Inferring hominin history with recurrent gene flow from single unphased genomes and a two-locus statistic

This study introduces a novel inference framework using multi-population two-locus statistics derived from single unphased genomes to reconstruct complex hominin demographic history, revealing multiple episodes of gene flow between early modern humans and Neanderthals, introgression from an unsampled lineage into Denisovans, and specific ancestry patterns in early European farmers.

The Gist

Imagine human history not as a straight line of ancestors, but as a massive, tangled ball of yarn. For a long time, scientists tried to untangle this ball by looking at single strands (individual genes). But to truly understand how the ball got so knotted, you need to look at how two strands twist together.

This paper introduces a new, powerful tool to untangle the history of our ancient human relatives—Neanderthals, Denisovans, and early modern humans—using a method that works even when we only have a tiny, messy piece of yarn to examine.

Here is the breakdown of their discovery in everyday language:

1. The Problem: The "One-Person" Puzzle

Usually, to figure out how populations mixed and moved, scientists need huge groups of people today. It's like trying to solve a jigsaw puzzle with 1,000 pieces; you can see the big picture.

But with ancient DNA (from fossils), we often only have one person's genome from a specific time and place. It's like trying to solve that same puzzle with only one piece. Most modern tools break when you give them just one piece. Plus, ancient DNA is often "unphased," meaning we can't tell which gene came from mom and which from dad—it's like a scrambled deck of cards where you can't see the suits.

2. The Solution: The "Two-Look" Statistic

The authors developed a new math trick called $H_2$.

• The Old Way (One-Look): Imagine looking at a single lightbulb to guess how the power grid works. You can see if it's on or off, but you don't know much about the wiring.
• The New Way (Two-Look): Now, imagine looking at two lightbulbs that are wired together. If they flicker in sync, you know they are on the same circuit. If they flicker differently, they are on different circuits.

The authors realized that even with just one ancient person, you can look at two spots on their DNA at the same time. By seeing how often those two spots "agree" or "disagree" with each other, they can reconstruct the history of the whole population. It's like deducing the entire layout of a city's subway system just by watching two trains move on the same track for a short while.

3. What They Found: The Ancient "Mix-and-Match"

Using this new tool on DNA from Neanderthals, Denisovans, and early humans, they uncovered a much more complex family tree than we thought.

• The "Back-Door" Visits: We knew modern humans (Africans) left Africa and met Neanderthals. But this paper shows that early modern humans actually visited Neanderthals twice.
  • Analogy: Think of it like a tourist visiting a foreign country. Most people think the tourist went there once. This study says, "No, the tourist went there, came back, and then went there again 100,000 years later."
• The "Ghost" Ancestor: Denisovans (a mysterious ancient human group) have DNA from a "ghost" lineage—a group of humans we haven't found fossils for yet.
  • Analogy: Imagine you are looking at a family photo album, and you see a great-grandparent who looks like no one else in the family. You know they exist because of a specific feature they passed down, but you've never seen their face. The authors found evidence of this "Ghost" ancestor mixing with Denisovans.
• The "Basal" Branch: Early European farmers (like the Stuttgart individual) weren't just a mix of local hunter-gatherers. They had a huge chunk of ancestry from a "Basal Eurasian" group that split off from everyone else before the main migration out of Africa.
  • Analogy: Imagine a family tree where one branch splits off so early that it never met the Neanderthals. When the farmers arrived in Europe, they brought this "Neanderthal-free" DNA with them, diluting the Neanderthal DNA found in other Europeans.

4. Why This Matters

Before this, scientists had to guess these stories based on limited data. This paper proves that even with one ancient skeleton, we can mathematically prove that these complex mixing events happened.

• The "Confounding" Trap: The authors also showed that if you try to figure out one event (like a visit from modern humans) without accounting for another (like the "Ghost" ancestor), your math gets confused. It's like trying to figure out how much sugar is in a cake without knowing if there's also honey in it. You have to taste for both at the same time to get the recipe right.

The Big Picture

This paper is like upgrading from a black-and-white, blurry photo of our ancestors to a high-definition, 3D movie. It tells us that human history wasn't a simple straight line of "us" replacing "them." Instead, it was a chaotic, messy, and fascinating dance where different groups of humans (and even groups we haven't found yet) kept bumping into each other, swapping DNA, and creating the complex family tree we are part of today.

In short: They built a new microscope that works on tiny, ancient samples, and it revealed that our ancient relatives were far more social, mobile, and genetically mixed than we ever imagined.

Technical Summary

1. Problem Statement

The study addresses the challenge of inferring complex demographic histories (divergence, population size changes, and gene flow) using Ancient DNA (aDNA). While aDNA has revolutionized our understanding of hominin evolution, it presents specific technical hurdles for statistical inference:

• Scarcity and Quality: Samples are often sparse, low-coverage, and chemically damaged, preventing long-read phasing.
• Sample Size Limitations: Many methods require large sample sizes or phased haplotypes to estimate higher-order statistics (e.g., $E[D^2]$ requires four haplotypes).
• Model Identifiability: Distinguishing between complex scenarios, such as ancestral population structure versus simple population size changes, or multiple episodes of gene flow, is difficult using standard one-locus diversity statistics (like $F_{ST}$ or heterozygosity) or methods like PSMC, which often assume panmixia.

