Introduction

Cat ecology

From an ecological perspective, the domestic cat is a very flexible animal. Despite being a hypercarnivore (i.e., highly dependent on a daily supply of meat), cats select prey opportunistically1,2. They can adjust to various conditions and successfully colonize a broad spectrum of environments1,3. Even when fed enough by their human owners, they search for other food sources and easily become feral (fully independent of people)1,3.

The history of the domestic cat and its ecological adaptations are currently highly debated topics4. Its domestication and rodent-controlling ability, followed by its global dispersal, is essential for our understanding of the diversity and evolution of human–animal relationships and the history of husbandry5,6,7,8,9,10. The domestic cat has become a widespread and abundant predator, and is considered one of the most detrimental species in the world that negatively impacts a wide range of wild native species3,11,12,13,14. Domestic cats also pose a serious threat to wildcat native populations through hybridization15,16 and competition17,18. Thus, the study of domestic cat ecology is important from various points of view—from the history of one of the world's most popular pet to conservation biology.

History of the domestic cat and its ecology

The earliest history of the human–cat relationship is still poorly documented, due to the sporadic occurrence of cat bones in archaeological and paleontological contexts, with records from the early stages of human–cat relationships containing long time gaps7,8,9,10,19,20. The Near Eastern ancestor of domestic cats had already emerged in Central Europe 6000 years ago; however, the possible relationships of this early population with modern domestic cats are unknown7,8,10. Still, the most plausible hypothesis assumes that the Roman Empire, with its military campaigns and far-reaching trade, played a major role in establishing a permanent domestic cat population across Europe7,21,22. However, archaeological evidence indicates that cats only became a constant and frequent component of the European anthropic landscape during the Middle Ages21,22. Viking expansion, followed later by trade contacts with the Ottoman Empire, played an important role in further distributing cats across the continent, and establishing the modern-day population7,19,21,23. During this period, cats became important pest controllers, and their numbers increased in parallel to the expansion of towns22. The high concentration of mice and rats and easy access to human food waste allowed dense cat populations to develop1,24.

Stable isotopes in tracking past cat ecology

The possibility of using stable isotopes to reconstruct diet became a powerful tool in archaeological studies focused on elucidating past human and animal ways of life25,26,27. For decades, this method has been used to detect past diets, providing insights into animal trophic ecology and adaptation to the anthropogenic landscape28,29.

The domestic cat is a complex object for stable isotope analysis because it opportunistically selects a wide range of food sources1,2. However, because it is a hypercarnivore, the bulk of potential food sources is relatively well defined. Until now, only few studies have investigated the ancient or modern cat diets using stable isotopes10,30,31,32,33,34. Here, we aimed to investigate the pattern of carbon and nitrogen isotopic variability in domestic cats from Northern Europe. Through analyzing a vast database of bulk collagen δ13C and δ15N values and comparing environmental and cultural conditions, we explore: (i) the range of isotopic variability among cats; (ii) predictability of cat isotopic signal in the context of its environment and presumed ecology; and (iii) usefulness of the stable isotope method to monitor the ecology and diet of this opportunistic and individualistic carnivore.

We used a collection of archaeological cat bones from medieval and post-medieval Europe, ranging from the 6th to the eighteenth centuries AD (Datasets S1 and S2). Due to the very low number of finds, which hindered statistical representativeness, we did not retain the earliest available cats (from Neolithic to Antiquity). We also excluded animals from the 19th to twenty-first century populations (i.e., cats from industrial times), as manufactured food and artificial pollution potentially strongly influenced their isotopic signal, making it difficult to interpret35,36. Our material originated from northern Europe, which is a region large enough to represent variable landscapes and socio-economic contexts, but is sufficiently restricted to avoid covering extremely different ecosystems, which could dominate the isotopic ecology and obscure minute differences.

Cat remains are scarce at archaeological sites. Complete or partial carcasses occur sometimes, but usually only single bones are found. Moreover, these finds often occur in archaeological contexts that provide limited information on cat ecology and relationship with people (such as latrines and waste heaps, or hideouts and natural shelters located far from human settlements). Therefore, our knowledge on the behavior, habitat, and place in the human cultural landscape of ancient cats is still limited. Thus, rather than deliberate over individual isotopic traits, we exploited isotopic trends among broad groups of cats in relation to their chronology, geography, and socio-economic contexts to determine population-scale isotopic patterns.

Results

Stable isotope results

We obtained new reliable isotopic results for n = 170 domestic cat individuals from 58 archaeological sites at 39 locations from Poland, Belgium, and Germany (Figs. 1 and 2; Datasets S1 and S2). Collagen quality checks (based on C:N atomic ratios) ranged from 2.9 to 3.6 (most were within 3.2–3.4), which was within the acceptable range for uncontaminated and non-degraded collagen37 (Dataset S2). Together with literature data taken from previously published studies on 8 additional locations38,39,40,41,42,43,44,45, we analyzed 188 individuals in total (Dataset S1). The isotopic values of domestic cats ranged from 6.0 to 13.8‰ for δ15N and from − 22.4 to − 15.5‰ for δ13C (Dataset S1).

