A genetic and linguistic analysis of the admixture histories of the islands of Cabo Verde

We considered four competing genetic-admixture scenarios described in Figure 6 and Table 2, tested separately for individuals born on each Cabo Verdean island and for the 225 Cabo Verde-born unrelated individuals grouped altogether, with MetHis–ABC machine-learning scenario-choice and posterior parameter inferences (Fortes‐Lima et al., 2021; Pudlo et al., 2016; Csilléry et al., 2012). ABC inferences are based on 42 summary statistics (Table 3), calculated for each simulation under each competing scenario separately using 60,000 independent autosomal SNPs in Cabo Verdean individuals, the African Mandinka and the European Iberian-IBS proxy source populations.

Figure 6

Download asset Open asset

Note that we did not explicitly simulate genotype data in the European and African source-populations. Instead, we built gamete reservoirs at each 21 generation of the forward-in-time admixture process, matching in frequency the observed allele frequencies at the 60,000 independent SNPs for the Iberian IBS and Mandinka populations, respectively. As in our previous MetHis-ABC investigation of the admixture history of the African-American ASW and Barbadian ACB populations (Fortes‐Lima et al., 2021), we therefore consider that the African and European proxy populations at the source of the admixture history of Cabo Verde are large and unaffected by mutation during the 21 generations of the admixture process; this assumption is reasonable provided that we consider only independent genotyped SNPs and the very recent demographic history of the archipelago, discovered un-inhabited and first settled in the 1460s. Therefore, although we cannot reconstruct the evolutionary history of the African and European source populations with our design, we nevertheless implicitly take the real demographic histories of these source populations into account in our simulations, as we use observed genetic patterns themselves, the product of this demographic history, to create the virtual source populations at the root of the admixture history of Cabo Verde.

We find that the summary-statistics calculated from the observed datasets fall well within the space of summary-statistics obtained from 10,000 simulated-datasets under each of the four competing scenarios (Appendix 1—figure 3, Appendix 1—figure 3—figure supplements 1–10), considering non-significant (α>5%) goodness-of-fit, visual inspection of summary-statistics PCA-projections, and each summary-statistic’s distribution, for each Cabo Verdean birth-island and for all Cabo Verde-born individuals grouped in a single population, separately. Prior-checks thus demonstrate that MetHis simulations are appropriate for further ABC scenario-choice and posterior parameter inferences using observed data in the African Mandinka and the European Iberian IBS source populations and each Cabo Verde islands separately or grouped altogether, as they allow to mimic the observed summary-statistics, despite the assumption that the European and African proxy source populations are at the drift-mutation equilibrium over the last 21 generations.

Overall (Figure 7B), MetHis-RF-ABC scenario-choices indicate that multiple pulses of admixture from the European and African source populations (after the founding admixture pulse, two independent admixture pulses from both Africa and Europe: ‘Afr2Pulses-Eur2Pulses’ scenarios, Figure 6 – Scenario 1), best explain the genetic history of individuals born on six of nine Cabo Verdean islands. Furthermore, we find that even more complex scenarios involving a period of recurring admixture from either source best explain the history of the remaining three islands. Scenarios with periods of recurring admixture from both Africa and Europe (‘AfrRecurring-EurRecurring’, Figure 6 – Scenario 4) are the least favored across Cabo Verde.

Figure 7 with 3 supplements see all

Download asset Open asset

Figure 7—source data 1

Historical landmark chronology for the peopling history of Cabo Verde as provided by previous historical work, respectively for each island.

https://cdn.elifesciences.org/articles/79827/elife-79827-fig7-data1-v3.xlsx

Download elife-79827-fig7-data1-v3.xlsx

RF-ABC cross-validation prior-errors for each of the 40,000 simulations used, in-turn, as pseudo-observed data indicate a reasonably good, albeit not perfect, discriminatory power of the RF (Appendix 1—figure 4A). RF-ABC scenario-choices identify the correct scenario in the majority of cross-validations for most scenarios and most islands. Furthermore, asymmetrical scenarios are the least confused with one-another (AfrRecurring-Eur2Pulses vs Afr2Pulses-EurRecurring, or Afr2Pulses-Eur2Pulses vs AfrRecurring-EurRecurring). As expected and previously shown empirically with MetHis-RF-ABC scenario-choice (Fortes‐Lima et al., 2021; Robert et al., 2010), these results are consistent with increased assignation-errors in the parts of the parameter-space where the different scenarios are highly nested and thus biologically equivalent. Finally (Appendix 1—figure 4B), the mean, variance, skewness, kurtosis, minimum, and maximum of individual admixture proportions’ distributions are systematically among the most informative summary-statistics for RF-ABC scenario-choice in every island or in Cabo Verde as a whole, consistently with theoretical expectations (Fortes‐Lima et al., 2021; Verdu and Rosenberg, 2011).

Finally, when considering all Cabo Verde-born individuals as a single random-mating population without distinguishing birth-islands, our MetHis-RF-ABC scenario-choice identifies the Afr2Pulses-Eur2Pulses scenario as the winning scenario (Appendix 1—figure 4B), thus consistent with the scenario most often found as the winner among Cabo Verde islands considered as the target admixed population in nine separate MetHis-RF-ABC analyses.

For individuals on each Cabo Verdean birth island separately, we performed NN-ABC joint posterior parameter estimation based on 100,000 MetHis simulations under the winning scenario (Fortes‐Lima et al., 2021): Afr2Pulses-Eur2Pulses in Santiago, Fogo, Santo Antão São Nicolau, Brava, and Maio; Afr2Pulses-EurRecurring in Boa Vista and Sal; and AfrRecurring-Eur2Pulses in São Vicente (Figure 7B). For each island separately, detailed posterior parameters’ distributions, Credibility Intervals (CI), and cross-validation errors are provided in Figure 7—figure supplements 1–3 and Appendix 5—Tables 1–10. We synthesized our results considering median point-estimates and 50% CI for each scenario parameter in the admixture history of each island in Figure 7C–D. We detailed our results and discussion for admixture history inferences for each island separately in Appendix 5, in the light of historical data about the peopling of Cabo Verde (Figure 7—source data 1).

Figure 7C shows that the reproductive population size of Cabo Verde islands remained very low until a strong increase in the last three generations for all islands but Santiago and Brava. In Santiago the population expansion was more linear since the founding of Cabo Verde in the 1460s, while the reproductive size of Brava remained almost constant and low until today.

In summary, Figure 7D shows that European and African admixture events throughout the archipelago occurred first during the early peopling history of each island, before the mid-17th century massive expansion of the TAST due to the expansion of the plantation economy (Eltis and Richardson, 2015). We find that other admixture events from Europe or Africa, or both, likely occurred much later during, or immediately after, the 19th century abolition of the TAST and of slavery in European colonial empires. Altogether our MetHis-ABC results support limited historical admixture having occurred in Cabo Verde during the most intense periods of the TAST between the mid-17th and early 19th centuries. Furthermore, note that we find admixture events often earlier than, or concomitant with, the first perennial peopling of an island. For the islands of Santiago, Fogo, Santo Antão and, to a lesser extent, São Nicolau, initial historical admixture events occurred synchronously to the first perennial settlement of the island. For the islands of Brava, Maio, Boa Vista, and São Vicente, early admixture events occurred long before their first perennial peopling, thus showing that their founding was already largely composed of already admixed individuals. Importantly, note that our MetHis-NN-ABC posterior parameter inferences cannot infer all scenario-parameters accurately, as some parameters hardly depart from their respective priors (Figure 7—figure supplements 1–3), and the admixture history of the island of Sal remains overall poorly inferred.

Interestingly, MetHis-NN-ABC posterior parameter inference results obtained for the 225 Cabo Verde-born individuals grouped in a single random-mating population instead of separately for each island of birth, are largely undifferentiated from their prior distributions, and have very wide CI and large cross-validation errors, for all admixture-history parameters except for the two parameters associated with the most recent pulse of admixture from the African source (Figure 7—figure supplements 1–3; Appendix 5—table 10). This contrasts with the substantial number of informative posterior-parameter estimations obtained for all islands of birth separately except Sal (Figure 7, Figure 7—figure supplements 1–3; Appendix 5—tables 1–9), despite the much smaller sample sizes used in each one of these separate analyses compared to the analysis considering Cabo Verde as a single population. These results further show that the history of admixture substantially differs across Cabo Verde islands and that considering the Cabo Verde archipelago as a single random mating population is inadequate to successfully infer the parameters of its admixture history, consistently with our results from ADMIXTURE, LAI, ROH, and Isolation-By-Distance analyses.

Numerous enslaved-African populations from Western, Central, and South-Western Central Africa were forcibly deported during the TAST to both Cabo Verde and the Americas, as shown by historical demographic records (Eltis and Richardson, 2015; Carreira, 2000). There is still extensive debate about whether enslaved-Africans remained or more briefly transited in Cabo Verde during the most intense period of the TAST, in the 18th and 19th centuries, when the archipelago served as a slave-trade platform between continents (Carreira, 2000; Patterson, 1988; Brooks, 2006); the question of the duration of stay of enslaved individuals at a given location being also of major interest throughout the Americas during the TAST (Eltis, 2002; Berlin, 2010). In this context, previous genetic studies considering a relatively limited number of populations from mainland Europe and Africa, and/or limited numbers of Cabo Verdean islands of birth, suggested that mainly continental West Africans and South Europeans were at the root of Cabo Verde genetic landscape (Micheletti et al., 2020; Verdu et al., 2017; Beleza et al., 2013).

In this context, our genetic results favor scenarios where mostly certain West Western African Senegambian populations only (Mandinka and Wolof in our study) contribute to the genetic makeup of Cabo Verde (Figures 2–3). Other Western, Central, and South-Western African populations historically also forcibly deported during the TAST seem to have had very limited contributions to the genomic diversity of most Cabo Verde islands, and virtually no contribution to that of Brava, Fogo, and Santiago.

This could be due to Cabo Verde being only a temporarily waypoint for these latter enslaved-African populations between Africa, the Americas, and Europe, but would also be consistent with additional socio-historical processes (see below). Interestingly, and further echoing these genetic results, the Cabo Verdean Kriolu language carries specific signatures mainly from the Mande language-family, and Wolof and Temne languages from Western Africa, and largely more limited signatures of Kikongo and Kimbundu Bantu languages from Central and South-Western Africa (Quint, 2000; Lang, 2009).

