The dispersal of domestic cats from North Africa to Europe around 2,000 years ago

Abstract

The domestic cat (Felis catus) descends from the African wildcat Felis lybica lybica. Its global distribution alongside humans testifies to its successful adaptation to anthropogenic environments. Uncertainty remains regarding whether domestic cats originated in the Levant, Egypt, or elsewhere in the natural range of African wildcats. The timing and circumstances of their dispersal into Europe are also unknown. In this study, the analysis of 87 ancient and modern cat genomes suggests that domestic cats did not spread to Europe with Neolithic farmers. Conversely, they were introduced to Europe around 2,000 years ago, probably from North Africa. In addition, a separate earlier introduction (1st millennium BCE) of wildcats from Northwest Africa may have been responsible for the present-day wild population in Sardinia.

The domestic cat (Felis catus) has a complex evolutionary history that ultimately resulted in a close association with humans (1). As one of the most successful mammalian domesticates, it has a worldwide distribution that extends even to remote islands. When including feral cats (i.e. populations descended from domestic cats currently living in the wild), the global population of Felis catus is approaching one billion (2, 3).

Despite their ubiquity, the timing and circumstances of cat domestication and dispersal remain uncertain. This is due to a variety of factors, including the paucity of felid remains in archaeological contexts, the difficulty of assigning species and domestication status to skeletal elements (since wild and domesticated forms overlap in size and morphology (4–6)), and the limited number of ancient and modern genomes analyzed so far. As a result, current hypotheses regarding when, where, and how cats were domesticated are poorly supported by empirical evidence.

Genetic findings from present-day wild and domestic cats have demonstrated that the African wildcat (Felis lybica lybica), currently distributed across North Africa and the Near East, is ancestral to all modern domestic cats (7). The burial of a cat in association with the complete skeleton of a human dated to ~7,500 BCE in Cyprus (8) led to the hypothesis that cats had been domesticated in the Levant during the Pre-Pottery Neolithic (9,600-7,000 BCE), where they may have acted as pest-controllers in early farming communities (9). A more traditional view, built upon iconographic and funerary evidence dating as far back as ~2,200 BCE, pointed to Pharaonic Egypt as the original locality of cat domestication (10, 11). Yet, attempts to tame cats might also have started earlier, in the Predynastic period ~3,700 BCE (12).

A clear understanding of the dispersal of domestic cats is hampered by the limited knowledge of the natural distribution ranges of African (F. l. lybica) and European (F. silvestris) wildcats. Past hybridization between these two species might have occurred naturally in areas where their ranges overlapped or were adjacent (13, 14). Ancient mitochondrial DNA (mtDNA) evidence suggests that wild populations of F. l. lybica were present in Anatolia and Southeast Europe ~10,000 years ago (13). The spatiotemporal distribution of maternal lineages in Europe, Southwest Asia, and North Africa led to the hypothesis that domestic cats carrying the haplogroup IV-A were initially dispersed by people from Anatolia via a southeastern route into central Europe as early as the mid-5th millennium BCE, possibly alongside a Neolithic expansion (13, 15). No later than ~2,000 years ago, during Classical Antiquity, domestic cats carrying the haplogroup IV-C reported from Egypt were dispersed across Europe and Southwest Asia (13, 15). However, recent genome-wide data from ancient and modern cats have questioned the use of mtDNA to draw conclusions about former natural distributions of wildcats and the spread of domestic populations (14). This is due to discordant nuclear and mtDNA evolutionary reconstructions observed in present-day European wildcats (7, 16) most likely resulting from potential past hybridization.

The reconstruction of cat dispersal is further complicated by the uncertain origins of wild F. l. lybica populations on the Mediterranean islands of Sardinia and Corsica (17). Though it is claimed that their origin can be traced to early domestic cats introduced by farmers from the Near East during the Neolithic (18–20), archaeological evidence suggests a much later introduction (21), and an analysis of morphological markers concluded that the modern populations shared an affinity with North African wildcats (22). In addition, morphological (23) and genetic analyses (20) of present-day specimens have indicated that they are distinct from domestic cats. Combined, these lines of evidence suggest that wildcats in Sardinia and Corsica are not descendants of a feral population of early domestic cats, but of a separate wildcat lineage. The poor documentation of small carnivores in the archaeological record of Sardinia and Corsica (21), however, and the lack of genetic data from North African and Near Eastern wildcats have made it difficult to test this hypothesis and to identify their source populations.

Here, to assess the timing and possible routes of dispersal of domestic cats into Europe, as well as their relationships with wildcats from Sardinia, we conducted paleogenomic analyses of 225 ancient cat specimens from 97 archaeological sites in Europe and Anatolia (Table S1, S2). We generated 70 low-coverage (~0.03-fold to ~3.82-fold) ancient genomes (Fig. 1A, Table S3) spanning a period of more than ten millennia from the 9th millennium BCE to the 19th century CE, and 17 low- to mid-range coverage (~0.7- to ~18-fold) genomes of present-day and museum specimens of wildcats from Italy (including Sardinia), Bulgaria, and North Africa (Table S4). To ensure a high degree of temporal resolution for the reconstructed genetic variation, we directly radiocarbon-dated 37 cat remains from 30 archaeological sites (Table S5).

Fig. 1. Samples and spatiotemporal distribution of cat ancestries.

(A) Geographic provenance of cats for which ancient and modern genomes were generated in this study. The approximate chronology of each sample is coded with shapes and gray scale colored as in the legend panel (chronology). Three samples from Israel previously analyzed (14) are also reported in the map. (B) Genome-wide autosomal SNPs maximum-likelihood (ML) tree of high- and mid-range coverage modern cats from this study and the literature. Support values are based on 500 bootstrap replicates; the asterisks indicate bootstrap values=100. Main clades are colored as in the legend panel (ancestry) (C) Principal component analysis built by projecting low coverage samples onto the coordinate space defined by modern mid- to high-range coverage wild and domestic cats. Present-day samples are color-coded based on the clades of figure 1B, while symbols and colors of ancient samples are chronologically coded as in figure 1A. (D) Spatiotemporal representation of cat ancestries where each dot represents a cat. Symbols are as in the geographic map (Fig. 1A). Colors are as in the ML tree (Fig. 1B) for the present-day wildcats, and in the ancient cats they represent genome-wide ancestries as inferred by the clusters of the PCA (Fig. 1C). (E) Bayesian tree of complete mtDNAs of ancient cats for which genome-wide data were obtained in this study, and present-day wild and domestic cats from this study and the literature. Symbols are as in the geographic map (Fig. 1A). Colors are as in the autosomal ML tree for the present-day samples (Fig. 1B) and in the ancient cats they represent genome-wide ancestries as inferred by the clusters of the PCA (Fig. 1C). Some clades were collapsed for clarity and for each symbol the total number is reported in brackets. Posterior probabilities between 0.86 and 1 are indicated by an asterisk in the main nodes.

Distinct North African ancestries in present-day domestic cats and Sardinian wildcats

To provide an updated phylogeny of present-day wild and domestic cats, we first reconstructed a Maximum-Likelihood (ML) autosomal tree (Fig. 1B, Fig. S1) using the genomes of present-day wildcats generated in this study along with samples from the literature (Table S6). We identified four main clades corresponding to the different wildcat taxa investigated, which corroborates previous studies illustrating that F. l. lybica, including its domestic form F. catus (hereafter F. l. lybica/F. catus when referred to as a single genetic cluster), is a sister group to F. l. ornata, F. silvestris and F. bieti (14). This analysis also showed that F. l. lybica populations are geographically structured with two distinct, highly-supported (bootstrap=100) clades corresponding to Levantine and African wildcats. Domestic cats form a distinct sister clade to the African wildcats, thus suggesting a closer genetic proximity to these than to the modern Levantine population. The present-day Sardinian wildcats clustered with the African wildcats and formed a monophyletic group with the Moroccan specimen outside of the variation represented by domestic cats.