Figure 1
figure 1

Locations with medieval and post-medieval cat remains studied for stable isotopes (1—York, 2—Wharram, 3—Koksijde, 4—Nieuwpoort, 5—Tournai, 6—Brussel, 7—Aalst, 8—Tongeren, 9—Logne, 10—Stavanger, 11—Bremen, 12—Dalheim, 13—Sigtuna, 14—Visby, 15—Ridanäs, 16—Lubin, 17—Wolin, 18—Kołobrzeg, 19—Stargard, 20—Żółte, 21—Stare Drawsko, 22—Słupsk, 23—Puck, 24—Gdańsk, 25—Malbork, 26—Elbląg, 27—Krosno Odrzańskie, 28—Zawada, 29—Bytom Odrzański, 30—Chełmno, 31—Poznań, 32—Grudziądz, 33—Kałdus, 34—Starogród, 35—Papowo Biskupie, 36—Zamek Bierzgłowski, 37—Toruń, 38—Gronowo, 39—Napole, 40—Karczyn, 41—Tum, 42—Grudusk, 43—Deszczowa Cave, 44—Perspektywiczna Cave, 45—Shelter in Udorka Valley I, 46—Miechów, 47—Sandomierz). Black circles—locations with new data; open circles—locations with literature data (see “Classification of the sites” section). Dotted lines show contours of the North Sea and Baltic Sea catchment basins.

Full size image
Figure 2
figure 2

δ13C and δ15N values vs. specimen chronology in the cats analyzed in this study. The chronology reported for each specimen is the central point (± 1-year accuracy) of the site chronology range or the central point of the 95.4% probability range of the radiocarbon date. For full chronology ranges see Datasets S1 and S2 and for each socio-economic context shown separately see SI Appendix, Figs. S1 and S2.

Full size image

Cat isotope signal along the timeline

The studied dataset included cat remains dating between AD ~ 600 and AD 1800 (Fig. 2; Datasets S1 and S2). Precise temporal fluctuations in the isotopic signal were difficult to detect. Many samples had wide chronological ranges (covering up to 550 years in some specimens, due to the lack of diagnostic archaeological artifacts or weak stratigraphic dating of some sites). However, a distinct shift in ranges of both δ13C and δ15N values appeared around the thirteenth century. This finding was supported by statistical analysis, which revealed significant differences in the variance of pre-AD 1250 and post-AD 1250 samples (Kruskal–Wallis test's p = 0.0001405 for δ13C and p < 10–7 for δ15N values) (SI Appendix Tables S2 and S3). This change included an increase in mean values (from − 20.3 to − 19.6‰ for δ13C, and from 9.5 to 10.6‰ for δ15N values) (SI Appendix Table S1). However, overall variability did not change (standard deviation from 1.1 to 1.2 for δ13C, and 1.2 to 1.3 for δ15N values) (SI Appendix Table S1). This shift coincided with the emergence of new types of human settlements, such as towns and castles46, and is represented in our database by the prevalence of different site types among the earlier and later periods (Fig. 2). Of note, the pre-AD 1250 cats in our database almost exclusively originated from the eastern part of the studied area (mainly Poland), while the post-AD 1250 samples originated from the entire studied area. Regarding the eastern region only, the statistical differences in δ13C and δ15N values between the pre- and post-AD 1250 cats were still significant (SI Appendix Tables S2 and S3).

Cat isotope signal vs. distance from the sea

We observed a significant correlation between δ15N values and distance from the sea for all our samples (Fig. 3, supporting statistics shown in Dataset S8). The δ15N values tended to decrease with increasing distance, with the rate of this shift being higher in sites located in the North Sea catchment area (Belgium, Germany, Norway, and UK; on average, 1.49‰ per 100 km) compared to the Baltic Sea region (Poland and Sweden; on average, 0.36‰ per 100 km) (Fig. 3). In the case of δ13C values, this correlation was weaker and only visible for cat samples from the North Sea catchment area (Fig. 3).

Figure 3
figure 3

δ13C and δ15N values in cats vs. site distance from the sea.

Full size image

δ13C and δ15N values differed in coastal cats from the North Sea and Baltic Sea regions (averages being higher for the North Sea, SI Appendix Tables S4–S7), following an isotope pattern already known for marine fish47,48. This was probably connected to general isotopic differences in the nutrient bases of the two basins. Baltic fish tend to have lower δ13C and δ15N values47,48; consequently, the isotopic signal of marine resources in the coastal areas of the Baltic region was less dissimilar to the terrestrial signals of inland resources. This observation supported the subdivision of our dataset into two geographic groups of cats, which were connected to two different marine-derived isotope baselines: cats from the North Sea region and cats from the Baltic Sea region. For this reason, we analyzed the data separately for these two sea catchment areas.