These results contrast with the admixture patterns identified in other enslaved-African descendant populations in the Americas in our dataset (African-American and Barbadian, Figures 2–3), and in previous studies (Micheletti et al., 2020; Ongaro et al., 2019; Mathias et al., 2016; Gouveia et al., 2020; Martin et al., 2017). Indeed, the origins of African ancestries in numerous populations throughout the Caribbean and the Americas traced to varied continental African regions, from Western to South-Western Africa, thus qualitatively consistent with the known diversity of slave-trade routes used between continents and within the Americas after the Middle Passage.

After the initial settlement of Cabo Verde by Portuguese migrants, temporary changes of European dominion in certain islands, newly developing commercial opportunities, and intense piracy during the 16th and 17th centuries have triggered different European populations to migrate to the archipelago (Albuquerque and Santos, 1991; Albuquerque and Santos, 1995; Soares, 2011).

Nevertheless, these latter historical events do not seem to have left a major signature in the genetic landscape of Cabo Verde (Figures 2–3). Instead, we find that Cabo Verdean-born individuals in our dataset share virtually all their European genetic ancestry with Iberian populations, with extremely limited evidence of contributions from other European regions, consistent with previous studies (Micheletti et al., 2020; Verdu et al., 2017). Interestingly, the reduced diversity of European populations’ contributions to the genomic landscape of Cabo Verde is also identified in other enslaved-African descendant populations in our study, as well as in previous studies in Caribbean and American populations (Moreno-Estrada et al., 2013; Baharian et al., 2016; Gouveia et al., 2020). Our results thus show that European admixture in enslaved-African descendant populations on both sides of the Atlantic, as represented here by Cabo Verde, the Barbadians ACB and the African American ASW, mainly stem from the gene-pool of the European empires historically and socio-economically dominant locally, rather than subsequent European migrations (Fortes-Lima and Verdu, 2021).

Altogether, note that, in our local-ancestry inferences, we considered as reference source populations varied existing populations from continental mainland Africa and Europe, categorized as such from sampling information and geographic location only, prior to any genetic investigation. Therefore, in these analyses, we cannot disentangle the fraction of European admixture in Cabo Verdean genomes stemming directly from European migrations after the 1460s, from the fraction stemming from the European genetic legacy in continental African source populations acquired whether during more ancient migrations which occurred before the peopling of Cabo Verde (e.g. Busby et al., 2016), or since then during the European colonial expansion in Sub-Saharan Africa. Symmetrically, we cannot disentangle the fraction of African admixture in Cabo Verde stemming directly from continental Africa after the 1460s, from the fraction inherited from Africa-Europe admixture events that may have occurred in Europe prior to the peopling of Cabo Verde or during the colonial era until today. Disentangling both genetic heritages will require, in future studies, the explicit modelling of such possible admixture histories within African and European ancestral populations at the source of the Cabo Verde genetic landscape, and would also benefit from including data from North-African populations in our reference panels.

While we expected recurring African admixture processes due to the known history of regular forced migrations from Africa during most of Cabo Verde history (Carreira, 2000), our MetHis–ABC scenario-choices indicate that, qualitatively, these demographic migrations did not necessarily translate into clearly recurring gene-flow processes to shape genetic patterns in most of Cabo Verde islands (Figure 7B). Indeed, African admixture processes in all islands, except São Vicente, seem to have occurred during much more punctual periods of Cabo Verdean history. Our MetHis–ABC posterior parameter estimations further highlighted often largely differing admixture histories across Cabo Verde islands (Figure 7C–D and Appendix 5).

We find that admixture from continental Europe and Africa occurred first early during the TAST history, concomitantly with the successive settlement of each Cabo Verdean island between the 15th and the early 17th centuries (Figure 7D). Furthermore, we find that the most intense period of enslaved-African deportations during the TAST via Cabo Verde, between the mid-17th and early-19th centuries during the expansion of the plantation economic system in the Americas and Africa (Eltis and Richardson, 2015; Carreira, 2000; Albuquerque and Santos, 1991; Albuquerque and Santos, 1995), seem to have left a limited signature in the admixture patterns of most Cabo Verdean islands today. Interestingly, previous studies also highlighted that admixture in enslaved-African descendants in the Caribbean may have occurred first early in the European colonial expansion in the region in the 16th century, and then much later towards the end of the TAST at the end of the 18^th century, and had thus been relatively limited during most of the plantation economy era (Moreno-Estrada et al., 2013; Fortes-Lima et al., 2018). Together with our results, this illustrates the apparent discrepancy between intense demographic forced migrations during the TAST and genetic admixture signatures at least in certain populations on both sides of the Atlantic. Indeed, in contrast with these results, numerous other enslaved-African populations in the Americas have, instead, shown substantial historical admixture inferred to have occurred during the plantation economy era, hence exemplifying the diversity of admixture histories experienced locally by enslaved-Africans descendant populations during the TAST (Ongaro et al., 2019; Baharian et al., 2016).

The inferred lack of admixture events in Cabo Verde during the height of the TAST could be due to newly deported enslaved-Africans being only transiting via Cabo Verde before being massively re-deported to other European colonies during this era (Carreira, 2000; Patterson, 1988; Brooks, 2006). Furthermore, and not mutually exclusive with this latter hypothesis, historians reported, in Cabo Verde and other European colonies in the Americas, that relationships between enslaved and non-enslaved communities largely changed with the expansion of plantation economy at the end of the 17th century. These changes are often referred to as the shift from Societies-with-Slaves to Slave-Societies (Eltis and Richardson, 2015; Berlin, 2009; Chaudenson, 2001). Slave-Societies legally enforced the socio-marital and economic segregation between communities, and coercively controlled relationships between new enslaved-African migrants and pre-existing enslaved-African descendant communities more systematically and violently than Societies-with-Slaves (Berlin, 2009)^{p.15-46,95-108},(Carreira, 2000)^p.281-319. The high prevalence of segregation during the height of the plantation economy could have limited genetic admixture between enslaved-African descendants and non-enslaved communities of European origin, as well as admixture between new migrants, forced or voluntary, and pre-existing Cabo Verdeans; notwithstanding the known history of dramatic sexual abuses during and before this era. This could consistently explain our observations of a relative lack of diverse African or European origins in Cabo Verdean genomes despite the known geographical diversity of populations deported and emigrated to the archipelago throughout the TAST. Furthermore, with legally enforced segregation, we might expect more marital pairings than before to occur among individuals with common origins; i.e. between two individuals with the same continental African or European origin. Such ancestry-specific marriages triggered by socio-cultural segregation would be consistent with our ROH patterns (Figure 4), also depending on how long such mate-choice behaviors persisted after the end of legal segregation. We note, however, that we have not formally tested this influence on ROH and ancestry patterns and that a careful consideration of alternate explanations, such as temporally varying admixture contributions over time or a severe bottleneck in one of the ancestral populations, would be important to consider in such future analyses.

In this context, the diversified African ancestries here found in the Americas (Figures 2–3), consistently with previous studies showing admixture events occurring before or during the height of plantation economy in the Americas (Micheletti et al., 2020; Ongaro et al., 2019; Baharian et al., 2016), would suggest that the gene-pool of enslaved-Africans communities admixing with Europeans in the Americas since the 16^th century often involved, at a local scale, multiple African source populations, thus reflecting the multiple slave-trade routes between continents and within the Americas. Conversely in Cabo Verde, the early onset of the TAST during the 15th century likely privileged commercial routes with nearby Senegambia (Carreira, 2000)^{p.31-54,281-319}, thus favoring almost exclusively admixture events with individuals from this region and from certain populations only. Altogether, our results in Cabo Verde contrasting with certain other enslaved-African descendant populations in the Americas, highlight the importance of early admixture processes and socio-cultural constraints changes on intermarriages throughout the TAST, which likely durably influenced genomic diversities in descendant populations locally, on both sides of the Atlantic.

Finally, we find that recent European and African admixture in Cabo Verde occurred mainly during the complex historical transition after the abolition of the TAST in European colonial empires in the 1800s and the subsequent abolition of slavery between the 1830s and the 1870s (Figure 7D). Historians have shown that these major historical shifts strongly disrupted pre-existing segregation systems between enslaved and non-enslaved communities (Eltis and Richardson, 2015)^-p.271-290, (Carreira, 2000)^-p.335-362, (Cooper et al., 2000). In addition, an illegal slave-trade flourished during this era, bringing numerous enslaved-Africans to Cabo Verde outside of the official routes (Carreira, 2000)^-p.363-384 (Conrad, 1969). Altogether, our results indicate increased signals of European and African admixture events in almost every island of Cabo Verde during this period, and were thus consistent with a change of the social constraints regarding admixture and forced displacements of enslaved-descendants that had prevailed over the preceding 200 years of the TAST. These results were largely consistent with previous studies elsewhere in the Caribbean (Moreno-Estrada et al., 2013; Fortes-Lima et al., 2018), and Central and South America (Ongaro et al., 2019; Gouveia et al., 2020); showing, at a local scale, the major influence of this recent and global socio-historical change in inter-community relationships in European colonial empires on either sides of the ocean.

The 261 Cabo Verdean individuals were each interviewed to record a classical familial anthropology questionnaire on self-reported life-history mobility. In particular, we recorded primary residence location, birth location, parental and grand-parental birth and residence locations, and history of islands visited in Cabo Verde and foreign experiences (Verdu et al., 2017). Furthermore, we also recorded age, sex, marital status, and cumulative years of schooling and higher education (for 211 individuals only), and languages known and their contexts of use.

GPS coordinates for each reported location were acquired on site during field-work, supplemented by paper maps and Google Earth for non-Caboverdean locations and islands that we did not visit (Sal and Maio). While participants’ birth-locations were often precise, increasing levels of uncertainty arose for the reported parental and grand-parental locations. We arbitrarily assigned the GPS coordinates of the main population center of an island when only the island of birth or residence could be assessed with some certainty by participants. All other uncertain locations where recorded as missing data. Figure maps were designed with the software QuantumGIS v3.10 ‘București’ and using Natural Earth free vector and raster map data (https://www.naturalearthdata.com).

The 261 participants each provided 2 mL saliva samples collected with DNAGenotek OG-250 kits, and complete DNA was extracted following manufacturer’s recommendations. DNA samples were genotyped using an Illumina HumanOmni2.5Million-BeadChip genotyping array following manufacturer’s instructions. We followed the quality-control procedures for genotypic data curation using Illumina’s GenomeStudio Genotyping Module v2.0 described in Verdu et al., 2017. Genotype-calling, population-level quality-controls, and merging procedures are detailed in Appendix 1—figure 1.

