Zinc isotopes from archaeological bones provide reliable trophic level information for marine mammals

In marine ecology, dietary interpretations of faunal assemblages often rely on nitrogen isotopes as the main or only applicable trophic level tracer. We investigate the geographic variability and trophic level isotopic discrimination factors of bone zinc 66Zn/64Zn ratios (δ66Zn value) and compared it to collagen nitrogen and carbon stable isotope (δ15N and δ13C) values. Focusing on ringed seals (Pusa hispida) and polar bears (Ursus maritimus) from multiple Arctic archaeological sites, we investigate trophic interactions between predator and prey over a broad geographic area. All proxies show variability among sites, influenced by the regional food web baselines. However, δ66Zn shows a significantly higher homogeneity among different sites. We observe a clear trophic spacing for δ15N and δ66Zn values in all locations, yet δ66Zn analysis allows a more direct dietary comparability between spatially and temporally distinct locations than what is possible by δ15N and δ13C analysis alone. When combining all three proxies, a more detailed and refined dietary analysis is possible. McCormack et al. present the first comparison of archaeological δ66Zn from bone material compared with traditional δ15 N and δ13 C values derived from collagen samples over a large geographic area. This study shows that the inclusion of δ66Zn analysis can provide otherwise inaccessible supplementary dietary information and more robust trophic level estimations.

I n ecology, archaeology and palaeontology, accurately reconstructing trophic levels can be challenging. Among others, these reconstructions are required for effective management and conservation strategies 1 , understanding changing predatorprey and foraging ecology related to climate change 2 and studying long-term environmental changes through the analyses of modern to fossil faunal assemblages (and their bone collagen stable isotope values) 3,4 . Stable isotope analysis is an effective tool for analysing marine food webs, complementary to and often more reliable than non-stable isotope approaches 5 . Bone collagen and soft tissue bulk δ 15 N and δ 13 C values are the traditional geochemical proxies used for dietary and trophic level reconstructions [6][7][8] . These isotope analyses can now be conducted on single collagen amino acids and provide more refined information on the trophic level of animals (e.g., corrected for baseline variability) 9,10 , but these studies are still time consuming and expensive.
Only recently, studies of the bone's mineral phase (bioapatite) non-traditional isotope systems such as calcium, magnesium and zinc (Zn) have shown potential as (palaeo)dietary and trophic level proxies in the terrestrial [11][12][13] and marine realm 14,15 . Element and isotope ratios observed in a diagenetically more resistant mineral phase (e.g., enamel) can preserve dietary information beyond the scope of application of collagen, as recently demonstrated for 66 Zn/ 64 Zn ratios (expressed as the δ 66 Zn value) 16 . In addition, even when collagen is well preserved, combining traditional collagen δ 15 N and δ 13 C analyses with Zn isotope analyses of bioapatite may provide complementary dietary information as proven by the lack of correlation of those tracers within individuals of the same species, implying independent controlling mechanisms 17 . Therefore, analysing δ 66 Zn alongside traditional δ 15 N analyses may help verify δ 15 N results and provide a muchneeded δ 15 N-independent additional geochemical trophic level and dietary indicator for ecological and archaeological studies.
Nitrogen-15 becomes relatively enriched in the tissues of aquatic consumers with successive trophic level 18 . The δ 13 C values behave much more conservatively with trophic level, increasing typically by <1‰ with trophic level for most tissues, compared to on average 3.4‰ higher δ 15 N values between a predator and its prey 19,20 . Carbon isotopes are therefore more commonly used to infer the source(s) of primary production at the base of the food web. For Zn, studies have shown a distinct 66 Zn depletion in carnivore bioapatite relative to that of herbivores 13,16 . As muscles and most organs are typically 66 Zn depleted relative to the animal's diet and its bulk body δ 66 Zn composition [21][22][23] , bones of carnivores (and their bulk body composition) have lower δ 66 Zn values than their prey's. Diet thus exerts control on the δ 66 Zn values of soft tissue and bioapatite. Most non-diet related factors, such as sex and age of an animal, have so far shown no effect on the isotope values and relative isotopic variability in body tissues 23,24 . Provenance, on the other hand, appears to affect the Zn isotope compositions of terrestrial vertebrates, though it remains unclear to what extent 13,16,25 . Provenance is known to play an important role for bulk δ 15 N and δ 13 C studies in marine food webs 26 but has until now not been studied for δ 66 Zn.