The authors aim to develop a framework capable of modeling recurrent gene flow and ancestral structure using single, unphased diploid genomes, a common scenario in aDNA studies.

2. Methodology

The core innovation is the development and application of a new family of multi-population two-locus statistics, denoted as $H_2$.

• The $H_2$ Statistic:

  • Defined as the probability of observing differences between two sampled genome copies at two loci.
  • Crucially, the authors derive an estimator for $H_2$ that works on single unphased diploid genomes. By averaging over all possible phasings (assuming equal probability), they can estimate the statistic without needing haplotype resolution.
  • $H_2$ is a linear combination of summaries from the Hill-Robertson (HR) system, capturing the covariance of pairwise coalescence times ($Cov(T_L, T_R)$) between linked loci.
  • Unlike one-locus statistics, $H_2$ decays as a function of recombination distance ($\rho$), providing information about the joint distribution of genealogies. This decay pattern is sensitive to historical events like population bottlenecks, expansions, and, most importantly, ancestral population structure and introgression.

• Inference Framework:

  • Model Fitting: The authors use a composite likelihood approach. They bin site pairs by genetic distance (recombination fraction $r$) and compare observed $H_2$ values against model expectations generated by the moments.LD engine (an extension of the HR system).
  • Model Choice: To rigorously test hypotheses (e.g., "Did gene flow occur?"), they employ a Bootstrap Likelihood Ratio Test (BLRT). Since the null distribution of the likelihood ratio statistic for composite likelihoods is unknown, they use simulations to generate an empirical null distribution and calculate $p$-values.
  • Data Handling: They account for local variation in recombination and mutation rates by weighting site pairs. They utilize two distinct recombination maps (Bhérer et al., 2017; Zhou et al., 2020) to test robustness.

• Dataset: The study analyzes eight individuals:

  • 1 Contemporary Human (Yoruba).
  • 3 Ancient Eurasian AMH (Ust'Ishim, Loschbour, Stuttgart).
  • 3 Neanderthals (Altai, Chagyrskaya, Vindija).
  • 1 Denisovan (Denisova-3).

3. Key Contributions

1. Methodological Breakthrough: Demonstrated that complex demographic parameters (divergence times, migration rates, introgression proportions) can be jointly inferred from single unphased diploid genomes using two-locus statistics. This bypasses the need for large sample sizes or phasing, which are often impossible for ancient samples.
2. Identifiability of Ancestral Structure: Showed theoretically and empirically that $H_2$ can distinguish between models of ancestral population structure (subdivided demes) and panmictic population size changes that produce identical marginal coalescence time distributions.
3. Joint Inference of Recurrent Gene Flow: Developed a framework to disentangle multiple, overlapping gene flow events that are often confounded when modeled separately.

4. Key Results

The authors inferred a comprehensive demographic model integrating seven ancient hominin individuals and one modern human.

• Neanderthal-AMH Gene Flow:

  • Confirmed two distinct episodes of gene flow from Anatomically Modern Humans (AMH) into Neanderthals:
    1. An early pulse into the ancestral Neanderthal population (250 kya) with an introgression proportion of **5.4%**.
    2. A later pulse into the Western Neanderthal lineage (110 kya) with a proportion of **1.15%**.
  • Joint modeling was essential; modeling these separately led to confounding and inflated estimates.

• Denisovan "Superarchaic" Introgression:

  • Detected evidence of gene flow from an unsampled "Superarchaic" lineage (potentially related to Homo erectus) into Denisovan ancestors (~700 kya).
  • The inferred proportion is ~1.35%.
  • The study highlights that failing to model this "ghost" introgression biases estimates of AMH-Neanderthal gene flow, demonstrating the necessity of joint inference.

• Basal Eurasian Ancestry in Early Farmers:

  • Modeled the ancestry of the Stuttgart farmer (Neolithic Europe).
  • Found that ~57.5% of Stuttgart's ancestry derives from a Basal Eurasian (BE) lineage that diverged from other non-African AMH before the Neanderthal introgression event.
  • This explains why early European farmers show reduced Neanderthal ancestry compared to hunter-gatherers (Loschbour).

• Population Structure and Size:

  • Inferred a Neanderthal-Denisovan split time of ~688 kya and a divergence from AMH of ~798 kya.
  • Observed population size collapses in sampled Neanderthal demes, consistent with PSMC curves, but noted that these patterns can also be explained by spatial structure rather than just panmictic bottlenecks.

5. Significance

• Resolving Complex Histories: The paper proves that two-locus statistics provide the statistical power necessary to resolve "reticulate" evolutionary histories (networks of gene flow) that are invisible to one-locus methods.
• Leveraging Sparse Data: By enabling the use of single unphased genomes, this method unlocks the potential of the vast archive of ancient DNA, which is often limited to single high-coverage individuals per population.
• Correcting Biases: The study demonstrates that ignoring "ghost" introgression (e.g., from Superarchaics) or ancestral structure leads to significant errors in estimating other parameters (like AMH-Neanderthal admixture rates).
• Future Applications: The framework is flexible and can be applied to other systems with sparse sampling or to combine ancient and modern datasets, offering a path toward a more unified understanding of human evolutionary history.

In summary, Collier et al. provide a robust statistical toolkit that transforms the limitations of ancient DNA (unphased, single samples) into a strength for detecting subtle, recurrent gene flow events that shaped the hominin lineage.