Cat isotope signal vs. socio-economic context of the site

Among the sites located in the Baltic region, cats from strongholds, castles, rural, natural and urban sites were statistically similar to each other in terms of δ13C values (analysis of variance p > 0.1 for each of these pairs for the null hypothesis of equal means, except for the castle–natural pair) (Figs. 4 and 5; SI Appendix Table S8). For δ15N values, similarities between these sites were weaker. According to ANOVA (or Welch F test), cats from rural sites differed too from those from other sites, except natural sites (SI Appendix Table S9). This result was the effect of much lower δ15N values in specimens from rural (and natural) locations.

Figure 4
figure 4

δ13C and δ15N values in cats vs. site socio-economic context, site geomorphology and locations (divided by sea catchment area). Colored fields are convex hulls covering samples attributed to each group. In locations panels, the locations that yielded < 5 specimens are denoted as ‘others’.

Full size image
Figure 5
figure 5

Variability of δ13C and δ15N values in medieval and post-medieval cats grouped by: (a,b) site socio-economic context; (c,d) site geomorphology; (e,f) pre- and post-AD 1250 chronology. For detailed values see SI Appendix Tables S1, S4–S7.

Full size image

Cats from seaports held a distinct position compared to other sites. These cats exhibited statistically significant different δ15N values to cats from other types of sites; except urban sites and castles. Their δ15N elevated values were attributed to higher contribution of marine food resources. For δ13C values, seaports statistically differed to strongholds, castles, and rural sites.

For the North Sea region, isotopic values significantly differed between most site types, in terms of both δ13C and δ15N (Figs. 4 and 5; SI Appendix Tables S10 and S11). In particular, there was a significant difference between urban sites and seaports (p < 0.001), which was not detected in the Baltic region (though, in the Baltic region, the null hypothesis of equal means was fulfilled for urban sites and seaports with weak probabilities, p < 0.15). The lowest δ13C and δ15N values were found at castle and rural sites, whereas the highest values related to seaports and a monastery located on the coast. The urban sites exhibited a wide range of values. The δ15N signal of single samples overlapped with the highest δ15N values detected at seaports, whereas the lowest signal fell in the range of castle and rural sites.

Cat isotope signal vs. geomorphological setting

To detect geomorphological factors that might shape the isotope signal of cats, we grouped the sites into five categories: inland, river valley, lakeshore, sea coast, and lagoon/fjord (Fig. 4; Dataset S1; definitions presented in “Classification of the sites” section).

For cats from the Baltic region, we found no statistically significant difference between sites next to freshwater and brackish water bodies (lakes, lagoons, fjords, and larger rivers). This lack of differences was visible in terms of both δ13C and δ15N values (high p-values of the analysis of variance, usually p > 0.2, SI Appendix Tables S12 and S13). Inland and sea coast locations differed from each other and from all other geomorphological settings in terms of δ13C values. In contrast, δ15N values differed for certain pairings (e.g., inland vs. sea coast, inland vs. lagoon/fjord, sea coast vs. river valley, and sea coast vs. lagoon/fjord), with a weak similarity for inland vs. lakeshore pairs (p = 0.0535, SI Appendix Table S13). River valleys and inland settings had the broadest range of δ15N values (Fig. 5).

For the North Sea region, we detected statistically significant differences between most of site geomorphology categories in terms of both δ13C and δ15N values (analysis of variance p < 0.005, but was much lower in most pairs, SI Appendix Tables S14 and S15). The only statistically similar pair is sea coast vs. river valley, for δ15N values. However, p value for this pair is close to the 0.05 confidence level. Similar to the Baltic Sea region, cats from river valleys had the widest range of δ15N, as well as a wide range of δ13C values (Fig. 5d).

Discussion

Potential sources of variability in the stable isotope values of cats

In general, the carbon and nitrogen isotopic signal (δ13C and δ15N) of a carnivore is controlled by many factors. These include biological factors (choice of prey, spectrum of prey, and isotopic variability of prey) and indirect environmental factors, which are responsible for the isotopic signal in prey (connected to vegetation type, aquatic/terrestrial ecosystem, soil development stage, climate, altitude)26,27,49,50,51,52,53. For domestic carnivores, direct or indirect anthropogenic impacts might also play an important role in shaping isotopic signals. These include, but are not limited to, various human activities: (i) feeding on consumption/production waste or on selected food; (ii) creation of new landscapes, opening up habitats for new prey species; (iii) farming selected plants, which serve as a food base for herbivores (for pest rodents hunted by carnivores, or for domestic stock delivering meat and/or dairy that is further fed to carnivores); and (iv) supplying food that would otherwise be of limited availability to a terrestrial carnivore (such as food acquired from fishing or distant trade). The feeding behavior of cats is also driven by local conditions, including currently available food resources, their relationship with people (indoor, free-roaming, feral), and individual behavioral characteristics1,4. Recent studies of modern cat diet indicate high behavioral and feeding differentiation in domestic cats1,17,18.