In the marine realm, the δ 15 N and δ 13 C values of particulate organic matter (POM), consisting of phytoplankton, bacteria, microzooplankton and detritus, show a substantial spatial variation within and among ocean basins 5,[26][27][28] . Variation in the isotopic composition at the base of the marine food web is passed along to higher trophic levels. Particularly for bone collagen, with its long turnover time 29 , trophic level reconstructions can be compromised when animals frequently migrate between areas of differing food web baseline isotopic composition, or when comparing animals from spatially and temporally distinct locations.
Here, we aim at exploring Zn isotopes as a reliable tracer of marine trophic levels. To do so, we investigate combined bone δ 66 Zn, δ 15 N and δ 13 C from the same species across 13 locations (17 sites) in a large geographic area stretching across the Arctic from the Hudson Strait in the east to the Bering Strait in the west. We include 5 locations and two single samples with already published δ 15 N and δ 13 C values 4,15,30,31 , as well as one site with already published δ 66 Zn values 15 . For this study, we analysed 167 archaeological bones, concentrating on ringed seals Pusa hispida (Phoca hispida) and polar bears (Ursus maritimus). Focusing on these species allows us to investigate δ 66 Zn trophic level isotope discrimination factors between predator and prey geographically. Both species have a circumpolar Arctic distribution and are abundant throughout the Arctic today 32,33 . Particularly, P. hispida remains are frequently found in archaeological assemblages with a large temporal and spatial range 30,34,35 .
As we focus here on archaeological material, we have no direct food web baseline information for the isotope systems analysed herein. Therefore, we take advantage of multiple studies documenting P. hispida as the main prey species of U. maritimus and subsequently both species being close to one trophic level apart 8,36-39 (Supplementary Note 1.1). We can therefore use these taxa to estimate species to species relative trophic level variability across the here studied sites. Studying the isotopic composition of high trophic level predators such as U. maritimus and P. hispida also has the advantage of their tissues' isotopic composition dampening the effects of short-term environmental variation and integrating multiple food web channels. This effectively leads to less isotopic variability and "noise" in the animal's tissues compared to those of lower trophic levels 30 . Therefore, these two species are prime targets to investigate the geographical variability of dietary proxies and trophic level isotopic discrimination factors. Anticipating an advantage of combining δ 66 Zn with δ 15 N and δ 13 C values, we also test in a series of bone dissolution experiments whether collagen-bound Zn influences bone δ 66 Zn values, which would preclude combining collagen extraction protocols with δ 66 Zn analysis of the mineral phase without resampling material.

Isotopic context
Marine phytoplankton from high latitudes shows particularly high variability in δ 13 C values 40 . Colder surface water temperatures lead to increasing aqueous CO 2 content, and therefore a net transfer of isotopically light CO 2 to the ocean and a depletion of 13 C in the surface water 41 . Other factors influencing spatial POM δ 13 C values include phytoplankton growth rates, cell size and cell lipid content (see ref. 5,40 and ref. therein). Additional spatial variability may arise from the relative contribution of sea ice POM (sympagic-POM) and open water phytoplankton (pelagic-POM) to a food web. Coeval pelagic-and sympagic-POM have differing δ 13 C values, with the former being 13 C depleted relative to the latter by 2-10‰ 37,[42][43][44] . Subtle shifts in Arctic consumers' δ 13 C values for a specific area over time may occur with largescale shifts in the relative importance of sympagic versus pelagic production related to changes in sea ice extent 4,30 .