We then assessed population structure in present-day wild and domestic populations using Multidimensional Scaling (MDS) based on identity-by-state (IBS) (Fig. S2) and an unsupervised ADMIXTURE analysis (Fig. S3). Both the MDS and ADMIXTURE at K=4 defined clusters that mirrored the main clades from the ML tree. Higher K values (K=5 to 7) identified substructure within the European wildcats and discriminated F. l. lybica from F. catus groups. Domestic cats shared more genetic affinities with wildcats from Africa, particularly with the Tunisian one, than with those from the Levant and Sardinia (K=5 to 7). F. l. lybica ancestry was better resolved at K=8, where a distinct component was assigned to the wildcats from the Levant, and the Sardinian wildcats shared more ancestry with the Moroccan sample. This was also illustrated by the MDS, in which Moroccan and Sardinian wildcats clustered together, consistent with the ML tree topology. Comparable results were found by conducting ADMIXTURE analyses on several subpanels of populations including only F. l. lybica wildcats (n=8) and different random subsamples (n=3) of domestic cats (Fig. S4, Supplementary text), to account for potential biases due to uneven sampling.

To further test whether present-day domestic cats are genetically closer to African than to Levantine wildcats, we employed outgroup-f₃ and f₄ statistics using the jungle cat (F. chaus) as an outgroup. Our results showed that all the present-day domestic cats shared more genetic drift with the present-day Tunisian wildcat (Fig. S5). Outgroup-f₃ and f₄ statistics conducted on the Sardinian wildcats showed that they shared more drift with the present-day Moroccan individual than with domestic cats (Fig. S6, S7). Although preliminary, due to the limited sample of wildcat genomes (three from the Levant and two from North Africa), these results allow hypothesizing that domestic cats and Sardinian wildcats derived from two genetically distinct populations in North Africa, represented in our dataset by Tunisian and Moroccan wildcats, respectively.

Domestic cats did not spread to Europe during the Neolithic

Previous studies using diagnostic SNPs from a fragment of the mitochondrial genome suggested that clades IV-A and IV-B, typical of F. l. lybica populations from the Near and Middle East, were amongst those that dispersed from Anatolia to central via southeastern Europe during the Neolithic (13, 15). Recently, this wave of Neolithic cat dispersal into Europe has been questioned by the demonstration that present-day Eastern European wildcat (F. silvestris) populations naturally possess some degree of Near Eastern wildcat ancestry probably due to admixture along a hybrid zone during the Late Pleistocene or Early Holocene (14). This study concluded that mtDNA haplogroups IV-A and B are thus not suitable diagnostic markers for tracing the dispersal of domestic cats. To address this, we analyzed genome-wide transversions and tested whether samples from Neolithic/Chalcolithic Bulgaria (n=4) and Neolithic Anatolia (n=4) (ranging 9,900-4,300 BCE) in which mtDNA IV-A was previously detected (13), were European wildcats.

We found that, at the nuclear level, all the samples (n=22) that dated from the 9th millennium to the 3rd century BCE project onto the European wildcat (F. silvestris) PCA cluster (Fig. 1C-D, Fig. S8). Crucially, this cluster includes Neolithic and Chalcolithic cats from Anatolia (n=4) and Bulgaria (n=4) dated from the 7th to the 5th millennium BCE that possessed a IV-A haplogroup (Fig. 1E, Fig. S9, S10), typically associated with F. l. lybica. All the other samples from across the rest of Europe dating from the 9th millennium to the 3rd century BCE (n=14), as well as one cat from Chalcolithic Bulgaria possessed both nuclear European wildcat ancestry and a mtDNA haplogroup I that is characteristic of F. silvestris (7). The mtDNA clade IV-A was also found in wildcats dating to the Roman era (Belgium and Italy, n=2), and Modern periods (Germany, n=3, Bulgaria, n=1) (13), thus implying the persistence of IV-A in southeastern and central European wildcats for the past seven millennia.

Conflicting mitochondrial and nuclear evolutionary histories (i.e. mitonuclear discordance) can be due either to hybridization, or to incomplete lineage sorting (ILS) (24). In this case, given the clear biogeographic pattern of mitonuclear discordance as illustrated in the ancient European wildcats (Fig. S11), ILS can be ruled out (25). By considering the effective population size estimated with a Bayesian Skyline plot and from the literature (Supplementary text, (26)), assuming neutrality and a complete mtDNA turnover in Neolithic Anatolian wildcats (~8,000 BCE), we computed the mean time required for haplogroup IV-A to reach fixation in the ancient Anatolian wildcat population (27, 28). Our results show that F. l. lybica IV-A mtDNA would reach fixation in not less than ~5,000 years in a small European wildcat population (N_e=1000), thus placing the admixture event no later than the Late Pleistocene (Supplementary text, Table S8, Fig. S12). Additional genome data from modern wildcats in Anatolia and the Caucasus, which are currently lacking, will help to test different admixture scenarios.

Our results demonstrate that cats previously found to possess a mtDNA clade IV-A in Neolithic Anatolia (~8,000-6,000 BCE) and Neolithic and Chalcolithic Southeast Europe (~5,500-4,000 BCE) (13) were not F. l. lybica/F. catus cats introduced by humans, but F. silvestris wildcats whose ancestors hybridized with F. l. lybica. This implies that European wildcats have been present in these regions for at least the last ten millennia (17, 29). Our genomic data thus corroborate previous evidence showing that haplogroup IV is not restricted to F. l. lybica, and that the matrilineage IV-A alone should not be used to infer either ancient wildcat distributions, or human-mediated translocations across continental Europe in the past (14). Although cats are traditionally viewed as commensals that frequented the human niche in the Neolithic Levant, recent zooarchaeological evidence suggests that wildcats were also hunted and exploited for food and fur (30). Regardless of the nature of the Neolithic cat-human relationships in the Levant, our data indicate that these did not lead to the spread of F. l. lybica/F. catus cats into Europe.

Of our 22 F. silvestris samples dating from the 9th millennium to 3rd century BCE, 20 originate from either archaeological settlements (i.e. Neolithic and Chalcolithic Anatolia and Bulgaria, Bronze Age Italy and Spain, Hellenistic Greece, n=15) or cave sites with evidence of anthropogenic activity (Late Paleolithic, Mesolithic and Neolithic Italy, n=5). The relationship between humans and wildcats in Europe was possibly based on exploitation for fur (31, 32), and food, as suggested by the Mesolithic samples from Galgenbühel/Dos de la Forca (33). However, more complex socio-cultural and symbolic relationships should not be discounted (34), considering the wildcat remains analyzed here from Bronze Age Partanna (Sicily, Italy) collected in a bell-shaped vase (35), and a feline clay head from Chalcolithic Bulgaria (36).