In summary, we extracted a preliminary dataset of 259 Cabo Verdean Kriolu-speaking individuals, including relatives, genotyped at 2,118,835 polymorphic di-allelic autosomal SNPs genome-wide. We then merged this dataset with 2504 worldwide samples from the 1000 Genomes Project Phase 3 Auton et al., 2015; with 1307 continental African samples from the African Genome Variation Project (Gurdasani et al., 2015; Network et al., 2019) (EGA accession number EGAD00001000959); and with 1235 African samples (Patin et al., 2017; Perry et al., 2014) (EGA accession number EGAS00001002078). We retained only autosomal SNPs common to all data sets, and excluded one individual for each pair of individuals related at the 2nd degree (at least one grand-parent in common) as inferred with KING (Manichaikul et al., 2010), following previous procedures (Verdu et al., 2017).

After merging all datasets (Appendix 1—figure 1), we considered a final working-dataset of 5157 worldwide unrelated samples, including 233 unrelated Cabo Verdean Kriolu-speaking individuals, of which 225 were Cabo-Verde-born, genotyped at 455,705 autosomal bi-allelic SNPs (Figure 1; Figure 1—source data 1). Note that, for this working-dataset, the fraction of missing genotypes was very low and equaled 7.0 × 10^–4 on average within Cabo Verdean islands of birth (SD = 3.0 × 10^–4 across islands).

We collected linguistic data for each of the 261 Cabo Verdean individuals using anthropological linguistics questionnaires, and semi-directed interviews. Each participant was shown a brief (~6 min) speech-less movie, ‘The Pear Story’ (Chafe, 1980), after which they were asked to narrate the story as they wanted in ‘the Kriolu they speak every day at home’. Discourses were fully recorded without interruption, whether individuals’ discourses were related to the movie or not. Then each discourse was separately fully transcribed using the orthographic convention of the Cabo Verdean Kriolu official alphabet “Alfabeto Unificado para a Escrita da Língua Cabo-verdiana (ALUPEC)” (Decreto-Lei n° 67/98, 31 de Dezembro 1998, I Série n° 48, Sup. B. O. da República de Cabo Verde).

Building on the approach of Verdu et al., 2017, we were interested in inter-individual variation of “ways of speaking Kriolu” rather than in a prescriptivist approach to the Kriolu language. Thus, we considered each utterance as defined in Croft, 1996^p.107: “a particular instance of actually-occurring language as it is pronounced, grammatically structured, and semantically interpreted in its context”. Transcripts were parsed together and revealed 4831 (L=4831) unique uttered items among the 92,432 uttered items transcribed in total from the 225 discourses from the genetically-unrelated Cabo Verde-born individuals. To obtain these counts, we considered phonetic, morphological, and syntactic variation of the same lexical root items that were uttered/pronounced differently, and we excluded from the utterance-counts onomatopoeia, interjections, and names. Note that we were here interested in the diversity of realizations in the Kriolu lexicon, including within the same individuals. In other words, we are interested in both between-speaker and within-speaker variation. Also note that a very limited number of English words were pronounced by particular individuals (10 utterances each occurring only once), and were kept in utterance-counts.

The list of unique uttered items was then used to compute individual’s specific vectors of uttered items’ relative frequencies as, for each genetically unrelated Cabo Verde-born individual i (in [1, 225]) and each unique uttered item l (in [1, L=4831]), $f_{i,l}=n_{i,l}/{\sum }_{j=1}^L{n}_{i,j}$, where n_i,l is the absolute number of times individual i uttered the unique item l over her/his entire discourse, f_i being the vector $\left(f_{i,1},f_{i,2},\dots ,f_{i,L}\right)$. We compared vectors of individuals’ utterance-frequencies by computing the inter-individual pairwise Euclidean distance matrix as, for all pairs of individuals i and j, $d\left(f_i,f_j\right)=\sqrt{{\sum }_{l=1}^L{\left(f_{i,l}-f_{j,l}\right)}^2}$, (Verdu et al., 2017).

We categorized each of the 4831 unique uttered items separately into five linguistic categories (Verdu et al., 2017). Category A included only unique utterances directly tracing to a known African language and comprised 88 unique items occurring 3803 times out of the 92,432 utterances. Category B included only items with a dual African and European etymology, that is items of a European linguistic origin strongly influenced in either meaning, syntax, grammar, or form by African languages or vice versa, attesting to the intense linguistic contacts at the origins of Cabo Verdean Kriolu, and comprised 254 unique items occurring 6960 times. Category C included 4432 items (occurring 73,799 times) with strictly Portuguese origin and not carrying identifiable traces of significant African linguistic origin or influence. Category D included 26 items occurring 6762 times with potential, not attested by linguists, traces of African languages’ phonetic or morphologic influences. Finally, Category U included the 10 English utterances occurring 10 times and the 21 unique Kriolu utterances occurring 1089 times of unknown origin as they could not be traced to African or European languages.

Following (Verdu et al., 2017), we defined an ‘African-utterances score’ based, conservatively, on the utterance frequencies obtained separately for items in Category A, Category B, and merging Categories A and B, as, for individual i and the set of utterances in each category denoted Cat (in [A; B; A&B]), $Z_{i,Cat}=\sum _{l=1}^{L_{Cat}}{f}_{i,l}$, with L_Cat the number of uttered items in the corresponding category, and f_i,l defined as previously.

We calculated pairwise Allele Sharing Dissimilarities (Bowcock et al., 1994) using the asd software (v1.1.0a; https://github.com/szpiech/asd; Szpiech, 2020), among the 5157 individuals in our worldwide dataset (Figure 1; Figure 1—source data 1), using 455,705 autosomal SNPs, considering, for a given pair of individuals, only the SNPs with no missing data. We then projected this dissimilarity matrix in three dimensions with metric Multidimensional Scaling using the cmdscale function in R (R Development Core Team, 2020). We conducted successive MDS analyses on different individual subsets of the original ASD matrix, by excluding, in turn, groups of individuals and recomputing each MDS separately (Appendix 2—figures 1–4; Figure 2—figure supplement 1). Lists of populations included in each analysis can be found in Figure 1—source data 1. 3D MDS animated plots in gif format for Figure 2, Figure 2—figure supplement 1, and Appendix 2—figures 1–4 were obtained with the plot3d and movie3d functions from the R packages rgl and magick.

Recently admixed individuals are intuitively expected to be at intermediate distances between the clusters formed by their putative proxy source populations on ASD-MDS two-dimensional plots. A putative estimate of individual admixture rates can then be obtained by projecting admixed individuals orthogonally on the line joining the respective centroids of each proxy-source populations and, then, calculating the distance between the projected points and either centroid, scaled by the distance between the two centroids (Appendix 1—figure 2; Fortes‐Lima et al., 2021; Paschou et al., 2007). We estimated such putative individual admixture rates in Cabo Verdean, ASW, and ACB individuals, considering sets of individuals for the putative proxy-source populations identified visually and resulting from our ASD-MDS decomposition (Appendix 2).

Based on ASD-MDS explorations, we focused on the genetic structure of individuals born in Cabo Verde compared to that of other admixed populations related to TAST migrations. Therefore, we conducted ADMIXTURE analyses (Alexander et al., 2009) using 1,100 individuals from 22 populations: four from Europe, 14 from Africa, the USA-CEU, the African-American ASW, the Barbadian-ACB populations, and the North Chinese-CHB population as an outgroup (Figure 1—source data 1). To limit clustering biases due to uneven sampling, we randomly resampled without replacement 50 individuals, once, for each one of these 22 populations. Furthermore, we also included all 44 Angolan individuals from four populations in the analyses, as no other samples from the region were available in our dataset. In addition to the 1100 individuals hence obtained, we included the 225 Cabo Verde-born individuals.

Following constructor recommendations, we pruned the initial 455,705 SNPs set for low LD using plink (Purcell et al., 2007) function --indep-pairwise for a 50 SNP-window moving in increments of 10 SNPs and a r² threshold of 0.1. We thus conducted all subsequent ADMIXTURE analyses considering 1,369 individuals genotyped at 102,543 independent autosomal SNPs.

We performed 30 independent runs of ADMIXTURE for values of K ranging from 2 to 15. For each value of K separately, we identified groups of runs providing highly similar results (ADMIXTURE ‘modes’), with Symmetric Similarity Coefficient strictly above 99.9% for all pairs of runs within a mode, using CLUMPP (Jakobsson and Rosenberg, 2007). We plotted each modal result comprising two ADMIXTURE runs or more, for each value of K separately, using DISTRUCT (Rosenberg, 2003). We evaluated within-population variance of individual membership proportions as Fst/Fst^max values using FSTruct (Morrison et al., 2021) with 1000 Bootstrap replicates per population, for the ADMIXTURE mode result at K=2 (Figure 3—figure supplement 1).

Finally, we computed, using ADMIXTOOLS (Patterson et al., 2012; Maier et al., 2022), f₃-admixture tests considering as admixture targets each Cabo Verdean birth-island, the ASW, and the ACB separately, with, as admixture sources, in turn all 108 possible pairs of one continental European population (Source 1) and one continental African population (Source 2), or the East Asian CHB (Source 1) and one continental African population (Source 2). For all tests, we used the same individuals, population groupings, and genotyping dataset as in the previous ADMIXTURE analyses (Figure 3—figure supplement 2), and considered the no-inbreeding option in ADMIXTOOLS.

We first phased individual genotypes using SHAPEIT2 (Delaneau et al., 2012) for the 22 autosomal chromosomes separately using the joint Hap Map Phase 3 Build GRCh38 genetic recombination map (The International HapMap 3 Consortium, 2010). We considered default parameters using phasing windows of 2 Mb and 100 states per SNP. We considered by default 7 burn-in iterations, 8 pruning iterations, 20 main iterations, and missing SNPs were imputed as monomorphic. Finally, we considered a ‘Ne’ parameter of 30,000, and all individuals were considered unrelated.

We determined the possible origins of each Cabo Verdean, ASW, and ACB individual pair of phased haplotypes among European, African, USA-CEU, and Chinese-CHB populations using CHROMOPAINTER v2 (Lawson et al., 2012) with the same recombination map used for phasing. Following authors’ recommendations, we conducted a first set of 10 replicated analyses on a random subset of 10% of the individuals for chromosomes 2, 5, 12, and 19, which provided a posteriori Ne = 233.933 and θ=0.0004755376, on average by chromosome weighted by chromosome sizes, to be used in the subsequent analysis. We then conducted a full CHROMOPAINTER analysis using these parameters to paint all individuals in our dataset, in turn set as Donor and Recipient, except for Cabo Verde, ACB, and ASW individuals set only as Recipient. Finally, we combined painted chromosomes for each individual in the Cabo Verdean, ASW, and ACB population, separately.