A high variability in modern baseline δ 15 N and δ 13 C values is documented by the isotopic composition of POM, zooplankton, higher trophic level consumers, as well as filter feeders across the Arctic 6,45-50 . Pomerleau et al. 48 documented a significant spatial variability in zooplankton δ 15 N values among the Labrador Sea, Baffin Bay and the Canadian Arctic Archipelago (CAA), but not for δ 13 C among these areas. Subsequent studies documented a higher variability for δ 13 C of POM, zooplankton and high trophic level consumers among and within these areas as well 49,50 . In addition, a pronounced west-east 13 C depletion was observed throughout consumers from the Bering Sea (Bering Strait) through the Chukchi Sea to the Beaufort Sea 6,45,46,49 . A similar west-east trend was also found for sedimentary organic carbon accumulated along the Beaufort Shelf 40 . This eastward 13 C depletion trend reaches its maximum in the south-eastern Beaufort Sea. Terrestrial organic matter derived from the Mackenzie River has δ 13 C values of~−26 to −27‰ and dominates over autochthonous organic matter in the delta and at least parts of the Beaufort shelf 51 . Terrigenous 13 C depleted carbon is also thought to play an important role for some animals (gammarid amphipods) of the Mackenzie shelf's food web 52 . A similar variability in δ 15 N values between the Bering Sea, Chukchi Sea and Beaufort Sea is absent within animals of higher trophic levels 6,53 . However, geographic variations in δ 15 N values within these water bodies were observed for different zooplankton species 6,46,49 . Parson et al. 52 explained high δ 15 N values in POM of the Mackenzie estuary instead of a low terrigenous signal as a potential bacterial recycling of nitrogen.
The eastward decrease of baseline δ 13 C values does not seem to continue into the CAA 43,48 , but significantly lower δ 13 C values have been reported in consumer tissues close to the Canadian mainland and in semi-enclosed basins 54 . In accordance, De La Vega et al. 50 observed higher baseline δ 13 C values in inflow shelves connected to the Atlantic or Pacific Oceans (Barents Sea, Chukchi Sea) and the North Water Polynya (Northern Baffin Bay) compared to lower baseline values in the more freshwaterinfluenced Arctic shelves (Beaufort Sea, CAA, Hudson Bay).
Lower baseline values for carbon in the more terrestrialinfluenced areas are likely a result of terrigenous input and lower phytoplankton productivity. Higher stratification caused by inflowing freshwater hampers phytoplankton productivity on the interior shelves 55 . Indeed, Bering and Chukchi Sea annual primary production rates greatly exceed those of the Beaufort Sea 6 .
Zn isotopes are increasingly being used as tracers for past marine hydrochemistry 56,57 and culture experiments have investigated Zn isotope fractionation in different planktonic species 58,59 . Still, there is hardly any data on the Zn isotope composition of natural marine planktonic organisms 60,61 . Indeed, data on food web baseline δ 66 Zn values and variability is nonexistent for both the marine and continental realms. Because of biological uptake, dissolved Zn concentrations are highly depleted in marine surface waters, often much <1 nmol kg -1,62,63 and most oceans show a nutrient-like vertical distribution of dissolved Zn concentrations closely correlating with silicate concentrations 64 .
The isotopic composition of dissolved Zn below 500 m seems to be globally homogenous with values close to +0.5‰, despite variable Zn concentrations 65,66 . The bulk isotopic composition of dissolved marine Zn is enriched in 66 Zn relative to its major inputs from rivers and aeolian dust, which centre on the global crustal average of +0.3‰ 67 .
Although most studies on cultured phytoplankton demonstrate a preferential uptake of light Zn into the cell relative to the bulk growth medium 58,59 , Atlantic and Pacific vertical dissolved Zn isotope profiles generally show lower δ 66 Zn values in surface waters compared to that of the deep water 62,64,66,68,69 . These studies demonstrate that the isotopic composition of Zn is most variable within the surface water (<500 m), often with higher values in the uppermost surface (<20 m). Surface water dissolved Zn isotope ratios vary across a North Atlantic transect from −1.1 to +0.9‰ 62 and across a North Pacific transect between −0.9 and +0.2‰ 69 . Individual and combined mechanisms discussed to be responsible for this surface water δ 66 Zn variability include external inputs from rivers and aerosols 67,69 , scavenging of heavy Zn onto sinking organic matter 64 and biological uptake and shallow remineralisation 70 .
We are unaware of any δ 66 Zn data from dissolved Zn in the Arctic. However, a recent study on Western Arctic dissolved Zn concentrations highlighted a deviation of Zn concentration vertical profiles from the nutrient-type Zn profiles observed in the Atlantic and Pacific 71 . These authors documented higher than global average surface Zn concentrations (~1.1 nmol kg -1 ) with a maximum concentration at 200 m and uniformly lower concentrations in the deep water. Jensen et al. 71 hypothesises that Western Arctic surface water dissolved Zn originates primarily from incoming Pacific waters that are modified by shelf sediment fluxes from remineralised Zn-rich phytoplankton.