Human-mediated dispersal of F. l. lybica from the 1st millennium BCE

The earliest sample in our dataset that possesses F. l. lybica/F. catus nuclear ancestry is a cat from the site of Genoni (GSA01) in Sardinia (Italy) that has been directly radiocarbon dated to the 2nd century cal. BCE (Fig. 1C-D). In a PCA conducted specifically on F. lybica/F. catus samples (Fig. 2A), this individual clusters with three present-day Sardinian wildcats, and their genetic proximity was confirmed by outgroup-f₃ statistics used to assess the shared drift between GSA01 and ancient and modern F. l. lybica/F. catus samples. We found the highest f₃ values with present-day wildcats from Sardinia (Fig. S13). These results suggest that the origins of present-day Sardinian wildcats can be traced back over 2,000 years ago to an ancestral wild population to which GSA01 belonged.

Fig. 2. Genome-wide ancestries of ancient domestic cats.

(A) Principal component analysis built by projecting low coverage pseudohaploid samples onto the coordinate space defined by modern high- and mid-range coverage F. l. lybica and domestic (F. catus) cats. Similar clustering was obtained in analyses based on genotype likelihoods (Supplementary text, Fig. S14 and S15). Present-day samples are color-coded based on the clades of figure 1B, while symbols and colors of ancient samples are chronologically coded as in figure 1A. (B) z-scores resulting from the test f₄(F. chaus, ancient F. l. lybica; modern Levant F. l. lybica wildcats, modern African F. l. lybica wildcats). Each dot represents an ancient cat analyzed in this study. Significant values are reported as filled dots and non-significant values as empty dots. The dashed lines indicate the significance threshold (±3). (C) z-scores resulting from the test f₄(F. chaus, ancient F. l. lybica; modern African F. l. lybica wildcats, modern domestic cats). Colors and significance are as in Fig. 2B; the negative significant value of sample GSA01 is reported in red.

All other archaeological cats from Europe and Anatolia (n=42) that cluster with F. l. lybica/F. catus are dated from the 1st century CE onwards. These samples, including three specimens from Roman to 19th century Sardinia, cluster with present-day domestic cats (Fig. 2A). This was corroborated by f₄ statistics that we used to assess patterns of shared drift in F. l. lybica/F. catus between ancient and present-day wild and domestic cats. Similarly to GSA01, all the ancient domestic samples share more affinities with modern African wildcats than with their Levantine relatives (Fig. 2B, Fig. S16). In contrast to GSA01, ancient domestic cats were more closely related to present-day domestic cats than to North African wildcats (Fig. 2C).

A recent study has found evidence of gene flow between F. l. ornata, the Asian wildcat, and F. l. lybica wildcats from the Levant (14). Accordingly, our admixture analysis at K=4 (Fig. S3) detected F. l. ornata ancestry in the Levantine wildcats. To evaluate whether this ancestry could explain the reduced genetic affinity of domestic cats with the Levantine wildcats, we tested all the ancient and modern F. l. lybica/F. catus samples in our dataset for gene flow with the Asian wildcat by computing D-statistics. We found that all the tested individuals (Fig. S17) have an excess of shared alleles with F. l. ornata. This suggests that shared ancestry with the Asian wildcat is not affecting our results and that it is most likely due to gene flow before the divergence of F. l. lybica lineages from each other, or to ancestral genetic structure (37).

Overall, the human-mediated dispersal of F. l. lybica/F. catus into Europe may have occurred at least twice. The first translocation of F. l. lybica to Sardinia from a source population in Northwest Africa dates to the second half of the 1st millennium BCE at the latest, as is evidenced by the 200-50 cal. BCE GSA01 sample. It is possible that this population consisted of wild commensals in Northwest Africa that had not yet developed the behavioral features typical of domestic cats. After a second population that shares ancestry with modern domestic cats was brought to Sardinia from the Roman Imperial era onwards, the earlier introduced population remained largely isolated from their domestic counterparts, preserving its ancestral genetic signature.

Phoenicians, and later the Punic people established and maintained colonies in coastal northern Africa, Sardinia and the southeastern Iberian Peninsula from the 9th century BCE until their defeat by the Romans in the Third Punic War in 146 BCE (38, 39). Because Sardinia had been under Roman supremacy since 238 BCE (40), the translocation of F. l. lybica to Sardinia could have been carried out by either the Phoenicians and Punic people or by the Romans before the Imperial period starting 27 BCE (41).

Subsequently, since the Roman Imperial era, cats more genetically similar to present-day domestic cats were spread across Europe from a distinct North African population. The earliest sample carrying the ancestry found in present-day domestic cats was dated to 50 cal. BCE-80 cal. CE, from the site of Mautern, in Austria. Cats with this ancestry were then identified in Italy (n=4) and in Roman Imperial military sites along the Danube Limes in Austria and Serbia (n=6) (42–45), confirming previous claims that the Roman army and its entourage played a role in spreading domestic cats to central and eastern Europe (1, 46). This dispersal reached northern Europe relatively early in the Roman Imperial era, as testified by the genome of a cat from the site of Fishbourne, Britain, dated to 24-123 cal. CE (14). The ancestry typical of present-day domestic cats is then found continuously in all cats from Europe and Anatolia from the Byzantine era (n=7), the Medieval (n=23) and post-Medieval periods (n=2) until today (Fig. 1D).

The identification of the ancestral population that sourced the domestic cat dispersal remains challenging due to the lack of ancient and modern genetic data from African wildcats. Egypt is traditionally considered the core region for the domestication of cats and their subsequent spread (11–13). The close relationship between humans and cats in the Nile Valley is demonstrated by their role in the Egyptian state cult of Bastet, and as rodent-killers in the agricultural economy (10). This hypothesis is supported by the presence of the mtDNA haplogroup IV-C in eight out of 10 Roman Imperial cats, a haplogroup previously found in five Ptolemaic to Roman Egyptian mummies (13). Our observation that ancient and modern domestic cats share higher genetic drift with the Tunisian wildcat, however, may suggest an additional, if not alternative source population from the west/central North African coast of the Mediterranean. Large ports that served a fertile agricultural hinterland, as was the case in Phoenician-Punic Carthage (47), may have fostered the synanthropic association of cats with humans, and their subsequent maritime dispersal across the Mediterranean. Crucially, genome data from modern and ancient cats from Egypt, which are currently lacking, will allow these two hypotheses to be tested.

Different factors may have driven the translocation of cats to new cultural settings. Arguably, the association between cats and the cult of the goddess Bastet elevated the species’ prominence beyond Pharaonic Egypt, influencing Phoenicians (48), Greeks and Romans (49) alike. As with the human-mediated dispersal of domestic chickens (50, 51) or fallow deer (52), the initial translocation of cats could have been religiously motivated. Dispersal trajectories may also have been driven by the benefits of cats as pest controllers on ships, in view of the extensive maritime trade network of Carthage and the role of Egypt as a major grain supplier to the Roman Empire (53).

Overall, the 1st millennium BCE marks a pivotal period for the introduction of cats into Europe. This assumption is also corroborated by a Bayesian Skyline plot of F. l. lybica mtDNA lineages showing a rapid expansion starting ~2,000 years ago (Supplementary text and Fig. S18). Of three cats dated to this chronological range for which the genome was sequenced, two are F. silvestris from Greece, the only F. l. lybica being the 2nd century BCE cat from Genoni (Sardinia, Italy). More genome data spanning the 1st millennium BCE in Europe are required to better understand the chronological, spatial, and cultural settings of the human-mediated dispersal of domestic cats and test whether they were introduced to Europe prior to the Roman Imperial era.

