Local ancestry inference identifies robust evidence of selection in Neolithic Europe

By benchmarking six local ancestry inference methods on Neolithic European genomes, this study demonstrates that while multi-method validation can robustly identify selection signals in genes related to pigmentation and metabolism, inferred ancestry patterns and selection signatures are highly sensitive to the specific method used, particularly in complex regions like the HLA.

The Gist

Imagine the history of Europe during the Neolithic period (about 10,000 years ago) as a massive, ancient potluck dinner. On one side, you have groups of farmers migrating from Anatolia (modern-day Turkey), and on the other, the local hunter-gatherers already living there. When these two groups met, they didn't just swap stories; they mixed their families, creating a new generation with a genetic "salad" of both backgrounds.

This mixing happened at a time when life was changing fast—new foods, new environments, and new ways of living. These changes acted like a strict head chef, demanding that the new mixed population adapt quickly or risk falling behind. Scientists want to know: Which specific ingredients in this genetic salad were kept because they were super useful?

To find these "super ingredients," researchers use a tool called Local Ancestry Inference. Think of this tool as a high-tech detective that looks at a person's DNA and tries to label every single segment: "This part came from the farmer," and "That part came from the hunter-gatherer." If a specific part of the DNA from one group shows up way more often than chance would allow, it suggests that part was selected for because it helped people survive.

The Problem with the Detective Tools
The paper points out a tricky issue: most of these detective tools were built and tested on modern people, where we have huge, clear reference libraries to compare against. But ancient DNA is like a dusty, incomplete library. The data is sparser, the samples are older, and the "reference panels" (the comparison groups) are much smaller.

The researchers asked: Do these detective tools work as well on ancient, messy data as they do on modern, clean data?

The Experiment
To find out, the team acted like a quality control inspector. They took 176 ancient Neolithic genomes and ran them through six different detective methods. It's like hiring six different appraisers to value the same house; you want to see if they all agree on the price.

Here is what they found:

• The Big Picture: All six methods agreed on the general mix. They all said, "Yes, this person has about 60% farmer DNA and 40% hunter DNA."
• The Details: However, when it came to the length of the DNA chunks and exactly when the mixing happened, the methods disagreed wildly. Some said the chunks were short; others said they were long. It was as if one appraiser said the house was built in 1920, and another said 1950.

The Reliable Findings
Because the methods disagreed so much on the details, the researchers decided to only trust the signals that all (or most) methods agreed on. This "consensus" approach helped them filter out the noise and find the real winners:

1. Skin Color (SLC24A5): They found strong, consistent evidence that genes related to lighter skin were selected for. This makes sense, as farmers moved to areas with less sunlight.
2. Diet and Metabolism (FADS1/2): They found clear signs of selection in genes that help process fats and oils, likely because the diet was shifting from wild game to farmed grains and dairy.

The "Maybe" Findings
They also spotted some interesting candidates, like genes for circadian rhythm (PER3) and immune defense (IRAK4), but the detective tools didn't all agree on these. The signal was there, but it was shaky.

The "Messy" Zone
Finally, they looked at the HLA region (a part of the immune system). Previous studies claimed there was an excess of hunter-gatherer DNA here. However, in this study, the six methods gave totally different answers. Some said "yes," others said "no." The researchers concluded that this area is so complex that the detective tools might be getting confused, creating false alarms.

The Bottom Line
This paper teaches us that while we can definitely find real biological signals in ancient DNA, the tool you choose matters a lot. Just like using a different map app might give you different routes, using a different ancestry method can change the story you tell about the past. To get the truth, you can't rely on just one method; you need to cross-check your results with multiple tools to make sure you aren't being misled by the limitations of the data.

Technical Summary

1. Problem Statement

The study addresses a critical gap in the application of Local Ancestry Inference (LAI) to ancient DNA (aDNA). While LAI is a standard tool for detecting natural selection following admixture in modern populations, its performance in the context of ancient genomes remains unvalidated. Specifically, the authors highlight three unique challenges in Neolithic European data:

• Data Sparsity: Ancient genomes often have lower coverage and higher missingness compared to modern datasets.
• Reference Limitations: Reference panels for ancient populations are significantly smaller and less diverse than those available for modern groups.
• Temporal Depth: The admixture events (between Anatolian farmers and European hunter-gatherers) occurred thousands of years ago, leading to shorter ancestry tracts that are harder to resolve.