We used CHROMOPAINTER results aggregated for each Cabo Verdean, ACB, and ASW individual separately and conducted two SOURCEFIND (Chacón-Duque et al., 2018) analyses using all other populations in the dataset as a possible source, separately for four or six possible source populations (‘surrogates’), to allow a priori for symmetric or asymmetric numbers of African and European source populations for each target admixed population. We considered 400,000 MCMC iteration steps (including 100,000 burn-in) and kept only one MCMC step every 10,000 steps for final likelihood estimation. Each individual genome was divided in 100 slots with possibly different ancestry, with an expected number of surrogates equal to 0.5 times the number of surrogates, for each SOURCEFIND analysis. We aggregated results obtained for all individuals in the ACB, ASW, and each Cabo Verdean birth-island, separately. We present the highest likelihood results across 20 separate iterations of the SOURCEFIND analysis in Figure 3B. The second-best results were highly consistent and thus not shown.

Considering the same sample and SNP set as in the above local-ancestry analyses (Figure 1—source data 1), we called ROH with GARLIC (Szpiech et al., 2017). For each population separately, we ran GARLIC using the weighted logarithm of the odds (wLOD) score-computation scheme, with a genotyping-error rate of 10^–3 (a likely overestimate), and using the same recombination map as for phasing, window sizes ranging from 30 to 90 SNPs in increments of 10 SNPs, 100 resampling to estimate allele frequencies, and all other GARLIC parameters set to default values. We only considered results obtained with a window size of 40 SNPs, which was the largest window size associated with a bimodal wLOD score distribution and a wLOD score cutoff between the two modes, for all populations.

For each population and Cabo Verdean island separately, we considered three classes of ROH that correspond to the approximate time to the most recent common ancestor of the IBD haplotypes, which can be estimated from the equation g=100/2l, where l is the ROH length in cM and g is the number of generations to the most recent common ancestor of the haplotypes (Thompson, 2013). Short ROH are less than 0.25 cM, reflecting homozygosity of haplotypes from more than 200 generations ago; medium ROH are between 0.25 cM and 1 cM reflecting demographic events having occurred between approximately 200 and 50 generations ago; and long ROH are longer than 1 cM, reflecting haplotypes with a recent common ancestor less than 50 generations ago. In Figure 4A–B and Appendix 4—figure 1, we plot the distribution of the summed length of ROH per individual per size-classes.

Using the same phasing results as described above, we conducted 10 EM iterations of the RFMIX2 (Maples et al., 2013) algorithm to assign, for each Cabo Verdean individual and each SNP on each chromosome, separately, its putative source population of origin among the 24 African, European, Chinese-CHB, and USA-CEU populations. We collapsed the local ancestry assignments for each SNP in each Cabo Verdean individual hence obtained into three continental regions, representing broadly African, broadly European, and broadly East Asian ancestries respectively. Any ancestry call that was assigned a population from the African continent was assigned a category of AFR, any ancestry call that was assigned a population from the European continent was assigned a category of EUR, and any ancestry call that was assigned a population from the East Asian continent was assigned a category of ASN. We considered an approach similar to previous work (Szpiech et al., 2019), and intersected local ancestry calls with ROH calls (Figure 4C–D).

RFMIX2 only makes local ancestry calls at loci that are present in the dataset. Therefore, a gap of unclassified ancestry exists where inferred ancestry switches between two successive genotyped loci as called by RFMIX2. These gaps necessarily each contain an odd number of ancestry switch points (≥1) absent from our marker set. Therefore, when computing the total ancestry content of an ROH that overlaps one of these ancestry breakpoints, we assign half of the length of this gap to each ancestry classification, effectively extending each local ancestry segment to meet at the midpoint.

We then plotted the length distribution of long ROH for each island and we break out the distributions by ancestral haplotype background: those with only African ancestry, those with only European ancestry, and those that span at least one ancestry breakpoint (Figure 4C). We excluded long ROH in East Asian ancestry segments from this and the following analyses, as we found such ancestry overall very limited in the samples (Appendix 4—figure 2). We also excluded ROH called with heterozygous ancestry calls (e.g. called with one haplotype called as AFR and the other as EUR). These regions were also rare (Figure 4—source data 3).

Finally, we explored how total African/European ancestry in long ROH varies between islands. For each individual, we summed the total amount of each ancestry in long-ROH and plot the distributions across islands (Figure 4—figure supplement 1). High levels of a given ancestry in long ROH could stem from an overall high level of that ancestry in that individual. Therefore, for each individual, we computed their global ancestry proportions by summing up the total length of the genome inferred as a given ancestry and dividing by the length of the genome. We then plotted (Figure 4D), the difference of an individual’s long-ROH ancestry proportion and their global ancestry proportion, for African and European ancestries separately. Values above zero indicated that a given ancestry is overrepresented in long-ROH relative to genome-wide proportions of that ancestry.

To assess the significance of these deviations, we performed a non-parametric permutation test. For each individual in each island, we randomly permuted the location of all long ROH (ensuring that no permuted ROH overlap), re-computed the local AFR ancestry proportion falling within these permuted ROH, and then subtracted the global ancestry proportion. We then took the mean of this difference across all individuals for each island and repeated the process 10,000 times. As there is negligible ASN ancestry across these individuals, the AFR and EUR proportions essentially add to 1, and therefore we consider an over/under representation of AFR ancestry in long ROH to be equivalent to an under/over representation of EUR ancestry in long ROH. Permutation distributions with observed values are plotted in Figure 4—figure supplement 2 and permutation p-values are given in Figure 4—source data 2.

ABC inference relies on simulations conducted with scenario-parameter values drawn randomly within prior distributions set explicitly by the user. We used MetHis v1.0 (Fortes‐Lima et al., 2021) to simulate 60,000 independent autosomal SNP markers in the admixed population H, forward-in-time and individual centered, under the four competing scenarios presented in Figure 6 and Table 2 and explicated below. In all four scenarios (Figure 6; Table 2), we considered, forward-in-time, that the admixed population (Population H) is founded at generation 0, 21 generations before present. Generation 0 thus roughly corresponds to the 15th century when considering an average generation-time of 25 years and sampled individuals born on average between the 1960s and the 1980s and no later than 1990 in our dataset. This is reasonable as historical records showed that Cabo Verde was likely un-inhabited before its initial colonial settlement established in the 1460s on Santiago (Albuquerque and Santos, 1991). Due to the recent admixture history of Cabo Verde and as we considered only independent genotyped SNPs, we neglected mutation in our simulations for simplicity.

Following our descriptive analyses results, we considered scenarios with only one ‘European’ and one ‘African’ source population, and each Cabo Verde island, separately, as the ‘Admixed’ recipient population H. This corresponds to the ‘two-source’ version of the general admixture model from Verdu and Rosenberg, 2011, also explored with MetHis previously (Fortes‐Lima et al., 2021).We therefore considered a single African and European population at the source of all admixture in Cabo Verde, and further considered that both source populations were very large and at the drift-mutation equilibrium during the 21 generations of the admixture process until present. Furthermore, we considered that these source populations were accurately represented, respectively, by the Mandinka from Senegambia and the Iberian-IBS populations in a random genotyping dataset of 60,000 independent autosomal SNPs (see Results).

In brief (see below), at each generation, MetHis performs simple Wright-Fisher (Fisher, 1922; Wright, 1931) forward-in-time discrete simulations, individual-centered, in a randomly mating (without selfing) admixed population of N_g diploid individuals. Separately for each N_g individual in the admixed population at generation g, MetHis draws parents randomly from the source populations and the admixed population itself at the previous generation according to given parameter-values drawn from prior distributions separately for each simulation.

We computed the mean and variance of interindividual ASD (Bowcock et al., 1994) within Population H. Furthermore, we computed the mean and variance of average heterozygosities (where the average is taken over independent SNPs for each individual and then over individuals within an island) as in Nei, 1978. We also calculated the mean and variance of inbreeding coefficient F (Danecek et al., 2011). Intuitively, we a priori expected these statistics to be particularly sensitive to genetic-drift and possibly informative specifically about reproductive population size N_e parameters in our scenarios.

We analytically showed previously that the distribution of admixture fractions across individuals within admixed populations carried identifiable information about the history of admixture (Verdu and Rosenberg, 2011; Buzbas and Verdu, 2018). Furthermore, we showed with MetHis (Fortes‐Lima et al., 2021) that summary statistics describing this distribution could be successfully used in ABC inferences.

Therefore, we considered the 16 admixture-related summary statistics describing the mode, the first four moments (as mean, variance, skewness, and kurtosis), and the minimum, maximum, and all deciles of the ASD-MDS admixture estimates of individual admixture fractions from the African source population (or one minus that from the European source in a two-source admixture scenario). ASD-MDS admixture estimates were obtained as described in Material and Methods, represented schematically in Appendix 1—figure 2, considering the 2D-MDS centroids, respectively, of the African and European sources, and the projection on the line joining these two points of each Population H individual. Furthermore, to further capture the ASD-MDS admixture patterns, we calculated the distribution of the angles between population H’s individuals and the source populations’ centroids on the 2D ASD-MDS. We then considered as summary-statistics for ABC inference the mode, the first four moments (as mean, variance, skewness, and kurtosis), the minimum and maximum, and the deciles of this distribution of angles in radian (Appendix 1—figure 2).

Using MetHis summary-statistic calculation tools, we computed Fst values between either Source population (African or European) and Population H following Weir and Cockerham, 1984. We also computed the mean ASD between the Source Population and Population H. Finally, we computed the f₃ statistic (Patterson et al., 2012) with African and European populations as the two sources and Population H as the targeted admixed population.

Appendix 1—figure 2

Download asset Open asset

We followed the same empirical approach as in MetHis-ABC (Fortes‐Lima et al., 2021) to determine the most suitable number of neurons in the hidden layer and associated number of simulations closest to the target data and retained for training the Neural Network (called “tolerance level”) for further NN-ABC posterior parameter estimation for this target. Indeed, there is no rule-of-thumb to determine these parameters a priori in order to obtain the most informative posterior parameter inference while minimizing overfitting using a Neural-Network approach under a given scenario (Csilléry et al., 2012; Jay et al., 2019).

Therefore, we empirically tested three different tolerance-level values to be used in NN-ABC posterior parameter estimation: 1%, 5%, or 10% of the total 100,000 simulations generated under the winning scenario determined by our RF-ABC procedure for each Cabo Verdean island and for Cabo Verde as a whole, respectively. Furthermore, for each one of the three tolerance-level values, we tested associated numbers of neurons in the hidden layer between 5 and 11, or 12, the number of free parameters minus one in the winning scenario for each Cabo Verdean island and for Cabo Verde as a whole, respectively (Figures 6–7).

