Inferring Migration Networks with Time-Lagged F2 Statistics

This paper introduces a novel method that leverages time-lagged F2 statistics and linear regression on ancient DNA time-series data to infer directional migration rates and reconstruct dynamic pan-European migration networks over the last 6,000 years.

The Gist

Imagine you are trying to figure out how people moved around Europe thousands of years ago. Usually, historians look at ancient DNA like a snapshot: they take a picture of a population in the year 3000 BCE and another in 2000 BCE, then guess what happened in between. But this is like looking at two photos of a crowded party and trying to guess who danced with whom without seeing the music or the movement.

This paper introduces a new, clever way to watch the "movie" of human history instead of just looking at the still photos.

The Core Idea: The "Genetic Echo"

The authors developed a method to track migration (people moving) by listening to the "echo" of genetic changes over time.

Think of two neighboring villages, Village A and Village B.

• No Migration: If no one ever visits, the people in Village A will slowly change their genetic makeup just by chance (like a deck of cards getting shuffled). Village B does the same, but in a totally different, random way. Their "genetic stories" are unconnected.
• Migration: If people start walking from Village A to Village B, the "story" of Village A starts to leak into Village B. The genes in Village B begin to look more like the genes in Village A.

The tricky part is that ancient DNA is often noisy and incomplete. It's like trying to hear a conversation in a crowded, windy room where you only catch a few words here and there.

The New Tool: "Time-Lagged F2 Statistics"

The authors created a mathematical tool (a fancy way of measuring genetic distance) that acts like a time-delayed echo.

1. The Standard Way: Usually, scientists compare Village A today with Village B today.
2. The New Way: This method compares Village A from 200 years ago with Village B today.

The Analogy of the Leaky Bucket:
Imagine Village A has a bucket of blue water (genes) and Village B has a bucket of red water.

• If you pour water from A to B, the red water in B slowly turns purple.
• If you look at the buckets now, you see purple.
• But if you look at what A looked like 200 years ago and compare it to what B looks like now, you can see exactly how fast the blue water was leaking in.

By measuring this "lag" (the delay between when the genes left one place and when they arrived in another), the researchers can calculate the direction and speed of the migration. It tells them not just that people moved, but who moved where and when.

What They Found: The Great European Migrations

They applied this method to ancient DNA from the last 6,000 years and found some fascinating stories:

1. The Neolithic Wave (The Farmers Arrive)
They confirmed the well-known story that farmers moved from the Middle East (Levant) into Europe. Their method showed a strong, one-way flow: genes moved from the East to the Southeast of Europe, but not the other way around. It was a clear, directional river of people.

2. The Steppe Migration (The "Bronze Age" Boom)
This was the big discovery. Around 3,000 BCE, a massive migration happened from the Pontic Steppe (modern-day Ukraine/Russia) into Europe.

• The Old Story: People thought the Steppe people marched straight into Britain.
• The New Story: The "movie" shows a relay race. The Steppe people moved into Central Europe first. There, they mixed with local farmers. Then, their mixed descendants (who now carried a blend of Steppe and European farmer genes) moved on to Britain.
• The Takeaway: The people who arrived in Britain weren't "pure" Steppe nomads; they were a second-generation mix that had already picked up local European DNA along the way.

Why This Matters

Before this, we mostly knew the starting point and the ending point of history. We knew who was there at the beginning and who was there at the end.

This new method is like getting a GPS tracker for ancient populations. It allows scientists to:

• See the direction of movement (who went where).
• See the timing (when the waves hit).
• See the mixing (how populations blended as they traveled).

In a Nutshell

The authors built a mathematical "time machine" that uses the tiny, random changes in our DNA to reconstruct the movement of ancient populations. Instead of guessing based on static snapshots, they can now watch the flow of human history like a river, seeing exactly where the water (genes) came from, where it went, and how it mixed along the way. This helps us understand that human history isn't just a series of static events, but a dynamic, flowing network of movement and connection.

Technical Summary

1. Problem Statement

Reconstructing human demographic history, specifically the dynamics of gene flow and migration, is a central challenge in archaeogenetics. While ancient DNA (aDNA) has revolutionized our understanding of past populations, existing methods face significant limitations:

• Static vs. Dynamic: Most current tools estimate static ancestry proportions or effective migration surfaces based on time-aggregated data. They fail to capture how mobility patterns evolve continuously over time.
• Data Sparsity and Noise: aDNA datasets are often sparse, unevenly distributed in time and space, and contain high levels of missing data and sampling noise.
• Model Assumptions: Many methods require assumptions about the existence or composition of ancestral source populations, which can be difficult to verify.
• Complexity: Modeling discrete, long-distance migration events is difficult using continuous diffusion models (PDEs), while tracking individual-level events lacks population-level quantitative interpretation.

The authors aim to develop a method to infer directional migration rates and reconstruct time-resolved migration networks directly from sparse, time-series aDNA data without relying on predefined ancestral populations.

2. Methodology

The authors propose a novel framework based on the linear time evolution of the expectation of neutral allele frequencies.

Core Concept: Time-Lagged $F_2$ Statistics

The method leverages the $F_2$ statistic, a standard measure of genetic distance (mean squared allele frequency difference) between populations. The authors introduce a time-lagged extension, denoted as $F'_{AB}(t)$, which measures the genetic distance between population $A$ at time $t'$ and population $B$ at time $t$ (where $t' = t + \Delta t$).