Results
Bone δ 13 C, δ 15 N and δ 66 Zn values. Bone collagen δ 13 C, δ 15 N and bone δ 66 Zn values of P. hispida, U. maritimus, harp seal (Pagophilus groenlandicus) and beluga whale (Delphinapterus leucas) are reported in Table 1 and Supplementary Data 1. All collagen samples had yields and elemental (wt% C, wt% N, C: N atomic ) compositions characteristic of samples with isotopic compositions not altered by contaminant or degradation in the burial environment 72,73 (Supplementary Data 1). Likewise, δ 66 Zn values do not indicate a modification due to diagenesis or contamination for the majority of samples, but we cannot exclude it   Supplementary Fig. 1). Within a site, we observe typically higher bulk collagen δ 15 N values and lower bone mineral δ 66 Zn values in U. maritimus relative to P. hispida and other prey species (Table 1, Fig. 1). Our results from applying different dissolution methods to reference materials and bone samples also indicate that the presence of collagen-bound Zn, and thereby collagen preservation, has no effect on the mineral phase δ 66

Discussion
The maximum variability for inter-site mean P. hispida bone δ 15 N and δ 13 C values (3.55 and 3.40‰) exceeds the maximum intra-site (1.77 and 2.67‰, Fig. 1) and typical trophic level variability in δ 15 N between predator and prey (e.g., +3.4 to +3.8‰) 8,18,19 . The QjJx-1 site on Little Cornwallis Island is a notable exception with very high on-site P. hispida bone collagen δ 15 N variability (3.85‰) 15 . Post hoc Games-Howell and Tukey pair-wise comparisons demonstrate a large heterogeneity in δ 15 N and δ 13 C values between archaeological populations (Supplementary Figs. 6-7). Isotopic heterogeneity between populations is related to geographic location resulting in δ 15 N and δ 13 C values from populations of different regions plotting in distinct groups on a δ 15 N versus δ 13 C plot (Fig. 1b). Based on site proximity and sample isotopic composition, we grouped sites from the Bering/ Chukchi Sea, Amundsen and Coronation Gulf, CAA, North Water Polynya, Hudson Bay, and sites influenced by the Labrador Sea (Hudson Strait and Frobisher Bay, Fig. 1a). The δ 15 N and δ 13 C variability between the archaeological sites is in good agreement in both spacing and amplitude with modern geographical variations observed from zooplankton 28,45,46 and higher consumer soft tissue 6,49 including P. hispida 37,53,54,[74][75][76] and U. maritimus 33 . While dietary differences between populations may have contributed to the geographic spacing of δ 15 N and δ 13 C values, integration of regional baseline isotopic patterns is likely the main factor controlling the observed inter-site isotopic 25 KkDo-1 n=9 JfEl-4 n=9 SfFk-4 n=8 KcFs-2 n=4 KkJg-1 n=11 RbJu-1 n=6 QjJx-1 n=11 NaPi-2 n=7 NkRi-3 n=5 OkRn  In contrast to P. hispida bone δ 15 N and δ 13 C values, the highest variability for mean δ 66 Zn values between sites (0.23‰) does not exceed the maximum variability observed within a single site (0.36‰). In addition, mean δ 66 Zn values between sites never exceed the variability between mean U. maritimus and P. hispida bone δ 66 Zn values (Δ 66 Zn U. maritimus − P. hispida = −0.32‰). While ANOVA analysis of P. hispida δ 66 Zn values did reveal statistically significant differences between P. hispida populations, post hoc pair-wise comparison tests show considerably more homogeneity in δ 66 Zn values than for δ 15 N and δ 13 C values ( Supplementary Figs. 4-7). Similarly, Levene's tests for equal variance show that for all sites, P. hispida and U. maritimus δ 66 Zn values have an equal variance, whereas δ 15 N values are more heterogeneous. The low geographical δ 66 Zn variability in P. hispida and U. maritimus bones implies that Arctic food web baseline and/or low trophic level consumer δ 66 Zn values are more homogenous than for δ 15 N and δ 13 C values. This is remarkable considering the large surface water's isotopic variability observed for dissolved Zn across the Atlantic and Pacific of −1.1 to +0.9‰ and −0.9 to +0.2‰, respectively 62,69 .