Terrestrial vs. aquatic resources

The most important factor shaping the isotope signal of the studied cats in the current study was the proportion of terrestrial and aquatic food resources in the diet. However, the regional isotope baseline (particularly the marine baseline), determined the degree to which differences could be detected through isotope values. In our dataset, the terrestrial and marine isotope signals were significantly different for cats from the North Sea region compared to those from the Baltic Sea region. Furthermore, the local background of human activity (especially cultivation of C4 plants and location of the fishing grounds and the type of fish exploited) could enhance or obscure the isotope indicators of consumed resources.

Cats from variable locations situated near water bodies (both marine and freshwater: sites on the sea coast, lagoons/fjords, river valleys, and lakeshores) had similar isotopic patterns with high δ15N values, whereas inland (i.e., far from the water bodies) cats had lower δ15N values (Figs. 5d and 6). Such elevated δ15N values are typical for aquatic ecosystems53. Thus, water-derived food resources contributed significantly to the diets of cats close to water bodies. Considering the socio-economic context of sites, higher δ15N values were expected in cats from seaports (Fig. 5b). Elevated δ15N values are also typical for cats from castles and strongholds in the Baltic region (Figs. 5b and 6). This is due to the cross-effect of cultural function and geomorphological setting of these sites, which were usually defensive locations situated on the edge of lakes or rivers (23 out of 25 of these sites in our database were attributed to river valley, lakeshore, or lagoon/fjord).

Figure 6
figure 6

Variance (square of standard deviation) of δ13C and δ15N values in medieval and post-medieval cats grouped by site socio-economic context, site geomorphology, distance from the sea, and chronology. Dashed contour line used for groups with low number of specimens (n < 10). Groups with n < 3 (only North Sea region: monastery and 200–300 km distance from the sea) are not included. For detailed values see SI Appendix Tables S1, S4–S7.

Full size image

Cats from non-coastal sites near to water (i.e., lakeshores, river valleys, and lagoons/fjords) were statistically dissimilar from those on the sea coast. This result was due to slightly lower mean δ15N values and clearly lower minimum δ15N values (Fig. 5d). This finding supported the expected difference between freshwater and marine fish54,55. In parallel, much lower δ13C values were obtained for cats from near-water non-coastal sites compared to cats from seashores (Fig. 5c). Despite being located quite far from the seashore, Bremen did not follow this relationship, as it had marine-like δ13C and δ15N values. However, availability of the marine resources in Bremen might have been due to the towns' trade hub function (see further discussion in “Cat isotope signal and the level of urbanization” and “Local conditions” sections).

The nitrogen and carbon isotopic signals in cats suggest an opportunistic shift toward local aquatic resources when available and a high reliance on such resources. This phenomenon might be related to scavenging fish remains or cats hunting terrestrial piscivores. Such prey might include sea birds, common in port towns and characterized by a marine-like δ15N signal56, and omnivorous rodents, which forage on fishery wastes and exhibit elevated δ15N values in port towns57,58.

Cat isotope signal and the level of urbanization

Cats from urban sites and seaports had the most varied δ13C values (Figs. 5a and 6). Cats from seaports had the highest mean δ15N values and were characterized by the lowest variability in δ15N (Figs. 5b and 6). These differences in δ13C and δ15N values variability indicated different primary food sources. The nitrogen signal is mainly related to the input of terrestrial (low δ15N values) vs. aquatic (high δ15N values) resources. The carbon isotopic signal depends on marine (high δ13C values, but there are other differences between littoral and pelagic ecosystems) vs. non-marine (terrestrial and freshwater) resources. The presence of C4 plants might also elevate the latter. The observed low δ15N and high δ13C variation in seaport cats indicated that aquatic resources were an important component of their diet, with mixed marine and non-marine resources, or mixed marine-littoral and marine-pelagic resources, or both.

The diet of cats from coastal locations was mainly driven by marine-derived food, whereas that of urban cats included a greater variety of food resources. Urban cats had high mean δ15N, but also, highly varied δ15N values (Fig. 5b). However, high-δ15N values of freshwater and/or marine food was still available to urban cats, regardless of whether it was naturally present in the local ecosystem. This finding demonstrates the importance of anthropogenic resources for these cats, as the availability of water-derived food was certainly due to trade. Medieval logistics allowed transport of fresh sea-fish up to 150 km inland59. This heterogeneity in the diet of cats in urban areas has also been identified in studies of modern cats11,32,58,60,61: cats consume more anthropogenic food in urban habitats, while predation rate and prey diversity are highest in rural areas.