Then, for each pair of tolerance-level value and number of neurons for each targeted data separately, we performed 1,000 cross-validation NN-ABC posterior parameter inferences using the cv4abc function in the abc package (Csilléry et al., 2012). This procedure draws 1,000 simulations at random and considers them, in turn, as pseudo-observed data for posterior-parameter inference using the remaining 99,999 simulations as the reference table. For the 1,000 pseudo-observed cross-validations and each pair of tolerance level and number of hidden neurons, we calculated the error between the median posterior point-estimate of each parameter (${\displaystyle {\hat{\theta }}_i}$) and its’ known true-value used for simulation (${\displaystyle {\theta }_i}$) , as the mean-square error (MSE) scaled by the parameter’s variance across the 1,000 cross-validations (Csilléry et al., 2012):

Therefore, we conducted, in total, 219,000 NN-ABC cross-validation parameter inference procedures to identify the NN-ABC settings most suitable for further posterior parameter estimations using the observed data for each Cabo Verdean birth-island and for Cabo Verde as a whole. For each Cabo Verdean island and for Cabo Verde as a whole, separately under the corresponding winning scenario, we chose the pair of tolerance-level values and number of hidden neurons that minimized the average error across all estimated parameters, as the Neural-Network setting used in further ABC posterior parameter inferences based on the observed data (Appendix 1—table 1).

For each Cabo Verdean island and for all Cabo Verde-born individuals grouped in a single population, separately, under the winning scenario identified with RF-ABC, we jointly estimated the posterior distributions of all parameters using NN-ABC “neuralnet” method option in the function abc of the R package abc (Csilléry et al., 2012), with logit-transformed parameters (“logit” transformation option) using the tolerance level and number of neurons in the hidden layer identified previously in Appendix 1—table 1. For each birth-island and for Cabo Verde as a whole, separately, a synthetic schematic figure of the complex admixture processes identified using median point estimates of the posterior scenario-parameter distribution is provided in Figure 7, and full posterior parameter distributions with 95% Credibility-Intervals (CI) are provided in Figure 7—figure supplements 1–3 and Appendix 5—Tables 1–10.

We evaluated posterior parameter error rates and 95% CI accuracies in the vicinity of the observed data, for each island and for Cabo Verde as a whole, separately, and for each corresponding NN-ABC posterior parameter joint estimation.

First, we identified the 1,000 simulations closest to the observed “real” data for each target data separately using a 1% tolerance level in the “neuralnet” option from the abc function. We then used each one of these 1,000 specific simulations, in turn, as pseudo-observed target data for cross-validation NN-ABC posterior parameter estimation using the same NN settings as previously, logit-transformed parameters, and the 99,999 remaining simulations in the reference table. We then compared the median posterior estimate of each scenario parameter ${\displaystyle \left({\hat{\theta }}_i\right)}$ , with the original parameter value used for the simulation $\left({\theta }_i\right)$ , respectively for each 1,000 cross-validation posterior parameter estimation; and calculated the cross-validation mean absolute error (MAE), for each scenario parameter (Appendix 5—Tables 1–10):

Second, for each island and for Cabo Verde as a whole, and for each scenario-parameter separately, we calculated how many times the true parameter values $\left({\theta }_i\right)$ , used for simulating the 1,000 simulations closest to the observed data, fell within the 95% CI [2.5% quantile ${\displaystyle \left({\hat{\theta }}_i\right)}$ ; 97.5% quantile ${\displaystyle \left({\hat{\theta }}_i\right)}$ ] estimated using the observed data. Thus, if the 1,000 cross-validation true parameter values fell more than 95% of the time within the estimated 95% CI, the length of the estimated 95% CI was considered over-estimated and thus excessively conservative; and, alternatively, if it was the case less than 95% of the time, the length of the 95% CI was considered under-estimated thus indicating a less conservative behavior of our CI estimation procedure (Appendix 5—Tables 1–10).

For individuals born on Santiago island, we find that the most recent pulses of admixture from Africa and Europe, respectively, occurred after the official abolition of the TAST in the Portuguese Empire in the 1830s, and after the effective abolition of slavery in the Portuguese empire in the 1860s (Figure 7). Both admixture pulses are characterized by posterior distributions with narrow 95%CIs clearly departing from their respective priors, both for the timing of each pulse and their respective intensities, in particular for the European admixture pulse (Figure 7—figure supplements 1–3, Appendix 5—table 1). For these recent events, observed genetic patterns are consistent with a relatively intense African admixture event (median = 0.5913, mode = 0.6837, 95%CI=[0.0642–0.9038]), occurring between generation sixteen and seventeen after the founding of Cabo Verde at generation 0 (generation median = 16.6, mode = 17.2, 95%CI=[10.0–18.9]). Furthermore, we find strong evidence for the most recent European admixture pulse being weak (median = 0.0684, mode = 0.0258, 95%CI=[0.0028–0.6021]), and having occurred between one and two generations before the sampled generation (generation median = 18.7, mode = 19.1, 95%CI=[11.7–19.6]).

NN-ABC posterior parameter estimations perform less accurately to infer the dates of the earlier admixture pulses and their associated intensities, both from Africa and Europe respectively, as shown by posterior distribution departing less clearly from their respective priors and overall large 95% CI. Nevertheless, results indicate that both the African and European pulses may have occurred in a more remote past; the African pulse was inferred at generation 9.2 (median) after founding at generation 0 (mode = 13.9, 95%CI=[1.7–16.8]); the European pulse at generation 7.5 (median, mode = 3.4, 95%CI=[2.3–15.9]), towards the beginning of the plantation economy era in the Americas and during the intensification of the TAST in the second half of the 17th century (Eltis and Richardson, 2015; Fortes-Lima and Verdu, 2021; Figure 7D). Conversely, the initial founding admixture pulse, set to occur at generation 0 at the beginning of Cabo Verde settlement history in the 1460s, is more accurately estimated a posteriori, as indicated by substantial departure from the prior distribution and relatively narrow CI. Results here indicate a stronger contribution from Africa than Europe (median = 0.6365, mode = 0.8816, 95%CI=[0.0538–0.9858]), at the root of the genetic make-up of Santiago-born individuals in our sample set.

Finally, effective population sizes in Santiago indicate a strong, relatively linear (u_Ne median = 0.365, mode = 0.440, 95%CI=[0.150–0.494]), increase since the 1460s up to 80,077 effective individuals (median, mode = 92,109, 95%CI=[30,408–98,975]), in the sampled generation born on average between the 1960s and the 1980s.

Altogether, our results indicate that admixture in Santiago occurred mainly between Senegambian and South Western European populations (Figures 2–3), at the very beginning of the Portuguese colonization of the archipelago in the 15th century. This early admixture history is likely due to an initial peopling of Cabo Verde strongly biased towards solitary males (familial permanent migrations from Portugal being a minority), who reproduced with African enslaved women, and whose admixed offspring constituted the first Cabo Verdean generations, as shown by some recorded instances of legitimizations of the admixed offspring to allow them to legally inherit dating back to the early 16th century (Carreira, 2000).

Our results further indicate that two substantial admixture events from Africa and Europe, respectively, occur concomitantly to the early expansion of the TAST during the 17th-century, although we are unable to precisely date both events. Nevertheless, our results support a more tenuous genetic admixture history from Africa experienced by Cabo Verdeans from Santiago during the most intense period of the TAST (18th-century) and the height of plantation economy in European colonial empires (Eltis and Richardson, 2015; Fortes-Lima and Verdu, 2021). This apparent reduced influence of the most intense period of the TAST on admixture pattern in Santiago is also echoed in our ASD-MDS, ADMIXTURE, and SOURCEFIND results, where Santiago Cabo Verdeans today share ancestry with Senegambian Mandinka almost exclusively (Figures 2–3). However, historical records unquestionably demonstrated that numerous other populations from West Central and South Western Africa were enslaved and forcibly deported to Cabo Verde during this era (Carreira, 2000). Thus, our results highlight that the strong and recurrent demographic migrations of enslaved Africans brought to Santiago during the second half of the TAST do not seem to have left a similar genetic admixture signature. This is consistent with several, mutually non-exclusive, historical scenarios, where enslaved African populations brought to Santiago may have been mostly re-deported to the Americas and Europe after the 17th century, without substantially contributing to the genetic landscape of this island, whether due to rapid re-deportation, and/or to slave-owners and colonial society strongly preventing marriages and reproduction between newly arrived enslaved individuals and pre-existing enslaved or non-enslaved communities in Santiago (see Discussion in main text).

Conversely, our results indicate an intense introgression event from Africa in Santiago, occurring shortly after the TAST and the abolition of slavery in the Portuguese empire in the 1860s. This may have been caused by a relaxation of the socio-economic constraints that predominated during most of the TAST against marriages between enslaved and non-enslaved communities as well as the control of marriages among enslaved individuals imposed by slave-owners; concomitant with the strong intensification of the illegal slave-trade which deported numerous enslaved Africans in a short amount of time shortly after the end of the TAST (Carreira, 2000; Albuquerque and Santos, 1991; Albuquerque and Santos, 1995; Albuquerque and Santos, 2007).

Our results also clearly identify a European introgression event in Santiago having occurred at the beginning of the 20th century, which may be explained by the European migrations towards African colonies triggered by European empires’ policies during the first half of this century. Interestingly, we find that this recent admixture event was of very limited intensity, thus indicating that European emigration to Santiago during the 20th century only had a quantitatively limited, albeit precisely detectable, influence on genetic admixture patterns.

Finally, Santiago-born individuals also show the lowest ROH levels and generally some of the shortest ROH in Cabo Verde (Figure 4 and Appendix 4—figures 1–2). When examining ancestry patterns within ROH in Santiago-born individuals, although proportions within ROH are on average consistent with an individuals’ overall genetic ancestry, there are some outlier individuals with much higher European ancestry within ROH compared to the remainder of the genome. This could possibly reflect historical marriage prohibitions and other social constraints against intercommunity reproductions (see Discussion in main text). These ROH results combined with the evidence for the relatively constantly increasing reproductive population in Santiago over the course of the peopling of Cabo Verde (Figure 7C), was consistent with this island being the administrative historical capital, and principal entry to the archipelago throughout its’ history (Albuquerque and Santos, 1991; Albuquerque and Santos, 1995).