Based on post hoc Tukey pair-wise comparison, δ 66 Zn values from P. hispida populations from the sites QjJx-1 (Little Cornwallis Island), QkHn-13 (Devon Island) and SfFk-4 (eastern Ellesmere Island) were identified as statistically different from other populations (Fig. 2, Supplementary Fig. 4). Differences in the dietary Zn resources of these populations relative to others may have caused these statistical anomalies. Alternatively, they may reflect true variability in the δ 66 Zn regional food web baselines. QjJx-1 and QkHn-13 are located within the CAA. The CAA is composed of multiple channels and interconnected basins, in which water mass modification and transport are governed by its complex topography and shelf exchange processes 77 . Within this setting, baseline δ 66 Zn values are perhaps more variable on a regional scale than for the rest of the Arctic. For the SfFk-4 site, we observe in all three species analysed lower mean δ 66 Zn values compared to other sites indicating a regionally lower baseline δ 66 Zn value (Figs. 2, 3). The SfFk-4 site is located at the biologically highly productive 78 northern edge of the North Water Polynya, a region in which the reduced ice-cover or icefree conditions influence biological processes (e.g., by upwelling, increased nutrient renewal) 79 , which in turn may modify the δ 66 Zn baseline.
Both δ 66 Zn and δ 15 N values are controlled by diet but show a better correlation for U. maritimus samples across all sites than for P. hispida (Fig. 1c, e), perhaps related to the more specialised diet of U. maritimus 38,39 . As with δ 15 N values, bone δ 66 Zn values clearly demonstrate a trophic spacing between U. maritimus and P. hispida in all locations analysed (Fig. 2) 66 Zn is more homogenous in its value for a specific taxon, and possibly diet, δ 66 Zn may more reliably reflect trophic levels than bulk δ 15 N values, when investigating multiple species across multiple sites, proving a better inter-site comparability.
Ursus maritimus bone δ 66 Zn values are on average 0.32‰ lower than those of P. hispida (mean Δ 66 Zn U. maritimus − P. hispida = −0.32‰). Because P. hispida is typically the primary prey species of U. maritimus for most locations today 38,39 , we predict this Δ 66 Zn U. maritimus − P. hispida value to be close to the Zn bone trophic level discrimination factor between a carnivore and its prey, when soft tissue is consumed. Previously, estimations of trophic discrimination factors between bioapatite of terrestrial mammalian carnivores and herbivores were between −0.6 and −0.4‰, respectively for the Tam Hay Marklot (THM) cave 16 and the modern Koobi Fora region 13 . These studies, however, had a lower sample size and compared multiple carnivores and herbivores with varying diets. Predicted bone δ 66 Zn trophic level discrimination factors are between −0.36 and −0.38‰ when calculated using individual δ 15 N trophic levels 8 from all marine mammal taxa with available δ 66 Zn data (Supplementary Discussion 3.4, Supplementary Tables 1-2, Supplementary Fig. 8). However, these estimations are oversimplified, not considering population-specific dietary differences, location-specific baseline variations and organism-specific trophic and tissue-type enrichment factors. We cannot exclude different δ 66 Zn trophic level discrimination factors between tissues of P. hispida relative to their prey (which could not be analysed herein). The δ 66 Zn values of different tissues vary within an organism [21][22][23] . It is therefore possible that when different tissues are consumed (e.g., consumption of soft tissue only versus consumption of the whole organisms), δ 66 Zn trophic level discrimination factors might vary as well. Nevertheless, bone δ 66 Zn trophic level discrimination factors calculated from δ 15 N trophic levels 8 are close to the mean Δ 66 Zn U. maritimus − P. hispida value and should at least approximate true discrimination factors between bones of a carnivore and its prey.