Cats connected to natural locations exhibited relatively low variability for both carbon and nitrogen (Fig. 5a). They were characterized by low δ15N and δ13C values, which is typical for forest animals10,34. This observation indicates a narrow isotopic niche and trophic specialization, and shows that cats probably mainly inhabited forests and exploited wild prey. Studies of recent cats also support this explanation, showing that feral domestic cats tend to have a narrower trophic niche than house cats, with their diet similar to that of wildcats18.

The carbon and nitrogen isotopic ranges of rural cats overlapped with those from natural sites (Fig. 5a,b). However, rural cats exhibited a broader range of values, especially for δ13C. Like cats from natural sites, rural cats had lower δ13C values, and the lowest mean δ15N values, compared to other locations. These results indicate the use of just terrestrial prey resources, and predation over wild prey species. Increased predation rates in rural areas commonly occurs in modern cats11,34,60,61. While urban cats are more likely to be locked inside or living close to households, rural cats have lower population densities but have more extensive home ranges and live closer to natural habitats11,17,18,62,63.

Local conditions

Detailed analysis of our dataset allowed us to detect some abnormalities in the isotopic pattern. For instance, several sites from the Baltic region (Kołobrzeg, Puck, Napole, and Visby) stood out from the whole dataset due to their high δ13C signal (Fig. 4; Datasets S1 and S2). Several potential factors could be responsible for such a high δ13C signal: low forestation26, high elevation64,65, or significant contribution of marine food resources54,55. Another factor that elevates δ13C values might be attributed to the important contribution of C4 plants (e.g., millet) at the basis of the trophic web. These C4 plants were possibly stored cereals, which served as a source of food for pest rodents, which, in turn, were preyed upon by cats. The presence of millet was confirmed in archaeobotanical material from Puck66. However, in the case of coastal sites, such as Puck, Kołobrzeg and Visby, the marine resources could also play an important role. Cats had significantly different δ15N values in these seaports versus Napole. These seaports had the highest, whereas Napole had the lowest values out of the entire dataset. This difference followed the seaport vs. rural site function, and likely reflected differences in the proportion of marine food available at each site.

Another interesting example concerns fortified sites, namely strongholds and castles from the Baltic region. Cats from these sites exhibited the lowest isotopic variability for δ13C values and moderate variability for δ15N values. The absolute δ13C values were similar to natural and rural sites, whereas δ15N values were usually higher (Fig. 4). This result indicates that freshwater fish formed an important component of the diet, which is logical considering the location of these sites (23 out of 25 sites were attributed to the river valley, lakeshore, or lagoon/fjord geomorphological setting). Cats inside these closed settlements might have preferred living close to humans, and therefore had an easy access to table refuse.

The castle of Logne was the only castle situated in the North Sea region in our database. Logne cats had a clearly lower δ15N signal compared to cats from castles in the Baltic Sea region (Fig. 4). This result was unexpected when considering the generally higher δ15N values baseline in the North Sea region. Cats from this location (which is rather far from the coast), might have had limited access to fish and other water resources. The region has been densely forested until recent times, which suggests that the Logne cats could have had easy access to forest resources (via hunting or access to the table refuse of game animal meat), reflected in their low δ15N and δ13C values.

The opposite situation is visible in cats from Bremen, recording marine-like isotopic signals despite its location in the freshwater zone of the Weser67. Marine fish, however, were commonly available in Bremen as evidenced by zooarchaeological studies and historical records. Some anadromous species (e.g., sturgeon, salmon, sea trout, smelt, twaite shad, and flounder) used to enter the Weser during their spawning season and might have been fished locally68. The Bremen fishermen were organized in a guild from at least 1489, which held the fishing rights down to the river mouth69. Additionally, Bremen was an important trade hub since at least the eleventh century, particularly involved in the stockfish and herring trade70.