Ancient and recent gene flow in wild and domestic cats

A recent study revealed that following the introduction of domestic cats to northern Europe, gene flow between indigenous wildcats and introduced domestic cats was limited, and the proportion of wildcat ancestry in ancient domestic cats ranged from 0 to 14% over the last 2,000 years (14). Analyses of domestic introgression in present-day wildcats from Scotland, however, demonstrated that much greater rates of hybridization have taken place over the last 70 years (54), possibly owing to a general decline in the wildcat population.

We assessed the interactions between wild and domestic cats across a temporal and geographic range stretching from western Europe to central Anatolia. In the PCA (Fig. 1C, Fig. S19), we observed that ancient cats from Anatolia and modern European wildcats from Scotland, Italy, and Germany spread out of the core of the European wildcat cluster, which may reflect a history of admixture.

We tested for the incorporation of F. l. lybica/F. catus ancestry into ancient and modern European wildcats by computing D-statistics. As a reference for European wildcats, we used a group (n=5) of ancient samples from Spain and northern Italy dated from ~8,500 BCE to ~4,500 BCE (from the Mesolithic to the Chalcolithic) that we ascertained not to carry any F. l. lybica/F. catus ancestry (Supplementary text, Fig. S20). As admixture sources we used the wildcats from the Levant and a domestic cat (Ocicat breed) that was devoid of F. silvestris ancestry (Supplementary text, Fig. S21).

In nine of the 11 samples from Anatolia and the Balkans dated from the Neolithic to the Iron Age (~6,400 BCE to the 3rd century BCE), we detected gene flow when using Levantine wildcats (Fig. 3A, Fig. S22A). When using the domestic source, only five were found to yield detectable levels of gene flow (one from Turkey, three from Bulgaria and one from Greece). Regardless, D-values in ancient Anatolian and Balkan samples were found to be higher in all instances when using the Levantine wildcats as the source of admixture. This pattern is less evident in present-day European wildcats, in which the levels of gene flow from both sources (Levant and domestic) were more similar.

Fig. 3. Patterns of gene flow between wild and domestic cats across time.

(A) D-statistics resulting from D(F. chaus, Ocicat/wildcat Levant; F. silvestris-ancient, test), where gene flow into ancient and modern European wildcats was tested against two distinct sources, a domestic one (F. catus, represented by the Ocicat breed), and a wild one from the Levant (F. l. lybica, represented by three wildcats from Israel). Colors indicate z-scores significance (±3) and symbols represent the source used, as in the legend. Error bars represent ±1 and ±3 (lighter blue/red color bar) standard deviations. In radiocarbon dated samples the age in thousands of years ago (kya) is reported. (B) D-statistics resulting from D(F. chaus, F. silvestris-ancient; Ocicat, test) to test gene flow from the European wildcats into ancient domestic cats. Colors indicate z-scores significance (±3), as in the legend. Error bars represent ±1 and ±3 (lighter blue/red color bar) standard deviations. In radiocarbon dated samples the age in thousands of years ago is reported. (C) Proportion of F. l. lybica ancestry in ancient and modern European wildcats from this study and the literature returning significant values of gene flow in the D-statistics. The proportions were computed with f₄-ratio following the model described in Fig. S23. Cats from Scotland and Germany were pooled, the number of samples is reported in brackets. Chronological range of samples as from radiocarbon dates and stratigraphy is reported in thousands of years ago. More details are in the Supplementary text. (D) Proportion of F. silvestris ancestry in the ancient domestic cats returning significant values of gene flow in the D-statistics of panel 3B. The sample PNE02, which returned a slightly significant D-value, is excluded from the figure as it did not show detectable proportions of F. silvestris ancestry. The proportions were computed with f₄-ratio following the model described in the Fig. S23. Chronological range of samples as from radiocarbon dates and stratigraphy is reported in thousands of years ago (kya)

We quantified the degree of gene flow by estimating admixture proportions with f₄-ratio (Fig. S23). They revealed ~7% to ~11% of F. l. lybica/F. catus ancestry in the ancient F. silvestris samples from the Balkans, and higher proportions in the Neolithic F. silvestris wildcats from western (~24%) and central Anatolia (~34%) (Fig. 3C, Fig. S24A). Modern wildcats from Italy exhibited ~10 to ~22% of F. l. lybica/F. catus ancestry, thus showing intermediate values between wildcats from Germany (~14%) and Scotland (~23%).

Overall, our data reveal two distinct sources of gene flow in European wildcats: an ancient one from F. l. lybica populations, and a more recent one from domestic cats following their dispersal into Europe. This supports that admixture first occurred in the Late Pleistocene, prior to the introduction of domestic cats in Europe. This likely stemmed from a contact zone between the two species located either in the Levant or in eastern Anatolia. This admixture led to the acquisition of F. l. lybica mtDNAs via mitochondrial capture, with possible unidirectional mating of local European wildcat males with Near Eastern wildcat females (55).

As previously suggested (14), this scenario would explain the mitonuclear discordance detected in Neolithic and Chalcolithic cats from Southeast Europe and Anatolia, as well as the presence of a cat in possession of a IV-A1 mtDNA in Mesolithic Romania (7,700 BCE, (13)). Unidirectional mating is also supported by previous genetic research showing a high occurrence of European wildcats carrying a F. l. lybica mtDNA (26%) as opposed to much rarer occurrence of domestic cats with F. silvestris mtDNA (0.4%) (7).

Lastly, D-statistics and f₄-ratio tests for the incorporation of F. silvestris ancestry into ancient domestic cats (Fig. 3B and 3D, Fig. S22B, S24B) showed detectable levels of gene flow from European wildcats in only two of 10 cats dated to the Roman era (up to ~6% in two cats from Austria dated to the 1st-2nd centuries CE). In the Middle Ages, the degree of introgression increased slightly, and F. silvestris ancestry rose up to ~15% in domestic cats from Europe (n=16), apart from all the Iberian and Sardinian samples. Our analysis thus suggests that following their introduction to Europe in the Roman era, domestic cats started to incorporate increasing levels of wildcat ancestry over time. However, this pattern does not extend to the Iberian Peninsula, where we were unable to detect gene flow in any of the six Medieval samples we investigated.

Overall, the diachronic pattern of admixture that we detected supports the hypothesis that habitat degradation and an encroaching human presence disrupted the ecological and spatial separation between wild and early domestic cat populations in Europe (14). Conceivably, deforestation and the expansion of agriculture over time led to a greater overlap of the home ranges and thus to increased gene flow between wild and domestic cats.

Conclusions

Our results suggest that the dispersal of domestic cats occurred several millennia after the Neolithic, and likely from North Africa rather than the Levant. Mediterranean civilizations during the 1st millennium BCE were probably the main agents of F. l. lybica translocation, which involved at least two genetically distinct populations of North African origin. One population likely comprised wildcats from Northwest Africa that were introduced to Sardinia and founded the present-day F. l. lybica populations on the island. The second population contributed the genetic pool of modern domestic cats.

Our results offer a new interpretative framework for the geographic origin of domestic cats, suggesting a broader and more complex process of domestication that may have involved multiple regions and cultures in North Africa. Efforts should continue to narrow down the original source population(s) of present-day domestic cats, and to clarify the cultural and socio-economic processes that led to their domestication and promoted their global dispersal. Currently, the genetic diversity of wildcat populations in North Africa and the Levant is represented by five individuals (two from Morocco and Tunisia, three from Israel). For this reason, it is fundamental to generate more ancient and modern genomes from these regions in the future, and particularly from Egypt.