Mathematical Derivation

1. Dynamics Model: The authors model the frequency of neutral alleles in a metapopulation of $n$ sub-populations. They assume that the allele frequency in population $i$ at time $t+\Delta t$ is a linear combination of frequencies in all populations at time $t$, plus a noise term (genetic drift):
   $$X_i(t+\Delta t) = \sum_{j=1}^n A_{ij} X_j(t) + \eta_i$$
   Here, $A_{ij}$ is the backward-migration matrix, representing the proportion of individuals in population $i$ at $t+\Delta t$ that originated from population $j$ at time $t$.

2. Linear Relationship: By deriving the dynamics in terms of $F_2$ statistics, the authors show that the change in genetic distance is linearly related to the migration matrix. Specifically, they derive the following equation (Eq. 8 in the paper):
   $$F'_{ik} - F'_{ii} = \sum_{j} A_{ij} (F_{jk} - F_{ji})$$

   • $F_{jk}$ and $F_{ji}$ are standard $F_2$ distances at time $t$.
   • $F'_{ik}$ and $F'_{ii}$ are time-lagged distances.
   • Crucially, the genetic drift term ($\eta$) cancels out in this formulation, making the method robust to drift.

3. Inference Strategy:

   • The equation above forms a system of linear equations where the unknowns are the elements of the migration matrix $A$.
   • The problem is solved via constrained linear regression (specifically, non-negative least squares with a simplex constraint where rows sum to 1).
   • Robustness: To handle aDNA noise, the method uses block-bootstrapping of SNPs to estimate confidence intervals and accounts for dating uncertainty via temporal resampling.

Data Processing

• Interpolation: To address gaps in the aDNA record, the authors employ a convolution-based interpolation technique to estimate $F_2$ and $F'_2$ statistics across continuous time windows.
• Goodness-of-Fit: A Pearson correlation test between the left and right sides of the derived equation is used to identify time periods where the model fits the data well (indicating sufficient sample density).

3. Key Contributions

1. Time-Resolved Migration Networks: The first method to infer directional migration rates as a function of time directly from allele frequency time series, rather than static ancestry proportions.
2. Drift-Independent Formulation: By utilizing the difference of time-lagged statistics, the method mathematically eliminates the confounding effect of genetic drift, a major hurdle in demographic inference.
3. Minimal Assumptions: The framework does not require prior knowledge of ancestral source populations or assumptions about admixture proportions.
4. Scalability: The linear-time interpolation and pairwise-statistic framework allow the method to scale efficiently to large aDNA datasets.
5. Robustness to Sparsity: The method successfully infers migration signals even with small sample sizes and high missing data rates by integrating information across multiple time slices.

4. Results

The authors validated the method using simulations and applied it to real aDNA data covering the last 6,000 years.

Simulation Validation

• Accuracy: In Wright-Fisher simulations with 4 populations and realistic sampling noise (20% missing data), the inferred migration matrix showed a high Pearson correlation ($\rho = 0.9$) with the ground truth.
• Comparison: The $F_2$-based method consistently outperformed naive least-squares estimators applied directly to allele frequencies, particularly under high noise and low sampling regimes.

Application 1: Early Neolithic Expansion (6800–5600 BCE)

• Scenario: Migration from the Levant/Anatolia into Southeastern Europe.
• Finding: The method successfully detected a strong, unidirectional gene flow from the Levant to Southeastern Europe, consistent with archaeological records. The reverse flow was negligible.

Application 2: Pan-European Migration Network (4000–1000 BCE)

• Scenario: Analyzing 9 regions across Europe to study the Steppe migration (Corded Ware/Bell Beaker).
• Key Findings:
  • Steppe to Central Europe: Confirmed a massive influx from the Western Pontic Steppe into Central Europe starting around 3000 BCE.
  • Indirect Migration to Britain: The study revealed that Britain received Steppe ancestry indirectly via Central Europe, rather than directly from the Steppe. The peak influx into Britain occurred later than in Central Europe, suggesting a "wave of advance" where migrating groups acquired local ancestry along the route before reaching the British Isles.
  • Regional Dynamics: Identified complex interactions, such as gene flow from Iberia to Italy around 3000 BCE and the relative isolation of the Levant during this period.
  • Confidence Intervals: The method provided rigorous confidence intervals, highlighting periods of low confidence (e.g., early periods for Scandinavia) due to data sparsity.

5. Significance and Implications

• Dynamic History: This work shifts the paradigm from static "ancestry snapshots" to dynamic "migration movies," allowing researchers to track how ancestry is redistributed through time.
• Mechanistic Insight: It provides a quantitative framework to test hypotheses about range expansions, such as whether migrations are direct or involve intermediate admixture (as seen in the Britain vs. Steppe results).
• Future Potential: As aDNA datasets grow in spatial and temporal density, this method offers a scalable tool to reconstruct global migration networks over millennia.
• Limitations & Future Work: The current model detects migration primarily through the reduction of pre-existing genetic divergence. It cannot detect exchanges between already well-mixed populations. Future iterations could incorporate de novo mutations or rare variants to improve resolution.

In summary, Isacchini et al. present a mathematically rigorous, computationally efficient, and robust method to decode the "genetic archive" of human history, revealing the directionality and timing of major migration events that were previously difficult to resolve.