A model of cat isotopic ecology

In general, domestic cats may follow three main foraging strategies: (i) hunting, (ii) scavenging, mainly on human waste, and iii) direct feeding by humans. Considering these strategies, we may define the following factors responsible for an isotopic signal of cat diet, synthesized in Fig. 7:

  • Presence of cereal crops, cultivated in open fields, characterized by elevated δ13C values26,27, and available to synanthropic rodents (such as mice and rats, hunted by cats) either directly from fields, storages in barns or granaries, or from processed products;

  • Cultivation or import of C4 cereal crops with high δ13C signals (such as millet or maize)50,71;

  • Agricultural treatments, which may influence isotopic values of cultivated plants, such as manuring (elevating δ15N values in crops, especially in grains) and watering (lowering δ13C values in crops)49,51,52,72;

  • The proximity of forests, where cats may hunt for wild animals characterized by low δ13C and low δ15N values in natural habitats, due to canopy effect and forest soils26,73;

  • Access to meat/dairy from domestic animals that forage on agriculturally treated or untreated food (manured/not manured, watered/not watered, with/without C4 plants), and access to the meat of wild animals (provided by huntsmen, poachers, and, during medieval times, especially by knights and noblemen who were allowed to hunt in forests)74;

  • Access to fish (and other marine resources), usually characterized by an elevated δ15N signal and variable δ13C values, depending on fish ecology and physiology (i.e., differences between freshwater and marine fish54,55, geography (e.g., the difference between the Baltic Sea and North Sea47,48, or between pelagic and littoral waters53).

Figure 7
figure 7

Concept of the variability of carbon and nitrogen isotope signal in cats in typical medieval and post-medieval anthropogenic landscapes.

Full size image

In a human-dominated landscape, the isotopic signal of cat diet can be impacted by all of the listed factors. However, certain basic patterns related to variable anthropogenic and geographical settings can be predicted. In rural areas, crop pests are easy prey; however, the proximity of natural environments encourages the opportunity for cats to wander and feed on wild prey. Modern rural cats have a more diverse diet and a higher predation rate compared to those inhabiting urban sites11,60,61. Towns favor larger populations of synanthropic rodents, due to food stores and refuse. In addition, urban areas as trading spots provide various imported food resources, including marine fish, even in inland locations. However, even the largest inland towns cannot compete with seaports in terms of the availability of marine resources.

Conclusions

The ecology of the domestic cat is complex because it is an opportunistic exploiter that can maneuver among multiple available ecological niches1. Modern cats sharing the same environment, especially town-dwellers, exploit highly variable habitats and, hence, have highly varied trophic ecology1,12,16,17,31,72. Some cats may live as "backyard cats," (i.e., welcome but free-roaming mouse hunters), which are occasionally fattened by their human neighbors. Others might be "feral outcasts," which are the barely tolerated semi-wild dwellers of cellars, roofs, and hovels, living on their own and avoiding human contact. Still others might live as "indoor pets." Some cats even, exploit all these lifestyles.

The individualistic behavior and the taphonomic limitations mean that the isotopic signature of a single fossil cat specimen, or even of several specimens, might not represent the entire population. These limitations were revealed in our study through the high variability of the isotopic signal within the cats from some socio-economic types of sites (particularly those from seaports, urban and rural settlements) and geomorphological settings (especially river valleys and sea coasts). However, our data also revealed relatively low isotopic variability in cats from specific locations (mainly natural sites, outside of cultural context). This phenomenon most likely relates to feral cats, which can permanently or temporarily adopt a wildcat-like lifestyle. Similarly to modern cat populations, the medieval cats tended to have more heterogenous diet in urban areas, as they consumed more anthropogenic food. At the same time, the predation rate was higher in rural and natural areas.

The observed isotopic diversity means that stable isotope analysis could be a useful tool in elucidating the diet and lifestyle of individual cats. However, a local/regional isotopic baseline must always be carefully considered, as no global range may be arbitrarily attributed to the particular ecology of a cat. This study shows a promising perspective concerning the trophic ecology of fossil and recent cats, offered by an analysis of diet-related stable isotopes preserved in bones. Further investigations using sulfur stable isotope analysis and compound-specific carbon and nitrogen isotope analysis may deliver more in-depth insight into the domestic cat ecology.

Materials and methods

Zooarchaeological material

We collected the bones of small felids from archaeological and paleontological sites located in Poland, Germany, and Belgium (Fig. 1; Datasets S1 and S2). Our dataset included various geographical locations, covering human settlements of various socio-economic contexts, and natural sites. The available remains were either single bone specimens or almost complete skeletons. In total, 190 bones originating from 60 archaeological sites at 40 locations were sampled. The characteristics of the sampled specimens and sites included in this study are listed in (Datasets S1 and S2). We also included literature data of 18 felid bones from Sweden, Norway, UK, Belgium and Germany, from previously published studies38,39,40,41,42,43,44,45 (Dataset S3).

Bones most probably belonging to the same individual (based on: the same stratigraphic unit; different skeletal elements; similar size, individual age and preservation state), were sampled and analyzed separately, but their isotopic values were averaged (Dataset S4).