Materials and methods

Extended descriptions of the archaeological contexts and methods are provided in the Supplementary text. A total of 225 cat bone remains from 97 archaeological sites across Europe and Anatolia were processed in this study.

Radiocarbon dating

37 samples from 30 archaeological sites (Table S6), were submitted for radiocarbon dating at the KIK-IRPA AMS laboratory, Royal Institute for Cultural Heritage, Brussels (56). Samples were prepared following pretreatment protocols described in (57). Dates were calibrated using OxCal 4.4 and the IntCal20 calibration curve (58).

Sample preparation and DNA extraction

Pre-PCR laboratory analyses of the archaeological samples were performed in the dedicated ancient DNA (aDNA) facility of the Center of Molecular Anthropology for Ancient DNA Studies of the University of Rome Tor Vergata, Villa Mondragone, Monte Porzio Catone (Rome, Italy). The Centre features state-of-the-art laboratories for aDNA consisting of individual ‘cleanroom quality’ working spaces (ancient sample pre-treatment, milling, DNA extraction, genomic library and PCR set-up fully equipped rooms).

The workflows adopted followed the standard precautions for access to the facilities and decontamination described in the literature (59). In particular, access to the pre-PCR laboratory was restricted to a limited number of people and only after wearing clean overalls, two pairs of gloves, over-shoes, surgical facemasks, plastic spectacles. Access to the aDNA facilities was not permitted if amplified libraries had been handled the same day. The aDNA working spaces were routinely cleaned with bleach and RNAse Away (Molecular BioProducts, San Diego, CA, USA) and every item entering the room was extensively washed with bleach or RNAse Away and UV-irradiated. Modern DNA and post-PCR laboratory analyses were conducted in physically separate buildings at the Department of Biology of the University of Rome Tor Vergata.

Archaeological samples

To minimize contamination, samples were UV-irradiated for 15 minutes on each side using a crosslinker set to 254 nm. Approximately 1 mm of the outer surface of each bone element was removed with a drill (Proxxon), while teeth were wiped with a tissue moistened with a 2% bleach solution, followed by a rinse with ultrapure water. Fine bone powder (80-150 mg) was collected using either a drill or a mortar and pestle, then stored in UV-irradiated vials at 4°C until DNA extraction. After processing each sample, the drilling room was thoroughly cleaned with a 2% bleach solution and UV-irradiated for 15 minutes. Drill bits were decontaminated by cleaning with a 2% bleach solution, rinsed with ultrapure water and 80% ethanol, and UV-irradiated for 30 minutes.

To maximize the recovery of endogenous ancient DNA, a two-step decontamination extraction protocol was applied as previously described (60), with minor modifications. Briefly, bone powder was washed in 1 mL of 0.5% bleach solution in a rotating incubator for 15 minutes at room temperature. Following a 30-second centrifugation at 13,000 rpm, the supernatant was discarded, and the pellet was washed three times with 1 mL of ultrapure water. The second decontamination step involved incubating the pellet in 1 mL of 0.5M EDTA (pH 8) for 30 minutes at 37°C in a rotating incubator. After pelleting the powder, the supernatant was removed, and a digestion was performed by incubating the pellet in 1 mL of lysis buffer (0.5M EDTA, pH 8, and 0.25 mg/mL proteinase K) for 24-48 hours at 37°C in a rotatory wheel. Silica-based purification of the extracts was performed as described elsewhere (61), with a few modifications: 3M sodium acetate was added to the binding buffer preparation, and the extract (~1 mL) was combined with 12 mL of binding buffer using the High Pure Viral Nucleic Acid Large Volume Kit (Roche). Purified DNA was eluted in 62 µL of TET buffer (1M Tris-HCl pH 8.0, 0.5M EDTA, 10% Tween-20, ultrapure water) and stored at -20°C until genomic library preparation. Extractions were performed in batches of eight samples, with two negative controls placed in positions 5 and 10.

Historic samples

Claws and skin samples of wildcats from Bulgaria (Table S4) from the collection of the National Museum of Natural History in Sofia were already sampled for a previous study (13). Extraction of DNA and library preparation of these samples were done in dedicated facilities for the genetic analysis of historic samples (dated to the 19th and 20th century CE) at the Department of Biology of the University of Rome Tor Vergata. A piece of tissue was collected for each specimen using a sterile surgical blade and weighed with a precision scale. DNA extraction was carried out using the Tissue and Hair extraction Kit and DNA IQ™ System (Promega), following the manufacturer’s instructions, and including a negative control.

Library preparation

Archaeological and historic samples

Illumina double-stranded genomic libraries were prepared as described in the literature (62), with a few modifications: 20 µL of DNA input was used, adapter concentration was reduced to 0.15 µM, purification steps were performed using the MinElute PCR Purification Kit (Qiagen), and DNA was eluted into 1.5 mL DNA LoBind® Tubes (Eppendorf). The final cleanup step was replaced by enzyme thermal deactivation (20 minutes at 80°C). All PCR tubes were UV-irradiated for 15 minutes with lids open before use. A library negative control (20 µL ultrapure water) was included in each preparation batch. Finally, genomic libraries were set up for PCR amplification using a double indexing strategy (63). In an experimental phase of the project where single-stranded library protocols were tested, one sample from Belgium (BRPAcat24) was processed with the Santa Cruz protocol (64), along with other samples from other projects conducted in the lab. Special SCR adapters and splints were ordered from IDT as suggested in the reference protocol.

Modern samples

Present-day samples were processed in the dedicated facilities at the Department of Biology of the University of Rome Tor Vergata. High-quality DNA extracted from skin tissues opportunistically collected from found-dead animals between 1998 and 2010 from Italy (n=12) and North Africa (n=2) were selected from the Italian Institute for Environmental Protection and Research (ISPRA) Felis DNA biobank (65). Sampled individuals had been previously identified morphologically by collectors according to phenotype, life history traits, and biometric indices. Genomic libraries were constructed using the NEBNext® Ultra™ II FS DNA Library Prep Kit with Sample Purification Beads (New England Biolabs Inc.), following the manufacturer’s instructions. DNA extracts were quantified using the Invitrogen™ Qubit™ 4 Fluorometer (Fisher Scientific) to ensure the appropriate adapter concentration. A library negative control (20 µL ultrapure water) was included in each batch. Indexing-PCR was set up using the NEBNext® Multiplex Oligos for Illumina®, supplied in a 96-well plate format with pre-combined forward (i7) and reverse (i5) primers, according to the manufacturer’s instructions.