For individuals born on Fogo island, we find that both European admixture pulses were precisely inferred with NN-ABC, for the time of their occurrence and for their respective intensities, as posterior parameter distributions and 95% CI substantially depart from their respective priors (Figure 7, Figure 7—figure supplements 1–3, Appendix 5—table 2). The most recent European admixture pulse in Fogo occurred concomitantly to the abolition of the TAST and of slavery in the Portuguese empire in the 1860’s (generation median = 15.5 after the founding of Cabo Verde at generation 0, mode = 16.7, 95%CI=[6.7–18.2]), and was of substantial intensity (median = 0.2428, mode = 0.2101, 95%CI=[0.0157–0.8485]). The older European pulse occurred much more remotely in the past (generation median = 5.3, mode = 2.5, 95%CI=[1.9–13.9]), shortly after the first perennial census above 100 individuals in 1570 for this island, and long before the increase in TAST intensity due to the expansion of Plantation Economy in the Americas in the second half of the 17^th century. Furthermore, note that our results indicate that this European pulse is of relatively similar intensity compared to the most recent one (median = 0.2234, mode = 0.1149, 95%CI=[0.0223–0.8383]).

Interestingly, we find that the African admixture history of individuals from Fogo occurred at very different periods of time than the European ones, the timing for both African admixture pulses being also well estimated with substantial departure from the prior and relatively narrow 95% CI. Indeed, we find strong indications of a very recent pulse of African admixture of overall limited intensity (median = 0.2034, mode = 0.1174, 95%CI=[0.0120–0.8225]), having occurred around the first part of the 20^th century (generation median = 19.3 after the founding of Cabo Verde at generation 0, mode = 19.5, 95%CI=[15.6–19.8]). Furthermore, we find that the previous African admixture pulse likely occurred between six and eight generations before the sampled one (generation median = 12.0 after Cabo Verde founding, mode = 13.3, 95%CI=[4.7–16.9]), in the midst of the most intense period of the TAST and plantation economy in the Americas, around the 1760s. However, we cannot recover a precise estimation for the intensity of this older pulse, as its’ posterior parameter distribution seldom depart from its’ prior. We also find that the founding admixture event set to occur at the beginning of the colonization history of the archipelago in the 1460’s is substantially departing from its’ prior distribution and very similar to the founding admixture event found for individuals born on Santiago (median = 0.6831, mode = 0.7407, 95%CI=[0.1679–0.9600]).

Finally, our posterior estimation of the demographic history of Fogo-born individuals substantially differs from that of Santiago. Indeed, while we find similar effective sizes towards the present between the two islands (median = 72,023, mode = 90,370, 95%CI=[18,011–98,846]), we find that the demographic increase was much more recent and rapid in Fogo (u_Ne median = 0.150, mode = 0.115, 95%CI=[0.055–0.447]) than in Santiago.

Our results indicate a very similar founding admixture event for Fogo island compared to that of Santiago, strongly suggesting that the admixture event at the root of Fogo peopling occurred early in Cabo Verde history, likely in Santiago island. This scenario is consistent with historical data suggesting an initial peopling of Fogo from Santiago at the end of the 15th century and the beginning of the 16th century, rather than independent founding events between islands (Albuquerque and Santos, 1991; Albuquerque and Santos, 1995).

Interestingly, we find a European pulse of admixture concomitant with the perennial peopling of the island at the end of the 16th century and the establishment of Fogo as an important center of agricultural and free-range cattle herding in the economic triangle between Europe, Cabo Verde, and continental Africa. Later on, we identify traces of an African admixture pulse having occurred during the most intense period of the TAST in the mid-18th century, thus consistent with increased African enslaved individuals’ forced deportations to Fogo as showed by historical records, but we could not infer its intensity, suggesting the difficult identifiability of such admixture event in the genetic landscape of Fogo island today with the methods here deployed.

This history of admixture is likely related to the intense control on the island’s economic expansion and regulated immigration imposed after 1532 by the central political, administrative, and commercial power of Santiago, with the notable exception of direct commercial relationships allowed between Fogo and Europe under the rule of the Portuguese crown (Albuquerque and Santos, 1991; Albuquerque and Santos, 1995; Albuquerque and Santos, 2007). This could thus explain the limited African admixture during Fogo history and more intense European admixture on this island, identified with our ASD-MDS/ADMIXTURE, and SOURCEFIND results (Figures 2–3). Furthermore, these socio-political factors likely also explain the maintaining of a relatively low reproductive population size on Fogo until very recently, as opposed to the more continuous population increase on Santiago (Figure 7C). In addition, this demographic history may be further explained by the recurrent eruptions of the Fogo volcano, which unquestionably chronically disrupted the island’s development, although historical records showed limited direct mortality due to these events and that communities fleeing the island often migrated back after each cataclysm (Albuquerque and Santos, 1991).

We identify a much more recent pulse of European admixture, having occurred in the first half of the 19th century, shortly after the abolition of the TAST in the Portuguese empire, possibly due to the profound socio-cultural changes in marital relationships between enslaved and non-enslaved communities during the abolition of slavery in colonial empires (see Discussion in main text). Finally, we identify precisely a weak African admixture pulse having occurred in Fogo during the 20th century possibly consistent with more recent work-related migrations from Africa to Cabo Verde (Albuquerque and Santos, 2007).

Finally, ROH patterns in Fogo are similar to Santiago, in that ROH levels are relatively low and individual ROH are relatively short (Figure 4, Appendix 4—figures 1–2). This is consistent with Fogo’s history of economic control and influx of European individuals. Interestingly, when exploring ancestry proportions within ROH relative to the remainder of the genome, we see on average more European ancestry within ROH. This could possibly reflect the historical influx of Europeans onto the island and/or a legacy of preferential marriages between individuals of shared family origin.

For individuals born on Santo Antão island, we find that both European admixture pulses after the founding admixture event are substantially departing from their respective priors and have relatively narrow 95% CI, both for the timing of their occurrence and their respective intensities (Figure 7, Figure 7—figure supplements 1–3, Appendix 5—table 3). We find that Santo Antão-born individuals have experienced a recent European admixture pulse of weak intensity (median = 0.0847, mode = 0.0467, 95%CI=[0.0056–0.6382]), that occurred roughly three generations before the sampled generation, at the turn of the 20^th century (generation median = 17.2 after Cabo Verde founding at generation 0, mode = 18.1, 95%CI=[7.6–19.1]). Furthermore, we find a much more ancient European admixture pulse of strong intensity (median = 0.3592, mode = 0.1542, 95%CI=[0.0216–0.9522]), occurring concomitantly (generation median = 5.8, mode = 3.2, 95%CI=[2.6–15.6]) with the first perennial census in this island dated in 1580, and before the massive increase of TAST during the 17th century.

Interestingly, we find that the first African admixture pulse after founding at generation 0 occurred at generation 4.9 (median, mode = 2.4, 95%CI=[1.9–11.9]), almost synchronically to the oldest European admixture pulse and the first perennial settlement of the island recorded in 1580. Furthermore, this African admixture pulse was intense (median = 0.6255, mode = 0.7574; 95%CI=[0.0451–0.9749]), which summed close to 100% with the European admixture pulse having occurred at very similar times. This obliterated any older admixture events, consistently with the wide 95% CI obtained for the intensity of the founding admixture event for this island (median s_Afr,0=0.5681, mode = 0.8047, 95%CI=[0.0260–0.9839]). Importantly, our NN-ABC posterior parameter estimation fail to infer the second African admixture pulse identified by our RF-ABC scenario-choice. Indeed, the posterior distributions of both the timing of this most recent pulse and its’ intensity depart little from their respective priors.

Finally, despite the relatively poor performances of our inferences of demographic changes in Santo Antão, our results are indicative of a very reduced population in this island over the course of Cabo Verdean peopling history until a very recent increase up to 41,256 individuals (median, mode = 12,056, 95%CI=[3570–96,855]).

Our results for the admixture history of Santo Antão indicate that individuals born today on this island experienced an African-European admixture event concomitant with the first perennial settlement of the island in the 1580s, at the root of the genetic makeup of this island.

Most interestingly, we find no identifiable genetic admixture events, from Europe or Africa, throughout the height of the TAST on Santo Antão, albeit this island experienced a strong agricultural expansion during the plantation economy era which triggered recurrent African enslaved individuals’ deportations to this island, as shown by historical records (Albuquerque and Santos, 1991; Albuquerque and Santos, 1995; Albuquerque and Santos, 2007). Thus, even more strikingly than observed in Santiago, our results indicate that the enslaved-African forced deportations to Santo Antão during most of the TAST did not leave a statistically significant genetic contribution to the genetic landscape of the island, for likely the same complex socio-cultural reasons suggested for the rest of Cabo Verde (see Discussion in main text).

In this context, our ADMIXTURE and SOURCEFIND results identify two different Senegambian populations, the Mandinka and the Wolof, sharing significant African haplotypic ancestry with Santo Antão-born individuals, as well as a much more limited shared ancestry with the West Central Africa Igbo population (Figure 3B). Our ABC results would thus be consistent with a scenario where different enslaved-African populations from Senegambia and West Central Africa where brought together to settle Santo Antão in the early stages of the TAST in the late 16^th century, a pattern distinct to our findings for Santiago and Fogo. Alternatively, it is possible that the Igbo shared ancestry identified in our analyses stem from another admixture event that we failed to identify. This is likely due to an overall limited shared ancestry between this population and Santo Antão identified only with haplotypic local ancestry inferences, as our MetHis-ABC procedure considered a single Senegambian population at the root of the admixture history of the island and a much more limited number of independent SNPs compared to the haplotypic local ancestry analysis.

Finally, we find that a weak but significant European admixture pulse occurred at the turn between the 19th and the 20th centuries, substantially later than the end of the TAST and the abolition of slavery and before the 20th European colonial migrations, whose putative historical causes remain to be elucidated. Furthermore, we find that the reproductive population of Santo Antão remained overall small during its’ entire history until a very recent expansion, although much more limited than the expansions observed in Santiago and Fogo. This is somewhat surprising with the census records showing that Santo Antão was the second most peopled island after Santiago during most of Cabo Verde history (Figure 7—source data 1). Nevertheless, our results could be explained by the demographic collapse experienced by Santo Antão since the beginning of the 20th century, due to several intense famine episodes and rural-urban emigration from the island; a demographic history scenario thus much more complex than the relatively simple scenario here explored. Reflecting the historically small population size on the island, Santo Antão show high total ROH and generally quite long individual ROH among the islands (Figure 4 and Appendix 4—figures 1–2).

