Full circumpolar migration ensures evolutionary unity in the Emperor penguin

Defining reliable demographic models is essential to understand the threats of ongoing environmental change. Yet, in the most remote and threatened areas, models are often based on the survey of a single population, assuming stationarity and independence in population responses. This is the case for the Emperor penguin Aptenodytes forsteri, a flagship Antarctic species that may be at high risk continent-wide before 2100. Here, using genome-wide data from the whole Antarctic continent, we reveal that this top-predator is organized as one single global population with a shared demography since the late Quaternary. We refute the view of the local population as a relevant demographic unit, and highlight that (i) robust extinction risk estimations are only possible by including dispersal rates and (ii) colony-scaled population size is rather indicative of local stochastic events, whereas the species' response to global environmental change is likely to follow a shared evolutionary trajectory.

Dispersal and migration. When considering the movements of individuals in a population system, an unfortunate complication often arises because of the convergent choice of the term « migration » to describe very different phenomena, in different conceptual frameworks. The first and most common sense of migration, especially in an avian biology context, is the seasonal movements of groups of individuals between distinct breeding and overwintering grounds.
Migratory species have evolved particular adaptations 1 that allow them to achieve well-timed departure and arrivals to track the most beneficial environments year-round. Such migratory patterns will not be examined in this work, mainly because it is not classically observable in our focal system (although the inter-breeding foraging trips of penguins may arguably be related to migratory behaviour 2 ).
A simpler, one-way movement is the dispersal of individuals out of their original group. individual bringing a set of alleles from one location to another. 8 The second sense of migration, and the one we will use throughout this work, has been described by Dingle and Drake 4 in a biogeographical context as « range expansions of faunas or individual species », such as « the northward extension of ranges following the retreat of glaciers at the end of the ice ages ». More specifically, in a population genetics context, the (mutation-scaled) migration parameter M has been defined by the same authors as « the exchange of genes among populations by whatever means, including but not limited to migration as we consider it here » 4 .
It is used in that sense in the coalescent framework, in particular by Beerli and colleagues 5,6 .
Thus, in that context, migration is distinguished from dispersal by its larger scale: whereas dispersal is an individual-and generation-centred phenomenon that may be observed directly, migration is a time-averaged, population-centred event that is only detectable through indirect methods, such as gene flow reconstruction.

Supplementary Note 2. Re-analysis of mitochondrial DNA data published in Younger et al. 7 and comparison with novel data.
A recent study by Younger and colleagues 7 focused on Emperor penguin mitochondrial DNA population structure. Their conclusion was that colonies from the Ross Sea area are significantly isolated from the rest of the continent, and had a different demographic history. Our genomewide SNP data does not support this view. However, our low sampling size in the Ross Sea region does not permit any definitive conclusion. In order to assess how far this result could be We also re-analysed the data from Younger and colleagues 7 (GenBank accession numbers  Fig. 5B). As a side note, the over-representation of pseudo-replicate sequences in the Ross sea region, by violating the random-sampling assumption of coalescent reconstructions, may also account for the differences in past demographic trends inferences found by the authors (see Figure 2 in Younger et al. 7 ).
Cytochrome-b sequences, on the other hand, showed a standard level of variation in our dataset, as assessed in DnaSP 9 (10 haplotypes, gene diversity = 0.634, nucleotide diversity = 0.003, Fu's Fs = -0.948, Tajima's D = -0.216, non-significant). Haplotype network was built based on Fitch distances between sequences, using Fitchi 10 and a maximum-likelihood bifurcating tree built in RaxML 11 . In keeping with the results of Younger and colleagues 7 , Cytochrome-b sequences do not reflect geographical distribution of the samples in any way ( Supplementary Fig. 5A).
The particular case of the Ross Sea area may require further analysis, as mitochondrial HVR alone seems too unreliably sequenced to provide positive information. The available data only support the extension of a low-level isolation-by-distance model to the whole continent, in keeping with our observations on genome-wide polymorphism. However, the intensity of the gene flow between the Ross Sea region and the rest of the continent has little impact on our ability to model immigration rates at the colony level. Indeed, the origin of the immigration flux is less relevant than its intensity if we are to accurately model population dynamics from colonylevel data.