Sequencing

Genomic libraries (ancient, historical, and modern) were amplified in the post-PCR facility of the Department of Biology at the University of Rome Tor Vergata using a dual-indexing amplification strategy (63). Genomic libraries of archaeological and historic samples underwent 15 PCR cycles, while modern libraries underwent 4 to 8 cycles, depending on the initial extract concentration. PCR cleanup for ancient and historical samples was performed using AMPure XP magnetic beads (Beckman Coulter Inc.), following the manufacturer’s instructions. Genomic library profiles were analyzed using the 2100 Bioanalyzer instrument (Agilent). Amplified libraries were diluted to 2 nM, pooled in batches of 30-40, and subjected to shallow shotgun sequencing on the Illumina NextSeq550 platform (75 SE, High Output kit) at the University of Rome Tor Vergata. Libraries with at least 10% endogenous DNA were selected for deeper sequencing on the Illumina HiSeqX or NovaSeq6000 (Macrogen Inc.), with few exceptions: the samples TERT03 (Tertenia, Sardinia, 8.4%), MENT01 (Menteşe, Turkey, 6.2%), and ASK01 (Aşıklı Höyük, Turkey, 5.9%) were deep sequenced given their chorological and geographic relevance. Differently, the sample SCAT03 (Torre Santa Caterina, Nardò, Italy), was not further processed because a sample from the same site and chronology (SCAT01, 17-18th c. AD) was already deep-sequenced. The sample DUR06 (Durankulak, Bulgaria, 5th millennium BCE) was not further sequenced due to the lower degree of post-mortem damage compared with the other samples from the same site (~8% vs ~30-40%). Similarly, the samples BAY01-03 (Bayraklı, Turkey, 4th century BCE) were not further sequenced due to the low degree of post-mortem damage (~1-4%) detected, which suggested inconsistency with the stratigraphic evidence. To confirm that, the radiocarbon measurement of the sample BAY01 yielded a value of 103.26±0.27 pMC, indicating a modern (post-1950) age.

Modern specimens were sequenced on the Illumina NovaSeq6000 platform (Macrogen Inc.) without prior screening. The number of reads generated are reported in Tables S2 and S3. Most of the ancient genomic data used in this study consist of ultra-low-coverage genomes, which are suitable for population genomic analyses after conversion to pseudohaploid format. To broaden our analytical approach, we selected five ancient domestic cat specimens for higher-coverage resequencing, enabling genotype likelihood estimation and reducing data loss. These include: MTR01B and PNE03, two Roman Imperial period cats resequenced at approximately 3.8-fold and 2.0-fold mean depth of coverage, respectively; GRA01 and SBR01, two Medieval cats with coverage of 2.3-fold and 2.6-fold, respectively; DH06 a domestic cat from Ottoman Turkey resequenced at 1.1-fold.

Raw data pre-processing and alignment to the reference genomes

Modern samples

Read quality was inspected using FastQC v0.11.9 (66). Quality filtering and adapter trimming of the reads was performed using AdapterRemoval v.2.3.3 with “--minlenghts 30”, “--minquality 25”, and “--trims” options (67). Separate alignments were generated for nuclear and mitochondrial genomes. Full genomes were aligned against the latest domestic cat reference genomes F.catus_Fca126_mat1.0 (GCF_018350175.1), by using the mem algorithm in bwa v0.7.17 (68) with default parameters. Aligned reads were sorted and indexed using samtools v0.1.19 (69). We assigned read groups with GATK v3.8 (70). PCR duplicates were removed using dedup v0.12.8 (71) and local realignment around indels performed using GATK v3.8 (70). The same pipeline was utilized for aligning reads to the mitochondrial reference sequence of Felis catus NC_001700.1 with few differences since we used a pseudo-circular version of the mtDNA reference sequence as done for the ancient data (see next paragraph).

Ancient and historic samples

We inspected raw fastq files with FastQC and performed quality filtering and adapter trimming of the reads with AdapterRemoval using the same parameters as for the modern samples but adding the option “--collapse”. Collapsed fastq files were then used to generate full-genome alignment following the procedure used for the modern samples with only one difference: we used the bwa aln algorithm, instead of bwa mem, with relaxed parameters (“-o 2”, “-n 0.01”) and seed disabled (“-l 1024”) as typically done for ancient DNA (72). For the alignment to the mitochondrial reference sequence of Felis catus NC_001700.1, we prepared a pseudo-circular reference version using the tool CircularGenerator of CircularMapper v.1.93.5 (73). Reads were then aligned to the elongated mtDNA reference with bwa aln again with options “-o 2”, “-n 0.01”, “-l 1024”, and then realigned to the original reference using the tool RealignSAMFile of CircularMapper. Finally, we assessed post-mortem damage patterns and performed rescaling of the bam files using mapDamage2.0 (74) for both nuclear and mitochondrial alignments.

Mitochondrial DNA consensus sequences generation

Previous work has found mtDNA insertions in the nuclear genome (NUMTs) of domestic cats (75) that can negatively impact mtDNA alignments and subsequent consensus sequence generation by introducing variants stemming from the nuclear inserts when mapping to the mtDNA. Recent analysis of wild and domestic cats from northern Europe showed that using high-coverage mitogenomes (>10-fold) generated from genomic libraries that present a high ratio between reads mapped against the mitochondrial and the nuclear genomes and a majority consensus call (14) does not result in incorporation of NUMTs variants in the mtDNA consensus sequence. We therefore computed the ratio between the reads aligned to the mtDNA and nuDNA genomes scaled by the respective genome lengths (Table S7). We decided to exclude three historic samples from Bulgaria due to the low coverage (<10-fold) and/or the low ratio between molecules aligned to mitochondrial and nuclear genomes scaled by the length of their reference sequences (ratio<35). One sample, ASK01 from Neolithic central eastern Anatolia, was retained given its relevance and since it was showing a high ratio between reads mapped against the mitochondrial and the nuclear genomes (>700) despite a mean depth of coverage of ~8-fold. All other samples showing a mean depth of coverage higher than 10-fold and mitochondrial versus nuclear aligned reads ratio from 40 to 2,300 were retained.

For each sample we first generated a raw VCF containing only variant sites using bcftools v.1.15 (69) (bcftools mpileup -BI -q 25 -Q 25 --annotate FORMAT/AD, FORMAT/ADF, FORMAT/ADR, FORMAT/DP, FORMAT/SP, INFO/AD, INFO/ADF, INFO/ADR sample.bam -f reference.fasta | bcftools call -mv -Ov -o sampleX_raw.vcf). From the raw VCF we generated a raw consensus sequence using bcftools consensus (cat reference.fasta | bcftools consensus -H I sampleX_raw.vcf > sampleX.cns.fa). From the same raw VCF we also extracted, using bcftools view and filter functions, low-quality variants (QUAL < 29) and positions with multiallelic calls for which the majority allele has a frequency lower than 0.8. We use the function vcf2bed from BEDOPS v2.4.41 (76) to store these variants in a bed file that we used to mask the raw consensus sequence by setting such positions as “N”, using the function maskfasta from bedtools v.2.30.0 (77). We finally computed coverage statistics using the tool genomecov of bedtools v.2.30.0 and extracted all positions covered by less than 5 reads. This list of low-covered positions was used for a second round of masking, setting again such positions as “N” in the final consensus sequence. We then used a custom script to concatenate all consensus sequences to use for phylogenetic analyses.

Genome-wide SNP panels curation

To date, a curated panel of Single Nucleotide Polymorphisms (SNPs) isn’t available for the wild and domestic cat species complex. We therefore adopted the GATK v.4 short variant discovery pipeline (78) and then performed hard filtering following GATK best practices recommendations (79). We used modern >10-fold coverage genomes from this study (n=12) and from the literature (n=44). Briefly, we first called variants separately for each sample bam files using the tool HaplotypeCaller in GVCF mode (option --ERC GVCF); then per-sample variants were collected in the so called “GenomicsDB datastore” using the tool GenomicsDBImport; finally, we used the joint genotyping tool GenotypeGVCFs to generate the raw multisample (n=56) VCF. Hard filtering was performed using the GATK tool VariantFiltration to mark low-quality SNPs by read depth (--filter-expression “QD < 2.0” --filter-name “QD2”), poor mapping quality (--filter-expression “MQRankSum < -12.5” --filter-name “MQRankSum-12.5” --filter-expression “ReadPosRankSum < -8.0” --filter-name “ReadPosRankSum-8”), or strand bias (--filter-expression “SOR > 3.0” --filter-name “SOR3” --filter-expression “FS > 60.0” --filter-name “FS60”). Marked variants were then filtered out with GATK tool SelectVariants (--exclude-filtered). We further filtered the variants with vcftools v.0.1.16 (80) retaining only biallelic SNPs with minimum quality of 40 (QUAL>40) and for which we have genotype information for all individuals (--max-missing 1 --minQ 40 --remove-indels --min-alleles 2 --max-alleles 2 --recode --recode-INFO-all), resulting in 61,999,591 variants. We finally converted the filtered VCF to plink binary format using the software plink v1.9 (81).