The core problem is whether current LAI methods, developed for modern data, can reliably reconstruct ancestry tracts and identify selection signals in these complex, ancient scenarios without introducing methodological bias.

2. Methodology

The authors employed a rigorous benchmarking and validation framework:

• Dataset: The primary analysis utilized 176 imputed Neolithic genomes from Europe.
• Method Benchmarking: Six distinct LAI methods were applied to the same dataset. The study compared their outputs across three key metrics:
  1. Ancestry Proportions: The overall percentage of hunter-gatherer vs. farmer ancestry per individual.
  2. Tract Length Distributions: The size of continuous ancestry blocks, which informs admixture timing.
  3. Selection Signatures: Deviations in local ancestry that suggest positive selection.
• Validation Strategy: To ensure robustness, the authors did not rely on a single method. Instead, they:
  • Integrated results across all six methods to identify consensus signals.
  • Replicated findings in two independent, larger datasets ($n=378$ and $n=1,121$).
• Comparative Analysis: They specifically examined regions known to be under selection (e.g., SLC24A5, FADS1/2, HLA) to see if different methods agreed on the presence and direction of selection.

3. Key Contributions

• First Comprehensive Benchmark of LAI in aDNA: The study provides the first systematic evaluation of how different LAI algorithms perform on imputed Neolithic genomes, revealing significant discrepancies in tract length and admixture time estimates.
• Identification of Robust vs. Fragile Signals: The authors distinguish between selection signals that are consistent across multiple methods (robust) and those that are highly method-dependent (fragile).
• Methodological Warning: The paper establishes that while individual-level ancestry proportions are highly correlated across methods, inferred tract lengths and admixture times can vary by over an order of magnitude, fundamentally altering the interpretation of demographic history.

4. Key Results

The analysis yielded specific findings regarding selection and methodological consistency:

• Robust Signals of Selection:

  • Pigmentation (SLC24A5): All methods consistently identified a deviation in ancestry, supporting the hypothesis of strong selection for lighter skin pigmentation in Neolithic Europe.
  • Metabolism (FADS1/2): Consistent signals were found at the FADS1/2 locus, indicating adaptation to dietary shifts (likely related to the transition to agriculture).
  • Candidate Loci: The study identified PER3 (circadian rhythm) and IRAK4 (innate immunity) as potential targets of selection, though the signals were less consistent across methods compared to SLC24A5 and FADS1/2.

• Inconsistent/Problematic Signals:

  • HLA Region: The study replicated previous reports of excess hunter-gatherer ancestry at the Major Histocompatibility Complex (HLA). However, this signal was highly inconsistent across the six methods. The authors conclude that this specific signal is likely an artifact of bias inherent in LAI algorithms when applied to complex, highly polymorphic regions like the HLA, rather than a true biological signal of selection.

• Methodological Discrepancies:

  • While individual ancestry estimates were correlated, the variation in tract length distributions suggests that single-method studies of ancient admixture timing may be unreliable.

5. Significance

This paper fundamentally shifts the standard for analyzing selection in ancient genomes:

1. Validation of Biological Signals: It confirms that LAI can recover biologically meaningful signals of selection in ancient Europe, specifically regarding pigmentation and metabolic adaptation.
2. Caution Against Single-Method Reliance: The findings underscore that method choice critically influences inferred ancestry patterns. A signal detected by one method may be an artifact when viewed through another.
3. New Best Practices: The authors advocate for a multi-method validation approach. To claim a selection signal in ancient DNA, researchers must demonstrate consistency across multiple LAI algorithms and replicate findings in independent datasets.
4. Resolution of Controversies: By identifying the HLA signal as likely methodological bias, the study helps resolve conflicting literature regarding selection in immune-related genes during the Neolithic transition.

In summary, the paper demonstrates that while local ancestry inference is a powerful tool for ancient genomics, its application requires rigorous cross-validation to distinguish true evolutionary adaptation from algorithmic artifacts.