Taxonomic identification—domestic cat vs. European wildcat

Preliminary identification to the genus level (Felis sp. Linnaeus, 1758) was carried out in the Royal Belgian Institute of Natural Sciences and the Institute of Archaeology, Nicolaus Copernicus University in Toruń, using reference collections of domestic cat and European wildcat skeletons. The bones were measured according to the methodology by A. von den Driesch75, using a digital caliper (0.1 mm reading error) (Dataset S2). To determine the species (domestic cat, F. catus Linnaeus, 1758 vs. wildcat, F. silvestris Schreber, 1777, considered here to be European wildcat, F. silvestris silvestris) we adopted the discrimination method based on relative size using the log-ratio technique by T. P. O'Connor76 (Datasets S5, S6 and S7). The reference standards for domestic cat and European wildcat size ranges were based on Z. Kratochvíl's database77,78.

Due to limitations in distinguishing domestic cat bones from the European wildcat based on osteometric and morphological features, we also used mtDNA analysis, especially for specimens where the size ranges of domestic cats and European wildcats overlapped (SI Appendix Fig. S3; Dataset S2).

On the basis of morphometry and/or mtDNA analysis, we distinguished 139 domestic cats and 12 European wildcats. The specimens for which we could not obtain genetic or morphometric taxonomical information (n = 31) likely represent domestic cats due to their relatively small size, and have been regarded as such in further analyses. Specimens extracted from previous publications (n = 18) were also considered domestic cats.

Cat age criteria

We selected bones of individuals that were older than the nursing period, to avoid the milk diet obscuring trophic ecology of individuals. Cat dependence on the mother's milk is short and weaning is considered finished by about 7 weeks after birth1. Thus, for our study we selected adult individuals (with bone fusion completed) and subadult individuals, i.e. older than weaning age, based on combined bone fusion or tooth eruption79. For fragmented bones, when it was not possible to apply the criteria above, we used overall bone size and stage of bone ossification as indicators of post-weaning age (Dataset S2).

Classification of the sites

Chronology

The chronology of the specimens was presented as ranges in years AD. These ranges were based on the radiocarbon determination of the specimen if available, or/and on stratigraphic and archaeological dating (Datasets S1 and S2), using the published data (and in some cases, unpublished archaeological reports) (Dataset S2).

We used all radiocarbon dates available in the published literature for the previously studied specimens8,10,80, and new data obtained during the course of current projects by the authors (Dataset S2). In all cases, the dated fraction used was bone collagen. Most dates were obtained in the Poznań Radiocarbon Laboratory (Poznań, Poland), while some dates were obtained in the Royal Institute for Cultural Heritage (RICH, Brussels, Belgium). The presented chronology range was the end points of the 95.4% probability range, achieved through calibration in OxCal, v. 4.481,82. The ranges were provided with a ± 1-year accuracy.

The resolution varied among sites, from intervals of a part of a century to several centuries-long. We attributed chronological ranges to the stratigraphic data provided as centuries or parts of centuries in the following way:

  • a century: AD 1X00–1Y00;

  • early century: AD 1X00–1X25;

  • late century: AD 1X75–1Y00;

  • mid-century: AD 1X50;

  • 1st half of a century: AD 1X00–1X50;

  • 2nd half of a century: AD 1X50–1Y00.

On rare occasions the chronological data were more precise (i.e., narrowed to a period before or after a particular year). In such cases, we used these years as the range end. In cases where chronology was provided in a descriptive way, we adopted the following scheme: late medieval = AD 1250–1500; post-medieval = AD 1500–1800.

Geomorphology

We attributed a geomorphological context to the studied sites based on their geographic location (Dataset S1). We used the following types of geomorphological setting:

  • sea coast (less than 5 km from the coast of a sea or a bay; taking into account home size ranges and mobility to acquire food1,83,84);

  • lagoon/fjord (over 5 km from an open sea coast, but less than 5 km from a brackish basin, such as a lagoon or fjord, which was weakly connected with the open sea);

  • lakeshore (less than 5 km from a larger lake, or at an island);

  • river valley (within a valley or at the edge of a valley of a large river);

  • inland (none of the above, i.e., over 5 km from the nearest sea coast, brackish basin, lake or large river valley).

Site socio-economic context

We attributed a socio-economic context to each site based on data from the published literature (Dataset S1), also taking into account the chronology of the studied specimens for sites where the function changed during medieval/post-medieval times. We attributed the following functions:

  • natural (no historical nor archaeological indications for any anthropogenic use of the site during the considered period; mostly cave sites);

  • rural (small settlement, possibly farmland, often loosely constructed with wooden buildings);

  • urban (towns, administrative centers, densely built-up, densely populated);

  • seaport (important marine ports; all of them were also urban sites); in this category, we also include Bremen, which was involved in marine fishing and trade in the past68,69,70;

  • castle (late medieval fortified masonry seat of a noble family, a bishop, or knights of the Teutonic Order);

  • stronghold (mostly early medieval fortified settlement or refuge with wooden and earthen embankments; only in the eastern part of the studied area);

  • monastery (late medieval ecclesiastic settlement; only one example, Koksijde Monastery in Belgium).