Particularly for archaeological material, assigning a relative trophic level to multiple species when utilising δ 15 N values alone can be challenging, as shown by the large differences in mean δ 15 N trophic discrimination factors between U. maritimus and P. hispida (Δ 15 N U. maritimus − P. hispida ) for individual sites (+2.2 to +7.0 ‰, Fig. 2b). Besides locally differing diet, Δ 15 N U. maritimus -P. hispida variability (and Δ 66 Zn U. maritimus − P. hispida ) may be influenced by physiological effects or unknown archaeological assemblage effects related to human hunting and/or scavenging. Relative differences in the consumption of higher and lower trophic level prey alone are unlikely to explain the Δ 15 N U. maritimus − P. hispida variability. As P. hispida bones are the most abundant fauna remains in all archaeological sites analysed herein 30,34,35 , it stands to reason that they were a similarly important food item for the archaeological U. maritimus populations as they are today 38,39 . Therefore, we expect even substantial differences in P. hispida trophic levels among sites to have only a small effect on Δ 15 N U. maritimus − P. hispida . Feeding at substantially different trophic levels is incompatible with modern P. hispida and U. maritimus population trophic levels and diet variability 38,39,54 . In addition, most other U. maritimus prey species feed on lower or similar trophic levels relative to P. hispida 8,80 .
It is possible that due to the low intra-site sample size for both or either species, our site mean isotopic values do not capture the true means of the different populations. As the bones analysed are from individuals hunted or scavenged by humans, we cannot exclude differences in the segments of a P. hispida population hunted by humans and U. maritimus. For example, remains of P. hispida pups are very rare in archaeological assemblages 34,81 . Ursus maritimus, however, regularly preys on P. hispida pups and the contribution of pups to its diet may vary for different individuals, populations and with seal productivity 82,83 . As pups rely on their mother's milk, they effectively feed on a different trophic level leading to higher collagen δ 15 N values in pups than adults 84 . Consequently, a higher consumption of P. hispida pups by U. maritimus relative to humans can lead to higher Δ 15 N U. maritimus − P. hispida values within an archaeological assemblage. Additional uncertainties for inter-site Δ 15 N U. maritimus − P. hispida values may arise from a higher contribution of migratory species such as D. leucas 85 to the diet of certain U. maritimus populations 39 . Differences in the migratory behaviour of P. hispida 86 and/or U. maritimus 87 among sites could also lead to both species feeding along food chains with differing baseline δ 15 N values and therefore variable Δ 15 N U. maritimus − P. hispida values within the assemblages. It remains, as of yet, unclear if and how physiological effects may influence δ 66 Zn variability within a population. However, dietary differences as well as effects related to an archaeological assemblage (e.g., not capturing true population means) might have a similar effect on Δ 66 Zn U. maritimus − P. hispida as on Δ 15 N U. maritimus − P. hispida .
For δ 66 Zn, two sites from the same geographic area close to the Labrador Sea (KkDo-1 and JfEl-4) have a markedly lower trophic spacing between P. hispida and U. maritimus δ 66 Zn values of −0.22 and −0.24‰ (Fig. 2). Modern U. maritimus individuals from the area belong to the Davis Strait population 88 . In addition to P. hispida and contrary to most other U. maritimus populations, this one obtains a large percentage of its biomass from the consumption of harp seals (Pagophilus groenlandicus) 38,39 . However, bone δ 66 Zn values of P. hispida and P. groenlandicus from the same site are indistinguishable (Fig. 3, Supplementary  Fig. 9). Instead of the consumption of P. groenlandicus, the lower trophic discrimination factor for these sites may arise from unknown dietary contributions, population-specific physiological effects or unknown archaeological assemblage effects.