For individuals born on São Nicolau island, we find that the two most recent admixture pulses from Europe and Africa, respectively, are precisely estimated with posterior distributions substantially departing from their priors and reduced 95% CI, for the pulses’ timing and intensities (Figure 7, Figure 7—figure supplements 1–3, Appendix 5—table 4). We find that the most recent pulse of European admixture is weak in intensity (median = 0.0731, mode = 0.0297, 95%CI=[0.0038–0.6642]), and occurred very recently at the generation before that of the sampled individuals (generation median = 19.0 after the founding at generation 0, mode = 19.2, 95%CI=[10.3–19.8]). Furthermore, we find that the most recent African admixture pulse is very intense (median = 0.6398, mode = 0.7132, 95%CI=[0.0877–0.9380]), and occurred at generation 14.5 after founding (median, mode = 14.6, 95%CI=[8.3–16.4]), during the abolition of the TAST in the first half of the 19th century.

Interestingly, for both the more ancient admixture pulses from either Africa or Europe, our NN-ABC posterior parameter estimation provide reasonably informative posterior distributions pointing towards relatively weak African and European admixture intensities (median = 0.3133, mode = 0.1695, 95%CI=[0.0171–0.9262]; and median = 0.3240, mode = 0.1193, 95%CI=[0.0145–0.9460], respectively). However, our results only provide indications that both pulses were ancient and occurred synchronically (generation median = 5.2 after founding, mode = 2.4, 95%CI=[1.3–15.5]; generation median = 6.7, mode = 2.7, 95%CI=[2.0–16.1], for the African and European older pulses, respectively), as posterior distributions for the timing of these events are largely confounded with their respective priors.

Finally, we obtain a posterior distribution of the founding admixture intensity largely confounded with its prior, similarly as for the demographic parameters, thus showing a limit of our ABC inferences procedures.

Our results for the admixture history of São Nicolau indicate that two moderate admixture pulses from Africa and Europe occurred synchronically between the date for the first perennial settlement of this island in the 1580s and the onset of the intensification of the TAST at the beginning of the 17th century. Much later in the course of Cabo Verde peopling history at the beginning of the 19th century, we find an intense introgression event from Africa, probably occurring for the same reasons as in Santiago (see above) and other Cabo Verde islands (see below and Discussion in main text).

Interestingly, our haplotypic local ancestry inferences identify three African populations with shared ancestries with São Nicolau-born individuals (Figure 3); Senegambian Wolof and Mandinka shared-ancestry representing the majority similarly to the rest of Cabo Verde and as expected historically (see Discussion in the main text, Carreira, 2000; Albuquerque and Santos, 1991; Albuquerque and Santos, 1995; Albuquerque and Santos, 2007), and South West Central African Angolan Ovimbundu a small minority. Analogous to the Santo Antão inferences’ limitations, our analyses cannot disentangle whether either or both African pulses in São Nicolau involved the limited South West Central African populations shared ancestry here identified.

Finally, we identify, much later during the 20th century, a weak event of European introgression for the genetic admixture history of São Nicolau (Figure 7D), similarly likely due to labor-induced migrations in the former Lusophone empire as for other islands, and in particular possibly due to the expansion of fishing and cannery industry in São Nicolau at this time (Albuquerque and Santos, 2007).

Altogether, our genetic data contain little identifiable information for the demographic history of this island (Figure 7C, Figure 7—figure supplement 1). This may be due to the recurrent demographic collapses of the permanent settlement in this island due to intense starvation events, epidemic outbreaks, and destructive pirate raids that affected the peopling of this island throughout history (Figure 7—source data 1). Nevertheless, one genetic pattern that was consistent with these recurrent demographic collapses is the high total levels of ROH observed in the data (Figure 4 and Appendix 4—figures 1–2).

For individuals born on Brava island, we find that the most recent pulse of European admixture is relatively weak (median = 0.1697, mode = 0.1281, 95%CI=[0.0188–0.6504]), and occurred towards the abolition of slavery in the Portuguese empire in the second half of the 19th century (generation median = 16.3 after the founding of Cabo Verde at generation 0, mode = 17.9, 95%CI=[4.2–19.3]), as indicated by posterior distributions of parameters for this event largely differing from their priors (Figure 7, Figure 7—figure supplements 1–3, Appendix 5—table 5). We identify a similarly intense pulse of admixture from Africa (median = 0.2473, mode = 0.1261, 95%CI=[0.0191–0.8266]), for which our inferences only bore indications that it occurred during the second half of the 18th century during the TAST, as the 95% CI for the timing of this event is wide (generation median = 13.6 after founding at generation 0, mode = 15.7, 95%CI=[2.4–19.2]).

More remotely in the past, we identify strong indications that a very weak European admixture pulse (median = 0.0352, mode = 0.0157, 95%CI=[0.0007–0.7239]) may have occurred at the beginning of the 17th century (generation median = 7.3 after founding, mode = 3.6, 95%CI=[2.8–16.1]). Furthermore, we identify a strong signal for an intense African admixture pulse (median = 0.6688, mode = 0.7403, 95%CI=[0.1358–0.9685]), occurring at the beginning of Cabo Verdean peopling history, during the early 16th century, at generation 2.8 after the founding of the archipelago at generation 0 (median, mode = 1.6, 95%CI=[1.1–12.2]), and thus largely before the first perennial settlement of this island.

Finally, our results strongly support a very limited effective size linear increase (u_Ne median = 0.420, mode = 0.435, 95%CI=[0.243–0.494]) throughout the entire history of the island reaching 7337 effective individuals (median, mode = 3278, 95%CI=[647–73,114]) in the present.

We find an intense first African admixture pulse roughly three generations prior the first official and perennial peopling of the island in the 1580s (Figure 7—source data 1, Figure 7), thus strongly suggesting that Brava was initially peopled by individuals already admixed and originating from another Cabo Verdean island, similarly to what was suggested for Fogo (see above). We hypothesize that Fogo admixed individuals were originally at the root of this peopling, as the patterns of shared haplotypic ancestry in Brava-born individuals strongly resemble that of Fogo, rather than Santiago, with shared ancestry exclusively with Senegambian Mandinka and in similar proportion, as well as a significant (albeit very reduced) signal for a shared ancestry with Tuscan TSI otherwise only found in Fogo (Figure 3). We find a limited introgression event from Europe during the late 17th century, possibly consistent with a known immigration event from Madeira and the Açores, and from Fogo after a volcano eruption, around this time (Albuquerque and Santos, 1995). We also find a relatively weak African admixture pulse at some point during the height of the TAST for which the timing remains poorly inferred, and which will require further investigations.

Finally, we identify a more intense pulse of European admixture shortly after the abolition of slavery in the Portuguese empire in the second half of the 19th century. This later event may be due to the expansion of the whaling industry at this time, particularly in Brava due to favorable maritime currents and routes for this industry, which brought European and European-North American sailors to the island and increased locally contacts with Europe, and North America. Most interestingly, this historical scenario is also consistent with the signal, virtually unique across Cabo Verde, of a significant albeit reduced shared haplotypic ancestry between Brava-born individuals and the British GBR.

Notably, our results indicate that the effective population of Brava remained low during the entire peopling history of Cabo Verde and until today (Figure 7C), consistent with this island being the smallest inhabited and the most south-west ward in the archipelago (Figure 7—source data 1). Surprisingly, in spite of the historically low effective population size, Brava shows low total ROH levels and relatively short ROH lengths (Figure 4 and Appendix 4—figures 1–2), which remains to be elucidated in the future by considering larger sample sizes from the island.

For individuals born on Maio island, we find evidence that the most recent pulse of European admixture was of relatively modest intensity (median = 0.1472, mode = 0.1134, 95%CI=[0.0233–0.5957]), and likely occurred during the 18^th century, in the midst of the TAST (generation median = 12.7 after founding at generation 0, mode = 14.1, 95%CI=[3.7–17.9]), as both posterior distributions substantially departe from their respective priors (Figure 7, Figure 7—figure supplements 1–3, Appendix 5—table 6). Although our RF-ABC scenario-choice favors several pulses of admixture from both sources, our NN-ABC posterior parameter estimation fails to capture sufficient information to infer satisfactorily the most recent pulse of admixture from Africa, as both the posterior distribution of the timing of this event and its intensity seldom depart from their respective priors and have wide 95% CI.

However, we capture substantial information strongly suggesting two, much older, admixture pulses from Europe and Africa, respectively, occurring virtually synchronically in the middle 16^th century (generation median = 4.1 after founding for the European pulse, mode = 2.3, 95%CI=[1.7–14.2]; generation median = 4.2 after founding for the African pulse, mode = 2.4, 95%CI=[1.5–12.8]), thus much earlier than the first perennial settlement of this island dated in 1650 by historians (Figure 7—source data 1). Furthermore, we find that the respective intensities of the two pulses (median = 0.3260 for the European pulse, mode = 0.1075, 95%CI=[0.0117–0.9497]; median = 0.5589 for the African pulse, mode = 0.7703, 95%CI=[0.0346–0.9769]), sum close to 100%. This suggests a large replacement of the Maio population at that time, with indications that these synchronic admixture events occur on the basis of a founding admixed population largely of African origin (median s_Afr,0=0.6757, mode = 0.9135, 95%CI=[0.017–0.421]).

Finally, the demographic inference of reproductive size changes in Maio indicates, similarly as for Brava, that the reproductive size of the island remained relatively small during its entire history, and only very recently increased up to a median 21,454 effective individuals (mode = 7159), albeit with a large 95% CI ([914–93,656]).

Our results for the admixture history of Maio indicate a relatively simple scenario with a single pulse of European and African synchronic admixture having occurred in the mid-16th century in another Cabo Verdean island, long before the first perennial settlement of the island in 1650, similarly to Fogo or Brava. We hypothesize that this event occurred on Santiago, as shared local haplotypic ancestry patterns are highly resembling between islands (Figure 3), and also consistently with Maio being located very close to Santiago and historical records showing unequivocally the strong relationships between islands throughout history (Carreira, 2000; Albuquerque and Santos, 1991; Albuquerque and Santos, 1995; Albuquerque and Santos, 2007).

After the descendants of this initial admixture event settled Maio, they experienced only a single identifiable admixture pulse of weak intensity from Europe, which likely occurred at the turn between the 18th and the 19th century during the TAST, and no clearly identifiable further signal of African admixture in the island. As for some other Cabo Verde islands, we cannot conclude about the timing of the African admixture event that brought the reduced signal of West Central and South West Central Africa shared haplotypic ancestry with Maio, observed in our ADMIXTURE and SOURCEFIND analyses (Figure 3).

Finally, the reduced population size throughout the history of Maio is consistent with the island being located close to Santiago and with much less water resources, having made it difficult to maintain and expand the settlement on this island, despite its first peopling having occurred early in the history of Cabo Verde almost exclusively dedicated to free-range cattle herding (Carreira, 2000; Albuquerque and Santos, 1991; Albuquerque and Santos, 1995).