Latitude, longitude, and distance from the sea

For cave sites, we provided the geographic coordinates of cave entrances. For open-air sites, we provided the coordinates of the center of a village/small town, center of a castle, or center of a street. We used Google Maps to obtain coordinates (decimal N and E values, ± 0.00001°). To provide the distance from the sea, we used the "measure distance" option in Google Maps. We measured this as the shortest distance from the coordinate-collecting point to the nearest sea coast line (± 1 km).

Stable isotope analysis

The bulk stable isotope analysis of carbon and nitrogen was performed on collagen extracted from bones. Dentine was not sampled to avoid bias from inter-tissue fractionation.

For samples with lab. no. CAT, collagen purification method followed the protocol after Krajcarz et al. and Bocherens et al.10,85. Isotopic measurements (δ13C and δ15N) and elemental measurements (elemental C:N ratio, calculated into atomic C:N ratio) were performed in duplicate at the Stable Isotopes Laboratory at the Institute of Geological Sciences, Polish Academy of Sciences (Warszawa, Poland) using a Flash EA 1112HT elemental analyzer (Thermo Scientific) connected to a Delta V Advantage mass spectrometer (Thermo Scientific). Mean SEs were < 0.33‰ for δ13C and < 0.43‰ for δ15N values. We normalized all measurements to δ13C values of USGS40 and USGS41 standards and all δ15N values to the IAEA 600 standard. Details are provided in SI Appendix Supplementary Text and Table S16.

For all of the remaining samples, collagen extraction was carried out at the Royal Institute for Cultural Heritage (RICH) in Brussels. Samples were prepared according to the method of Longin86, which was modified with an additional NaOH-wash between demineralization and hydrolyzation87,88. Collagen subsamples were sent to the Division of Soil and Water Management of the KU Leuven for stable isotope (δ13C and δ15N) and elemental (%C, %N) analyses. The atomic C:N ratio was then calculated. Analyses were carried out on a Thermo Flash HT/EA elemental analyzer, linked to a Thermo DeltaV Advantage Isotope Ratio Mass Spectrometer via a ConfloIV interface (Thermo Scientific). Data calibrations were done using IAEA-600 (caffeine) and two in-house standards (Leucine and muscle tissue of Pacific Tuna), which were previously calibrated versus certified standards. Analytical precision was typically better than 0.15‰ for both δ13C and δ15N values, this was based on the average of four replicates of the different standards in each batch of samples.. Details are provided in SI Appendix Supplementary Text and Table S16.

Data were reported in delta (δ) notation: δ13C or δ15N = ([Rsample/Rinternational reference] – 1) × 1000 in ‰ (parts per thousand), where R is 13C/12C or 15N/14N. The international references were Vienna Pee Dee Belemnite (VPDB) for carbon and atmospheric nitrogen (AIR) for nitrogen.

Statistics

We used one-way Analysis of Variance (ANOVA; or alternatively Kruskal–Wallis or Welch F tests in the case of non-parametric distributions or unequal variances) to check for statistical differences between isotopic variances of selected groups of samples (such as samples representing: chronological intervals, site cultural functions or site geomorphologies). The homogeneity of variance and normality were tested with Levene's tests (from means and medians) and a set of normality tests (Shapiro–Wilk, Anderson–Darling, Lilliefors and Jarque–Bera tests)89. Because in many cases the groups exhibited non-normal distribution and unequal variances, post-hoc pairwise comparisons were difficult to perform. Therefore, analysis of variance was run separately for each pair. In cases where post-hoc tests were possible for the entire set of pairs, we applied Tukey's test (after a significant ANOVA) with the Bonferroni correction of p values for multiple comparisons. All tests were performed with PAST software, v. 4.0590.

Methods of ancient DNA analysis

Genetic analyses of 75 samples were performed at the Centre of New Technologies at the University of Warsaw, Poland. All samples were analyzed in facilities dedicated to working with ancient DNA following strict rules to avoid contamination. Samples were prepared and processed as described by Baca et al.8. DNA was extracted using silica-coated magnetic beads according to Rohland et al.91. Double-strand, double-indexed library preparation, target enrichment of mitochondrial DNA (mtDNA), sequencing on the Illumina platform, and processing of the sequencing reads were performed as in Baca et al.8, with the modification that only positions with a minimum of 3× coverage were called. We obtained mtDNA sequences for 57 specimens, which were used in phylogenetic analysis, along with 123 already published sequences5,8,10. The Maximum Likelihood Tree was generated using IQtree92. The implemented ModelFinder determined that GTR is the best-fit model and the branch support was assessed using ultrafast bootstrap method (SI Appendix Fig. S3). All mtDNA sequences obtained in this study were deposed in GenBank under accession numbers: OL654306–OL654362.