenomenon affects multiple dimensions of genetic variation.
• Coalescent Correlation: The results confirmed that the shift in inference aligns with which demographic epoch contributes the most to the total coalescent branch length in the genealogy for a given sample size.

Significance

This research highlights a critical caveat in population genetics: demographic inference is not absolute but relative to the sample size.

1. Interpretation of Past Studies: It suggests that conflicting results in previous literature regarding population history (e.g., whether a population expanded or contracted) may stem from differences in sample sizes rather than biological discrepancies.
2. Optimization of Analysis: It advocates for a multi-scale approach to data analysis. By analyzing data at varying sample sizes, researchers can disentangle ancient bottlenecks from recent expansions, gaining a more holistic and accurate view of the full demographic trajectory.
3. Model Robustness: The findings urge the development of inference methods that are robust to sample size variations or explicitly account for the weighting of coalescent events across different time epochs.