A previously reported bottleneck in human ancestry 900 kya is likely a statistical artifact

Yun Deng, Rasmus Nielsen, Yun S. Song

Abstract It was recently reported that a severe ancient bottleneck occurred around 900 thousand years ago in the ancestry of African populations, while this signal is absent in non-African populations. Here, we present evidence to show that this finding is likely a statistical artifact.

Introduction

In 2023, a study by Hu et al. utilized a method called FitCoal to argue that ancient human ancestors suffered a catastrophic population crash—a 98.7% reduction to just ~1,300 individuals—approximately 900,000 years ago (kya). This finding presented an evolutionary puzzle: the signal was detected in African populations but was absent in non-African groups. Given that African and non-African lineages only diverged within the last 100,000 years, any event occurring 900 kya should theoretically be visible in both lineages.

This paper, authored by researchers from UC Berkeley and the University of Copenhagen, re-evaluates these claims by addressing the fundamental mathematical challenge of identifiability in population genetics. Reconstructing the deep past from modern genetic data is difficult because many different historical scenarios can produce nearly identical genetic signatures today.

Methods

The researchers employed a "counter-example" strategy to test the reliability of the FitCoal method. First, they used a tool called mushi to find a "synthetic" demographic history that lacked a sharp bottleneck but produced a Site Frequency Spectrum (SFS)—a summary of genetic variation—nearly identical to the one used by Hu et al.

They then simulated genetic data (216 haplotypes of 100 Mb length) based on this "no-bottleneck" synthetic model using msprime, mimicking the Yoruba (YRI) population data from the 1000 Genomes Project. Finally, they benchmarked three inference tools—Relate, MSMC2, and FitCoal—on this simulated data, specifically using the original scripts and procedures from the Hu et al. study to ensure a fair comparison.

Results

The analysis reveals that the SFS is a "low-dimensional" summary of genomic data, meaning it loses significant information. As shown in Figure 1A, the SFS of a severe bottleneck model and a mild decline model are virtually indistinguishable. This lack of identifiability means that small amounts of noise or slight model misspecifications can lead to wildly different conclusions.

Comparison of Site Frequency Spectra and demographic inference across different methods.

When tested against the "Ground Truth" synthetic model (which featured only a mild decline), both Relate and MSMC2 correctly identified the general population trend. However, FitCoal falsely inferred a sharp, severe bottleneck (Figure 1B). This demonstrates that the previously reported 900 kya bottleneck is likely not a biological reality but a "statistical artifact" produced by the FitCoal algorithm's inherent bias toward over-fitting and favoring extreme scenarios.

Discussion

Reconstructing ancient demography remains a critical research challenge, but this study serves as a cautionary tale regarding the limits of statistical inference. The authors highlight that the original study failed to provide robust measures of statistical uncertainty.

Future work in the field must focus on developing valid measures of uncertainty for demographic inferences and moving beyond the SFS to higher-dimensional data summaries that offer better identifiability. Ensuring that new methods are robust against "hallucinating" extreme events when presented with noisy or slightly different data is essential for accurate reconstructions of human evolutionary history.

Summary updated

Abstract

It was recently reported that a severe ancient bottleneck occurred around 900 thousand years ago in the ancestry of African populations, while this signal is absent in non-African populations. Here, we present evidence to show that this finding is likely a statistical artifact.
BibTeX
@inproceedings{deng2024previously,
  author = {Yun Deng and Rasmus Nielsen and Yun S. Song},
  title = {{A previously reported bottleneck in human ancestry 900 kya is likely a statistical artifact}},
  booktitle = {Genetics},
  year = {2024},
  doi = {10.1093/genetics/iyae192},
}
Figure