For the ancient low-coverage genomes (n=70) as well as historic (n=3) and modern samples (n=2) up to middle-coverage (~8X), we called pseudo-haploid genotypes on the 61,999,591 positions obtained from modern genomes. Pseudo-haploid genomes were generated using mpileup function of samtools v.1.15 (69) and pileupCaller from sequenceTools (https://github.com/stschiff/sequenceTools) by selecting a single read randomly (‘--randomHaploid’ parameter) for each individual at each of the targeted SNP positions. We then merged in plink the pseudo-haploid panel with the modern reference panel (n=131). Finally, the merged panel was filtered in plink for minor allele frequencies (maf) and linkage disequilibrium (LD) (--maf 0.02 --indep-pairwise 50 5 0.2) resulting in 3,954,446 SNPs with a genotypic call rate of ~56%. From this combined and filtered dataset, we prepared three panels for downstream analyses. The first consists only of modern diploid genomes (n=56), thus excluding all ancient and pseudohaploid data (PANEL-A). To reduce computational burden of lengthy phylogenetic and ADMIXTURE analyses, PANEL-A was thinned in plink by imposing a minimum distance between SNPs of at least 5kb, resulting in a total of 403,868 SNPs (PANEL-B). The third panel included also the ancient genomes (n=131) thus, to limit potential biases from deaminated cytosines, we removed all transitions resulting in 919,415 transversions (PANEL-C).

To account for the uncertainty of genotype calls in low and middle coverage samples we built a panel based on genotype likelihoods by calling variants using ANGSD v.0.937 (82). We included publicly available present-day F. catus (n=18) and F. l. lybica wildcats from the Levant (n=3), as well as modern F. l. lybica wildcats from this study from Sardinia (n=4) and North Africa (n=2). We also included in this panel five ancient domestic cats that we resequenced at higher depth (1- to 3.8-fold) to minimize the impact of missing data in downstream analyses. We therefore excluded transitions and applied the following parameters on each chromosome separately: angsd -nThreads 16 -b bam_list.txt -ref F.catus_Fca126_mat1.0 -uniqueOnly 1 -remove_bads 1 -only_proper_pairs 1 -trim 0 -C 50 -baq 1 -minMapQ 25 -minQ 25 -minInd 30 -rmTrans 1 -noTrans -GL 2 -SNP_pval 1e-6 -minMaf 0.01 -doGlf 2 -doMajorMinor 1 -doMaf 1 -skipTriallelic 1. After merging the outputs, the panel resulted in 4,250,120 autosomal transversions (PANEL-D).

Data analysis

Nuclear phylogeny of present-day wild and domestic cats

Phylogenetic analysis of a SNP-based supermatrix was conducted using RAxML-NG v.1.2.2 (83). The supermatrix was prepared by converting PANEL-B to VCF format in PLINK 1.9. The VCF was then converted to PHYLIP format using the python script vcf2phylip.py (84). A Maximum-Likelihood tree was constructed under the GTR+GAMMA model with default parameters, using 500 bootstrap replicates and Felis chaus as an outgroup. The resulting tree was visualized using Figtree v1.4.4 (available at http://tree.bio.ed.ac.uk/software/figtree/) (Fig. 1B, Fig. S1).

Population structure analysis

Population structure was first examined using Multidimensional Scaling (MDS) analysis of PANEL-A (Fig. S2). For this we applied the Classical Metric Multidimensional Scaling algorithm using the cmdscale function in R v.4.2.0 (85), which operates on an Identity-By-State (IBS) pairwise distance matrix that we generated in plink (plink --bfile PANEL-A --distance square -- out distance.matrix).

Unsupervised ADMIXTURE analysis was run using PANEL-B. We used Admixture v.1.3 software (86) for K values between 2 and 13, with 200 bootstrap replicates and a 5-fold cross-validation (CV) procedure. To account for uneven sampling and to focus more specifically on F. l. lybica/F. catus, we extracted all the F. l. lybica wildcats (from Morocco, Tunisia, Israel and Sardinia, n=8) and domestic cats (n=18) data from PANEL-A and prepared eight subpanels, each including all the wildcats and three randomly chosen domestic cats. Therefore, each subpanel consisted of a total of 11 individuals and, after imposing a minor allele count of 3, resulted in ~950,000 SNPs. We run the analysis for K values between 2 and 6. Results for all ADMIXTURE runs were visualized with pong v.1.4.9 (87) (Fig. S3, S4).

A principal component analysis of modern and ancient was run with PANEL-C. First, we converted this panel to eigenstrat format using the convertf function of eigensoft v.7.2.1 (88). We then applied a principal component analysis (PCA) using smartpca v.16000 (89), with the option LSQproject set to YES, so as to project low- to middle-coverage samples to the coordinate space built with modern high-coverage individuals (Fig. 1C). To increase resolutions, we extracted all the modern and ancient F. l. lybica and F. catus individuals (n=70) from PANEL-C and ran smartpca with the same option used above (LSQproject set to YES) (Fig. 2A).

Additionally, we run a F. l. lybica and F. catus PCA using the software PCAngsd v1.10 (90) with PANEL-D, which is based on genotype likelihoods. Results are in Fig. S14. Finally, we further explored population structure in F. l. lybica/F. catus using PANEL-D by running an admixture analysis using the software NGSadmix, which is part of the package ANGSD (91). Results (Fig. S15).

F-statistics

We assessed patterns of shared genetic drift using F-statistics in Admixtools v.2.0.4 (92) and exploiting different panels. In all tests Felis chaus was used as the outgroup.

By employing PANEL-A we ran: outgroup-f₃ statistic in the form of f₃(F. chaus, wildcats, domestic cats) to measure shared drift between domestic cats (n = 18) and different groups of F. l. lybica (Levantine, Moroccan, Sardinian, Tunisian) (Fig. S5a); f₄ statistics in the form f₄ (F. chaus, Domestic; Tunisian, other wildcats) to evaluate whether domestic cats are significantly more closely related to Tunisian wildcats than to the Levantine, Moroccan, or Sardinian wildcats (Fig. S5b); outgroup-f₃ statistics in the form f₃(F. chaus, Levantine/African wildcats/domestic, Sardinian) to assess relationships of Sardinian wildcats with other wild and domestic cats (Fig. S6a); f₄ statistics in the form f₄(F. chaus, Sardinian; Moroccan, Levantine/Tunisian/Domestic) tested whether Sardinian wildcats are significantly more closely related to the Moroccan wildcat (Fig. S6b); f₄ statistic in the form f₄(F. chaus, Domestic/Sardinian; Morocco, Tunisia) to further compare affinities of domestic and Sardinian wildcats with North African wildcats (Fig. S7).