Unlike P. hispida, most P. groenlandicus leave their Canadian Arctic summering grounds, ahead of the formation of local pack ice in autumn 89,90 . However, when sympatric with P. hispida, P. groenlandicus feeds at a similar trophic level, consuming many of the same prey species, and both species show no statistical difference in muscle and liver δ 15 N values 91 . Indeed, bones of both seal species cover the same δ 66 Zn range for the same location (Fig. 3). However, bones of P. groenlandicus from Hudson Strait (KkDo-1) have almost 1‰ lower mean δ 15 N and δ 13 C values than those of P. hispida, perhaps related to this species' seasonal southwards migration (Fig. 3b, d). In contrast, some P. groenlandicus individuals remain in west Greenland waters during winter 92,93 , which may explain why bones of both seal species from eastern Ellesmere Island show a similar δ 13 C, δ 15 N and δ 66 Zn range (Fig. 3a, c). Due to its long turnover time, the bone collagen isotopic composition of P. groenlandicus likely represents an amalgamation of different food sources and local isotopic baseline values along their migration route and within their seasonal feeding grounds. P. groenlandicus bone δ 66 Zn values do not seem to record migratory signals, again arguing for lower baseline variability or Zn isotope homogenisation within low trophic level organisms. Despite a very low samples size (n = 2) beluga whale (Delphinapterus leucas) δ 66 Zn values fall within the same range as P. hispida and P. groenlandicus with slightly higher mean values (Fig. 3b, d). Indeed, all three species occupy a similar trophic level 8,91 . When sympatric with P. hispida, D. leucas typically has slightly lower soft tissue δ 15 N values likely due to migrating between areas with differing baselines or a more offshore/pelagic foraging 8,76 . Here, D. leucas δ 15 N values are higher than those of P. hispida and their δ 13 C values are highly variable (Fig. 3). Instead of only reflecting this species' trophic level relative to P. hispida, their bulk collagen δ 15 N and δ 13 C values are likely influenced by the high mobility of this species 94 and its foraging in locations with different isotopic baselines.
The trophic levels of U. maritimus, P. hispida and P. groenlandicus are reflected by their bone collagen δ 15 N and bone δ 66 Zn values across the Arctic. The analysis of Zn isotopes, however, offers additional advantages for studying marine trophic ecology, not only due to its greater preservation potential in fossil material 16 , but also due to the lower baseline controlled speciesspecific spatial isotopic gradients. The inclusion of δ 66 Zn analysis in ecological, archaeological and palaeontological studies may thus allow more robust interpretations of spatial and temporal trophic interactions. In addition, while both δ 66 Zn and δ 15 N generally record trophic levels, they do not record physiological and/or dietary effects equally, thus providing a strong incentive to combine δ 66 Zn with δ 15 N and δ 13 C analyses where possible.
In conclusion, this study compares archaeological bone δ 66 Zn values with traditional collagen δ 15 N and δ 13 C values for the same species across a large geographic area. Focussing on prey (P. hispida) and predator (U. maritimus) we investigate the baseline variability and trophic spacing of these dietary proxies. Our results show that: (1) Overall, δ 66 Zn values shows less site-specific variability within a species, likely due to a lower baseline variability than for δ 15 N and δ 13 C. As such, δ 66 Zn values are particularly valuable for dietary studies on highly mobile species (or consumers thereof) and for comparing geographically and temporally distinct populations.  Table 3.
Zinc analysis. All samples' surfaces were mechanically abraded (cleaned) to avoid sediment contamination, using a dental drill equipped with a diamond-tipped burr. Approximately 10 to 50 mg chunks were then sampled using a diamond-tipped cutting wheel. The chunks were then ultrasonicated in ultrapure water (Milli-Q water) for 5 min and dried in a drying chamber for a few days at 50°C. Bone samples and reference materials NIST SRM 1400 and NIST SRM 1486 were subjected to different dissolution methods (HCl and HNO 3 ) to investigate the impact of the organic bone phases on its Zn isotope signal (Supplementary Methods 2.1, Supplementary Discussion 3.2). The column chromatography steps (3.1.2) for quantitative recovery of sample Zn 95,96 was the same for all samples regardless of the dissolution methods used. Each column chromatography batch (n = 15) included up to 13 samples, one chemistry blank and at least one reference standard (SRM 1400 and/or 1486).
Zn purification was performed in two steps, following the modified ion exchange method adapted from Moynier et al. 96 , first described in Jaouen et al. 13 . Each step included AG-1 × 8 resin that was cleaned and conditioned prior to sample loading. One ml of AG-1 × 8 resin (200-400 mesh) was placed in 10 ml hydrophobic interaction columns (Macro-Prep® Methyl HIC). Resin cleaning involved 5 ml 3% HNO 3 followed by 5 ml ultrapure water. These cleaning steps were repeated. The resin was then conditioned with 3 ml 1.5 M HBr. After loading, 2 ml HBr were added for matrix residue elution followed by Zn elution with 5 ml HNO 3 . Following the second column step, the solution was evaporated for 13 h at 100°C and the residue re-dissolved in 1 ml 3% HNO 3 .
