A Common Genetic Thread Through Time and Space
SNP-level allele frequency trends in Europe and East Asia
When we talk about polygenic selection over deep time, we often focus on single regions such as Europe, East Asia, or a specific archaeological sequence. But a deeper question is rarely asked: do the same genetic variants tend to change over time in different parts of the world?
As far as I am aware, no study has explicitly tested this. Ancient DNA has given us increasingly detailed genetic time series, yet these series are almost always analyzed in isolation. The possibility that SNP-level temporal dynamics might replicate across continents has been largely ignored.
In this context, genetic variants refer to single-nucleotide polymorphisms, or SNPs, which are individual DNA positions where people can carry different bases. Most complex traits are influenced by thousands of such variants, each with a very small effect.
If selection is acting on shared biological constraints, we might expect temporal allele frequency trends to be at least partly aligned across populations. If instead dynamics are mostly local and contingent, then even SNPs associated with the same trait may show largely unrelated trajectories.
This distinction also provides a clear falsification test. If the time series inferred from ancient DNA were a statistical fluke, an artefact of sampling, or unrelated to the trait they are meant to predict, there would be no reason to expect any systematic agreement across continents. In that case, SNPs would trend up or down independently in different populations, and any apparent temporal signal would fail to replicate outside the region where it was first observed.
To examine this, I use ancient DNA samples from Europe (EUR) and East Asia (EAS), spanning roughly the last 12,000 years. For each population, I bin samples in time and estimate allele frequencies for SNPs associated with educational attainment, using two GWAS-derived polygenic scores (PGS), EA3 and EA4.
A naïve approach would be to regress allele frequency on time and compare slopes across populations. Raw slopes, however, depend on the scale of allele frequencies. A SNP moving from 2 percent to 4 percent and one moving from 20 percent to 40 percent have very different slopes despite identical directional behavior.
Instead, I focus on the correlation between allele frequency and time. This statistic is scale free and captures whether a SNP tends to increase or decrease monotonically over time. Importantly, it is equivalent to a standardized regression coefficient, which makes it comparable across SNPs.
For each SNP, I compute two quantities: the correlation between allele frequency and time in Europe, and the correlation between allele frequency and time in East Asia. Each SNP is therefore represented by a pair of numbers, (rEUR,rEAS).
The central question then becomes: are SNPs that trend upward or downward over time in Europe also likely to do so in East Asia?
A positive answer would suggest that ancient DNA time series are capturing a real, trait relevant signal that generalizes across populations. A negative answer would instead point toward predominantly local dynamics, or toward temporal patterns that do not meaningfully predict educational attainment at all.
Do SNPs Change in the Same Direction Over Time?
A very direct way to assess whether temporal allele-frequency dynamics are shared across populations is to ignore effect size altogether and focus only on direction. For each SNP, I compute the correlation between allele frequency and time separately in Europe and East Asia. Each SNP therefore has a sign—positive if the allele tends to increase over time, negative if it tends to decrease.
If SNP-level temporal trends were unrelated across populations, then a SNP would be just as likely to move in opposite directions as in the same direction. Under this null, 50% of SNPs should show concordant signs purely by chance.
That is not what the data show. Instead, a modest but consistent excess of concordant SNPs emerges for both GWAS definitions:
EA3: 1,799 out of 3,265 SNPs (55.1%) show the same sign of temporal change in Europe and East Asia
EA4: 2,148 out of 3,920 SNPs (54.8%) show concordant signs
Although these deviations from 50% are small in absolute terms, they are highly statistically significant. A binomial test strongly rejects the null of random sign agreement (EA3: p ≈ 6 × 10⁻⁹; EA4: p ≈ 2 × 10⁻⁹).
The modest size of the effect is expected. At the level of individual loci, allele-frequency trajectories are dominated by drift, sampling noise, and local demographic history. Selection typically shifts frequencies only weakly at each SNP, so strong concordance is not anticipated. Nevertheless, the consistent excess of same-direction changes across thousands of loci indicates that temporal dynamics are not entirely population-specific.
This directional test sets a lower bound on shared structure. In the next section, I move beyond sign agreement and ask whether the strength of temporal trends is also aligned between Europe and East Asia.