Several F-statistics were run employing PANEL-C. For most of the following tests, modern samples were grouped according to the clades detected in the phylogenetic ML tree (Fig. 1B), distinguishing Levantine, North African (Morocco and Tunisia), Sardinian and domestic cats. Outgroup-f₃ statistics in the form f₃(F. chaus, ancient/modern wild/domestic cats, GSA01/modern Sardinian) tested affinities of the ancient Sardinian sample GSA01 with modern groups (Fig. S13a and S13b). f₄ statistics in the form f₄(F. chaus, ancient F. l. lybica/F. catus; Levantine, African wildcats) assessed whether ancient samples were genetically closer to North African or Levantine wildcats (Fig. 2B). The same f₄ statistic was repeated using the two North African sources separately using the form f₄(F. chaus, ancient; Levantine, Tunisian/Moroccan) (Fig. S16a, S16b). f₄ statistics in the form f₄ (F. chaus, ancient F. l. lybica; African wildcats, Domestic cats) were used to evaluate whether ancient samples were more closely related to modern domestic cats or African wildcats (Fig. 2C).

D-statistics

We used D statistics to assess patterns of gene flow between European wild and domestic cats over time. First, we tested whether ancient and modern F. silvestris experienced gene flow from F. l. lybica/F. catus using the form D(Felis chaus, Levantine/Moroccan/Tunisian wildcat; modern Portuguese F. silvestris, other ancient and modern F. silvestris) (Fig. S20).

To test if we could also use the modern domestic cats of our comparative dataset as source of gene flow into the European wildcats we assessed if they were devoid of F. silvestris ancestry by running D statistics in the form D(Felis chaus, ancient pool F. silvestris; Tunisian wildcat, domestic cat) (Fig. S21). As European wildcat proxy we used five ancient samples (pooled to minimize data loss due to missing data) that were found non positive to Lybica gene flow in the previous test; the Tunisian wildcat was modelled in all the previous analyses (see MDS, admixture and outgroup-f₃ statistics) as the closest wildcat genome to the ones of domestic cats, so it was selected as the non-admixed F. l. lybica reference for this test (Fig. S21). Finally, we repeated the test of gene flow from F. l. lybica/F. catus into ancient and modern European wildcats using the form D(Felis chaus, F. l. lybica Levant/F. catus Ocicat breed, ancient pool F. silvestris, other modern and ancient F. silvestris), with results shown in Fig. 3A and Fig. S22A. Finally, we explored patterns of gene flow from European wildcats into our ancient domestic cats, using the test D(Felis chaus, ancient pool F. silvestris; F. catus Ocicat breed, ancient domestic cats), with results shown in Fig. 3B and Fig. S22B.

To investigate whether F. l. ornata ancestry could explain the lower genetic affinity of Levantine wildcats from domestic cats, we computed D-statistics in the form D(Felis chaus, F. l. lybica/F. catus; F. bieti, F. l. ornata), as originally tested in Jamieson et al. (2023) (14), in all the modern and ancient F. l. lybica/F. catus of our panel (Fig. S17).

f4-ratio

We estimated admixture proportions in all the ancient and modern European F. silvestris and F. l. lybica/F. catus samples that resulted positive to gene flow. To do so, we used the qpf4ratio test in admixtools2. This test uses the ratio (α) of two f₄ statistics under the phylogenetic model illustrated in Fig. S23: α = f₄(A,O;X,C)/f₄(A,O;B,C), where A = Mesolithic to Chalcolithic F. silvestris from Italy and western Europe, B = Bronze Age F. silvestris from Italy, C = F. l. lybica from the Levant (n=3) or domestic cats (Ocicat Breed), O = F. chaus, X = test. Under this model α represents the proportion of F. l. lybica/F. catus ancestry in ancient and modern European wildcats (Fig. 3C and fig. S24A) while 1–α represents the proportion of F. silvestris ancestry in ancient domestic cats (Fig. 3D and fig. S24B).

mtDNA phylogeny

We used the multi sequence mtDNA alignments fasta file (see above) to build Maximum-Likelihood (ML) and Bayesian phylogenetic trees using the software IQ-TREE v.2.2.0-beta (93) and BEAST v.2.7.6 (94) respectively.

For ML phylogenetic analysis we generated a tree using 1,000 bootstrap replicates in IQ-TREE (-B 1000) and 1,000 SH approximate likelihood ratio test replicates (--alrt 1000) for assessing branch supports and the ModelFinder Plus option for model selection (-m MFP). The tree was rooted by using the F. margarita mtDNA (NC_028308.1) as outgroup (iqtree2 -s Cats_multisequence_alignement.fasta -B 1000 -alrt 1000 -m MFP -o FMA08). The resulting tree (Fig. S9) was visualized in Figtree v1.4.4 (http://tree.bio.ed.ac.uk/software/figtree/).

For the Bayesian phylogenetic analysis, the multi sequence fasta file mtDNA alignment was converted to the nexus format with Aliview v.1.28 (95) and different partitions were specified based on the mtDNA F. catus NCBI annotation (accession number: NC_001700.1) for the D loop, RNAs, 1st, 2nd, and 3rd codon sites. The nexus file was then loaded into BEAUTi (which is part of the BEAST package) to convert it into the BEAST input file and using the following parameters: a strict molecular clock and a lognormal distribution with a mean in real space of 1.0E-8; a HKY+ G substitution model with gamma category count of 4; tip dates were specified using the mean radiocarbon date, when available, or alternatively the dates based on the archaeological contexts. The resulting tree (Fig. 1E and Fig. S10) was visualized using Figtree v1.4.4 (available at http://tree.bio.ed.ac.uk/software/figtree/).

Bayesian Skyline Plot

We made use of mtDNA data to infer demographic histories of both domestic cats (clade IV, n=70) and European wildcats (clade I, n=33) by estimating the effective population size (N_e) with the Coalescent Bayesian Skyline (BSP) method implemented in Beast v.2.7.6 (94). For both analyses we created two partitions, for the D-loop and the coding region; we used the HKY+ G substitution model with gamma category count of 4 for both partitions; we set tip dates using the mean radiocarbon date, when available, or alternatively the dates based on archaeological contexts; we did not fixed a mutation rate; we set a strict molecular clock; we ran three independent MCMC with 100,000,000 iterations each. The independent runs were merged using LogCombiner v.2.7.6 after discarding 10% of the states as burn-in, resulting in a total of 270,000,000 iterations for both analyses. Finally, we visualized the skyline plots with Tracer v.1.7.2.

Modelling mtDNA turnover in ancient Anatolian European wildcats

By using a mean generation time of 3 years (54), we extrapolated the N_e of the overall European wildcat population from the Bayesian Skyline Plot. At a more regional level, recent estimates of N_e based on nuclear data from present-day wildcat populations from Germany returned values ranging ~1,000-2,000 individuals in the last 40,000 years (26).

Assuming a complete mtDNA turnover, we used the effective population size estimates as upper (N_e=20,000) and lower (N_e=1,000) bounds to compute the mean time required for haplogroup IV-A to reach fixation in the ancient Anatolian wildcat population at varying F. l. lybica introgression proportions. This was computed using the formula from Kimura and Ohta (28) modified as in Posth et al. for the mtDNA (27).

Data and materials availability

Genomic sequencing data are available through NCBI SRA Project accession no. PRJEB81815.