Zn isotope ratios were measured using a Thermo Fisher Neptune MC-ICP-MS at the Max Planck Institute for Evolutionary Anthropology (Leipzig, Germany) and a Thermo Fisher Neptune Plus MC-ICP-MS at the Géosciences Environnement Toulouse -Observatoire Midi-Pyrénées (Toulouse, France). Instrumental mass fractionation was corrected by Cu doping following the protocol of Maréchal et al. 95 and Toutain et al. 97 . The in-house reference material Zn AA-MPI was used for standard bracketing. δ 66 Zn values are expressed relative to the JMC-Lyon reference material. Analysed sample solution Zn concentrations were close to 300 ppb as was the Zn concentration used for the standard mixture solution. Zn concentrations in the respective samples were estimated following a protocol adapted from one used for Sr by Copeland et al. 98 , applying a regression equation based on the 64 Zn signal intensity (V) of three solutions with known Zn concentrations (150, 300 and 600 ppb). δ 66 Zn uncertainties were estimated from standard replicate analyses and ranged between ±0.01‰ and ±0.03‰ (1 SD). Additional reference materials SRM 1486 and SRM 1400 were analysed alongside the samples. SRM reference materials and samples show a normal Zn mass dependent isotopic fractionation, i.e., the absence of isobaric interferences, as the δ 66 Zn vs. δ 67 Zn and δ 66 Zn vs. δ 68 Zn values fall onto lines with slopes close to the theoretic mass fractionation values of 1.5 and 2, respectively.
Carbon and nitrogen isotope analysis. Bone surfaces were cleaned with a dental drill equipped with a diamond-cutting wheel. Subsamples of bone chunks (100-200 mg) were demineralised in 0.5 M HCl at 4°C. After demineralisation, samples were rinsed to neutrality with Type I water (resistivity > 18.2 MΩ cm). Any bone samples with dark colouration were treated with 0.1 M NaOH for successive 30 min treatments under sonication at room temperature until the solution no longer changed colour. The samples rinsed to neutrality with Type I water and then the insoluble collagen residue was solubilised in~8 ml of 0.01 M HCl at 75°C for 48 h. The resulting solution containing the solubilised collagen was filtered through a 5-8 μm filter and then filtered using a Microsep® 30 kDa molecular weight cut-off (MWCO) ultrafilter (Pall Corporation, Port Washington, NY) to remove low molecular weight compounds 99 . The >30 kDa fraction was freeze-dried, and the collagen yield was calculated.
Analysis of variance (ANOVA) were performed across the dataset in order to determine statistical differences in δ 13 C, δ 15 N and δ 66 Zn values between P. hispida populations. A single P. hispida specimen from Little Cornwallis Island was excluded from the statistical analysis (δ 66 Zn = 1.00‰, from Jaouen et al. 15 ), as it could disproportionately influence the analysis (see Supplementary Fig. 4 versus 5).
It was singled-out as an extreme outlier lying more than three times the interquartile range above the third quartile, both within-site and for the whole P. hispida dataset. Where variance was found to be significant, post hoc Tukey pair-wise comparisons were carried out to determine which populations were significantly different from each other in terms of their δ 13 C, δ 15 N and δ 66 Zn values. To adhere to ANOVA's assumptions, each P. hispida populations' δ 13 C, δ 15 N and δ 66 Zn datasets underwent visual inspection to check for normally distributed and homogeneous residuals, as well as tested for equal variance using Levene's test. Accordingly, we report the results of ANOVAs and post hoc Tukey pair-wise comparisons (Supplementary Figs. 4-7, Supplementary Data 3). As the δ 15 N dataset violated the equal variance assumption, an alternative Welch ANOVA was conducted instead, with post hoc Games-Howell pair-wise comparisons. In order to investigate the homogeneity of δ 66 Zn values within the Arctic relative to δ 15 N values, a series of Levene's test for equal variance (with Bonferroni correction) was performed on Zn and N isotope values between P. hispida and U. maritimus, as well as between sites for which data are available for both species. All statistical analyses were conducted using the free program R software 102 .
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
All data generated during this study are included in this published article (and Supplementary Data 1-3). Provenance information including sample ID, Bordon code, sampling location, feature and sample source are given in Supplementary Data 1, with additional site information in Supplementary Table 3. Received: 12 November 2020; Accepted: 12 May 2021;