ROH patterns are also consistent with this history, with high total ROH indicating long term isolation and population bottlenecks. Similar to Santiago there were some outlier individuals with much higher European ancestry within ROH compared to the remainder of the genome (Figure 4 and Appendix 4—figures 1–2). This could possibly reflect historical marriage prohibitions and other social constraints against intercommunity reproductions (see Discussion in main text).

For the admixture history of individuals born on Boa Vista, we find an intense admixture pulse from Africa (median = 0.6553, mode = 0.7516, 95%CI=[0.1262–0.8949]), that occurred in the mid-19th century (generation median = 15.0 after Cabo Verde founding at generation 0, mode = 14.8, 95%CI=[11.0–17.4]), during the abolition of the TAST, with posterior distributions for both these parameters substantially departing from their respective priors and relatively reduced 95% CI (Figure 7, Figure 7—figure supplements 1–3, Appendix 5—table 7). Prior to that event, we fail to estimate the timing and intensity of the older African admixture pulse into Boa Vista, as posterior parameter distributions seldom depart from their respective priors for this event.

However, our results strongly support a scenario with an intense period of recurring monotonically decreasing admixture from Europe (median admixture intensity at the period’s start = 0.8057, mode = 0.9219, 95%CI=[0.2736–0.9907]; median admixture intensity at the period’s end = 0.340, mode = 0.120, 95%CI=[0.021–0.880]), between the first official settlement of the island in 1566 (generation median at the start of the European period of admixture = 3.9 after founding, mode = 2.2, 95%CI=[1.7–11.8]), until shortly after its first perennial peopling dated in 1650 (generation median at the end of the European period of admixture = 8.2, mode = 8.6, 95%CI=[3.1–14.9]). Interestingly, the initial founding admixture event at the root of the genetic patterns of Boa Vista-born individuals supports a substantially lower amount of African admixture (median = 0.3917, mode = 0.1706, 95%CI=[0.1814–0.6678]), compared to the rest of Cabo Verde where this parameter could be reliably identified.

Finally, our results show that the demography of Boa Vista remained relatively constant throughout the history of Cabo Verde, and indicate a possible very recent increase in effective size, albeit posterior parameter estimation of the most recent effective size was ambiguous due to large 95% CI (median Ne₂₀=47,744, mode = 13,128, 95%CI=[2657–98,587]).

Altogether, our results for the admixture history of Boa Vista born-individuals indicate a largely differing admixture history compared to all other islands of the archipelago. First, we find that the admixture event at the root of the genetic peopling of the island, although occurring long before the first perennial peopling of the island similarly to other islands such as Fogo, Brava, and Maio, is largely biased towards European admixture, as opposed to all other islands in the archipelago (Figure 7D). Second, we find the beginning of a period of recurring admixture from Europe before the mid-16th century, concomitantly with the first official settlement of the island, and until its first perennial settlement in the 1650s (Hakluyt, 2014)^pp.529. Interestingly, this corresponds to a period when Boa Vista has been used by Cabo Verde as a penitentiary island for non-enslaved individuals which may explain the recurring admixture process here identified (Carreira, 2000; Albuquerque and Santos, 1991; Albuquerque and Santos, 1995).

Later-on towards the end of the TAST in the 19th century, Boa Vista experienced an intense pulse of African admixture, similarly to Santiago and Saõ Nicolau and probably for similar complex socio-cultural reasons (see Discussion in the main text). However, overall limited historical records specific to this island render speculative the interpretation of our results based on genetics only (Figure 7—source data 1).

Finally, the effective size of Boa Vista remained constant and low throughout the history of this island (Figure 7C), which echoes its’ small historical census sizes, mainly due to limited water resources on the island, up until today (Figure 7—source data 1). ROH patterns are also consistent with this history, with the highest total ROH among the Cabo Verdean islands indicating long term isolation and population bottlenecks (Figures F4 and Appendix 4—figures 1–2).

For the admixture history of individuals born on São Vicente, we find a recent European admixture pulse of substantial intensity (median = 0.2840, mode = 0.2596, 95%CI=[0.0397–0.8044]), having likely occurred towards the end of the 19th century (generation median = 17.4 after founding of Cabo Verde at generation 0, mode = 17.7, 95%CI=[10.6–18.9]); both posterior distributions have relatively narrow 95% CI and clearly depart from their respective priors (Figure 7, Figure 7—figure supplements 1–3, Appendix 5—table 8).

This most recent admixture event is preceded by a period of recurrent decreasing African admixture likely spanning the second half of the most intense period of the TAST starting in the early 18th century (generation median at the start of the African period of admixture = 9.7, mode = 12.6, 95%CI=[2.5–16.9]), and ending during the end of the TAST and the abolition of slavery in the Portuguese Empire in the mid-19th century (generation median at the end of the African period of admixture = 15.8, mode = 17.5, 95%CI=[7.4–19.1]), albeit our results are ambiguous for the estimate of the start of this admixture period as indicated by the large 95% CI. Furthermore, our results are also ambiguous to estimate the intensity of the African introgression during the period of recurring admixture, as the posterior distributions for the onset and offset intensities, as well as the shape of the monotonic recurring admixture do not clearly depart from their respective priors and had large 95%CIs.

We find traces of an older European admixture pulse of substantial intensity (median = 0.3403, mode = 0.1772, 95%CI=[0.0149–0.9133]), which likely occurred at the beginning of the 17th century (generation median = 6.4 after founding, mode = 2.8, 95%CI=[1.9–15.1]). Our results indicate that this first admixture event occurred on a root-population that exhibited a large African admixture proportion (median s_Afr,0=0.5983, mode = 0.8596, 95%CI=[0.0529–0.9835]), albeit this result should be considered with cautious as CI is overall wide.

Finally, our results clearly indicate a strong effective population expansion since the perennial peopling of the island in the late 18th century, up to 66,715 effective individuals today (median, mode = 85,838, 95%CI=[12,159–98,375]).

Our results for the admixture history of individuals born on São Vicente indicate a strong European admixture event concomitant with the first official peopling of the island occurring at the same time as the nearby island of Santo Antão, consistently with historical records (Figure 7—source data 1). However, the perennial peopling of this island was difficult due to the very limited water resources, in particular compared to the much more hospitable, larger, and nearby Santo Antão. Thus, São Vicente essentially served as a coal and salt harbor depot during most of its history with a minimal settlement (Albuquerque and Santos, 2007), which may explain the initial European admixture pulse we identify (Figure 7), and our local haplotypic shared ancestry results (Figure 3).

Later-on and unique in Cabo Verde, our results support a long period of recurring African admixture occurring during the entire second half of the TAST and ending with the abolition of slavery in the Portuguese empire. However, only the final admixture event of this period could be precisely inferred to have occurred during the mid-19th^h century during the abolition of the TAST and of slavery in the Portuguese empire (Figure 7D). The small haplotypic ancestry shared between São Vicente-born individuals and West Central and/or South West Central African population may stem from illegal slave-trade known to have surged during this period. Nevertheless, the poor performances of our parameter inference for the other parameters of this recurring admixture period may indicate that more complex admixture scenarios than the ones considered occurred in this island, thus begging for future work further complexifying admixture histories to be reconstructed with ABC.

Interestingly, historical data show that São Vicente census only very recently started to increase, with newly available water sources and a protected deep-water harbor in the bay of Mindelo, unique in Cabo Verde, which favored the massive economic expansion of the island in the late 19th century and throughout the 20th century (Figure 7—source data 1, Duarte et al., 2015^p.36). Furthermore, the end of the agro-slavery system and end of plantation economy in Cabo Verde at the end of the 18th century and beginning of the 19th century are known historically to have triggered emigration from Santo Antão and rural-urban migrations during the 19th and 20th century, essential for São Vicente perennial peopling (Albuquerque and Santos, 2007; Patterson, 1988). This is further consistent with extensive familial grand-parental birth-places in Santo Antão and extensive familial relationships reported by São Vicente-born individuals in our familial anthropology interviews.

Altogether, these historical, economic, and demographic processes may well explain the onset of the large reproductive population increase identified in our analyses (Figure 7C), as well as haplotypic local ancestry patterns largely resembling those obtained for Santo Antão (Figure 3). Furthermore, they are consistent with the substantial European admixture pulse identified at the turn of the 19th and 20th century with our analyses. Reflecting the historical small population size, São Vicente had high total levels of ROH (Figure 4). Interestingly, individuals from this island, on average, show more African ancestry in ROH compared to the remainder of the genome. This could possibly reflect historical marriage prohibitions or the historically high levels of recurrent African admixture, or both.

Finally, the admixture history of individuals born on Sal inferred with NN-ABC is overall ambiguous a posteriori (Figure 7, Figure 7—figure supplements 1–3, Appendix 5—table 9). Indeed, while our RF-ABC scenario-choice leaned towards a complex admixture history with a period of recurring admixture from Europe and two African admixture pulses after the initial founding admixture event for the population of Sal, the duration and intensity of the European admixture period, as well as the first African admixture pulse, all showed posterior parameter distributions seldom departing from their priors. Nevertheless, we clearly identify a recent African admixture pulse of moderate intensity (median = 0.2401, mode = 0.1780, 95%CI=[0.0197–0.6590]), which likely occurred at the end of the 19^th century (generation median = 17.6 after the founding of Cabo Verde at generation 0, mode = 17.9, 95%CI=[12.5–19.0]). Furthermore, our results indicate a strong and very recent effective size expansion in Sal, up to 63,289 individuals (median, mode = 88,248, 95%CI=[5689–98,181]).

Sal was among the early Cabo Verdean islands to be officially settled and exploited, in 1529 (Figure 7—source data 1), as a source of sea-salt both exported and employed locally in the free-range herding meat-production and tannery industry at the historical root of Cabo Verde economy with Europe and Africa (Albuquerque and Santos, 1991; Patterson, 1988). However, its perennial peopling dated only from 1841, the last perennially peopled island of Cabo Verde, as water sources and agricultural surfaces are scarce on this island. This slow demographic expansion is well captured by our inference showing a reduced population size during most Cabo Verde history on Sal, followed very recently by a strong reproductive size increase (Figure 7C). Furthermore, in this context, our results clearly identifying only a recent admixture pulse of African origin at the beginning of the 20^th century could reflect the recent economic migrations of Western African populations to this island, during the development of its salt mining activities and its travel and tourism activities since then. Nevertheless, our scenario-choice results suggested the occurrence of a period of recurrent European admixture in Sal, which our posterior parameter inference largely failed to identify. This is possibly due to the complex recent peopling history of Sal that we fail to capture with genetic data, which will need to be clarified in the future in particular with additional samples from this island. Nevertheless, reflecting the historical small population size, Sal has high total levels of ROH (Figure 4).