population history - senecio7の日記

Gutenkunst RN, Hernandez RD, Williamson SH & Bustamante CD 2009 Inferring the joint demographic history of multiple populations from multidimensional SNP frequency data. PLoS Genet 5:e1000695.

combining our demographic model with a previously estimated distribution of selective effects among newly arising amino acid mutations accurately predicts the frequency spectrum of nonsynonymous variants across three continental populations (YRI, CHB, CEU)

African and Eurasian populations diverged around 100,000 years ago
this is earlier than other genetic studies suggest
our model includes the effects of migration, which we found to be important for reproducing observed patterns of variation in the data
we find no evidence for recurrent migration after East Asian and Native American populations diverged

simulations incorporating linkage are necessary to estimate variances and define critical values for hypothesis testing and model selection

numerics
we solve the diffusion equation on a regular nonuniform grid, using a finite difference scheme [36]
inspired by the method of Chang and Cooper [37]

data
we considered all diallelic SNPs in 5.01 Mb of sequence from noncoding regions of 219 autosomal genes

to estimate the ancestral allele, we aligned to the panTro2 build of the chimp genome
like other methods based on the unfolded AFS, our analysis is sensitive to errors in identifying the ancestral allele

the human-chimp divergence in the data is 1.13%
we assumed a divergence time of 6 My [45] and a generation time of 25 years
this yielded an estimated neutral mutation rate of μ = 2.35 × 10^{− 8} per site per generation
which is comparable to direct estimates [46]
there is some controversy as to the appropriate generation time to assume in human population genetic studies
the human generation time may differ between cultures and may have changed during our biological and cultural evolution

results
first, we explored how various demographic forces affect the AFS, building intuition for our subsequent applications to real data
we then compared the performance of diffusion versus coalescent methods for evaluating the AFS, finding that the diffusion approach is substantially faster
we then applied our diffusion approach to infer parameters for plausible demographic models for the history of continental human populations
we first considered the expansion of humans out of Africa and then the settlement of the New World
finally in an application incorporating selection, we predicted the distribution of nonsynonymous variation between populations in our Out of Africa model

computational performance
∂a∂i
Diffusion Approximations for Demographic Inference

the computational advantage of the diffusion method is even larger when placed in the context of parameter optimization
unlike the coalescent approach, there is no simulation variance
efficient derivative-based optimization methods can be used
the fits of three-population models with roughly a dozen parameters typically took a few hours to converge from a reasonable initial parameter set
this speed allows us to use extensive bootstrapping to estimate variances, overcoming the limitations of composite likelihood

expansion out of Africa
our analysis of human expansion out of Africa used data from three HapMap populations
12 Yoruba individuals from Ibadan, Nigeria (YRI)
22 CEPH Utah residents with ancestry from northern and western Europe (CEU)
12 Han Chinese individuals sampled in Beijing, China (CHB)

previous analyses found that the YRI spectrum is well-fit by a two-epoch model with ancient population growth [5,17]
we found this as well
previous analyses of the CEU and CHB populations found that both populations went through bottlenecks [5,11] concurrent with divergence
such models qualitatively fit our marginal CEU-CHB spectrum
combining these demographic features yields the model illustrated in Figure 2B
allowing for asymmetric gene flow yielded very little improvement in fit, as did allowing for growth in the Eurasian ancestral population or allowing the CEU and CHB bottleneck and divergence times to differ
our composite likelihood function assumes that polymorphic sites are independent
it thus overestimates the number of effective independent data points
to control for linkage, we performed both conventional and parametric bootstraps

the times for growth in the African ancestral population and divergence of the Eurasian ancestral population (T_AF and T_B) have particularly wide confidence intervals, likely a consequence of the high inferred migration rate m_AF–B between the African and Eurasian ancestral populations
T_AF shows high correlation with the ancestral population size N_A
T_B shows no strong linear correlation with any other single parameter
N_AS0 < N_EU0
the CHB population suffered a more severe bottleneck than the CEU population [11]

the improvement in fit to the real data upon adding contemporary migration to the model is much larger than would be expected if there were no such migration
the contemporary migration we infer is highly statistically significant
omitting ancient migration (m_AF–B) reduced fit quality even more
the data also demand substantial ancient migration

settling the New World
a model in which the CEU and CHB diverge from an equilibrium population did not reproduce the AFS well
a model allowing a prior size change in the ancestral population better fit the AFS but very poorly fit the observed LD decay
reproducing the AFS does not guarantee reproduction of LD

nonsynonymous polymorphism
diffusion approaches are particularly useful for studying such nonsynonymous polymorphism, because they easily incorporate selection
the diffusion approximation assumes that sites are unlinked
nonsynonymous segregating sites are rare enough that this is often a reasonable approximation

discussion
Keinan et al. [11] inferred no significant migration between CEU and CHB
our inferred divergence time of ≈22 kya between East Asians and Mexican-Americans is somewhat older than the oldest well-accepted New World archeological evidence
the divergence we infer may reflect the settlement of Beringia, rather than the expansion into the New World proper
the divergence time of ≈140 kya we infer between African and Eurasian populations is consistent with archeological evidence for modern humans in the Middle East ≈100 kya
it is much older than other inferences of ≈50 kya divergence from mitochondria DNA
this discrepancy may be explained by our inclusion of migration in the model
migration preserves correlation between population allele frequencies
an observed correlation across the genome can be explained by either recent divergence without migration or ancient divergence with migration
the sampled populations may not best represent those in which historically important divergences occurred

we consider only noncoding sequence in fitting our historical model
selection on regulatory or linked coding sites may skew the AFS

the AFS encodes substantial demographic information
the AFS does not capture all the information in the data
patterns of linkage disequilibrium encode additional information