## Introduction

Global distribution patterns of killer whales (Orcinus orca) indicate they are most abundant in coastal, high-latitude waters of both hemispheres, with lower densities occurring in less productive tropical and offshore waters1. In the northwest Atlantic (NWA), where there has been almost no directed research on killer whale distribution, incidental sightings and whaling records reflect a similar general pattern of decreasing abundance from the Arctic to the Gulf of Mexico and Caribbean Sea2,3,4. Over the past several decades, killer whales have been observed more frequently in both the eastern Canadian Arctic (ECA)5 and off the coast of Newfoundland and Labrador (NL)6, potentially reflecting shifting distributions and/or increasing abundance in response to declining sea ice5 or recovery from commercial whaling and culls7. Assessment of potential explanatory factors within the broader context of the NWA (e.g., where do these whales come from?), however, has been limited by existing observational data, which is largely qualitative and includes temporal and spatial biases.

Killer whales move into the seasonally ice-free waters of the ECA, where they are associated with summering aggregations of Arctic marine mammals2,8,9. They leave the region at the onset of sea ice formation for unknown overwintering areas10,11. Reeves and Mitchell2 suggested a possible continuous distribution between Canada and Greenland, but also hypothesized the range of ECA killer whales could extend along the North American coast as far south as the Caribbean, or into the open North Atlantic. Farther south, killer whales are sighted year-round off coastal NL, although less frequently during winter6,12. It is unclear whether this sightings pattern reflects seasonally variable observer effort, or a seasonal shift in distribution as has been documented in other killer whale populations13. Sightings of killer whales throughout the NWA during the summer months4,14, including both the ECA and NL6,15, provide evidence against a north–south migration, as does the lack of photo-identification matches between the two regions16. However, satellite telemetry10 and the presence of warm-water Xenobalanus barnacles17 indicate at least some killer whales observed in the ECA range into temperate and perhaps tropical waters.

One of the earliest ecological applications of stable isotope analysis has been the assignment of animal distributions through matching tissue isotope composition with underlying geographic isotope variation18. In both terrestrial and marine ecosystems, biogeochemical processes lead to systematic regional variation in baseline stable isotope composition known as ‘isoscapes’. For example, the interplay between physical and biological processes (e.g., air-sea gas exchange and primary productivity) affecting dissolved inorganic carbon (DIC) concentrations and associated 13C fractionation leads to pronounced latitudinal δ13C gradients in marine surface water carbon species and phytoplankton19,20. Baseline isotopic variation is incorporated into animal tissues via food and water (typically with some degree of isotopic discrimination that needs to be quantified and accounted for), thus serving as a marker of their geographic origins18. Although not commonly used in this context, the oxygen isotope composition (δ18O) of biogenic apatite is an effective marker of cetacean distributions across isotopically distinct regions of the ocean21,22. Marine surface water δ18O is governed by relative rates of evaporation and precipitation, resulting in a latitudinal gradient of several per mil in both hemispheres19,23. Cetacean body water closely tracks marine δ18O, as their dominant oxygen fluxes (transcutaneous water exchange, ingestion of water in food, and incidental or active ingestion of water while eating or drinking24,25) are not strongly fractionating26. Biogenic apatite in dentine and bone, in turn, precipitates in oxygen isotope equilibrium with a constant offset from body water in homeothermic mammals21,27,28, such that the δ18O of phosphate (δ18OP) and structural carbonate (δ18OSC) ultimately vary linearly with that of source marine water21,22,29.

The most pronounced variation in global marine δ18O occurs in the North Atlantic, where a latitudinal δ18O gradient spanning several per mil exists between the Arctic, which is characterized by 18O-depleted marine surface waters typical of high latitude regions, and the mid North Atlantic, which is characterized by 18O-enriched Gulf Stream waters23. The δ13C of DIC and phytoplankton also vary systematically across the North Atlantic, decreasing by up to 10 per mil moving northeast from the Caribbean to the Norwegian Sea19,20. These two elements together can thus provide complementary discriminatory power for evaluating North Atlantic whale distributions30,31. To that end, we measured δ18OP, δ18OSC, and δ13C of structural carbonate (δ13CSC) in biogenic apatite of dentine from killer whales that stranded in the ECA and NWA. Source marine water δ18O (δ18Omarine) was estimated from dentine δ18OP values using a published calibration22, and then compared to the North Atlantic δ18O isoscape23 to infer individual killer whale distributions. Our objectives were to evaluate whether ECA and NWA killer whale distributions (1) overlap within the broader North Atlantic, and (2) encompass the relatively 18O-enriched waters of the mid North Atlantic, as earlier hypothesised2 and recently demonstrated using satellite telemetry10. The δ18O values of most ECA and NWA killer whales indicated their distributions were restricted to the 18O-depleted, high-latitude waters surrounding the ECA, NL, and Greenland, although several individuals originated from a broader area encompassing 18O-enriched, low-latitude waters. These findings are discussed within the contexts of population structure and purported killer whale ecotypes in the ECA and NWA.

## Methods

### Specimen collection and sampling

Teeth from killer whales that stranded in ECA (n = 10 individuals) and NWA (n = 7) (Fig. 1) and mandibular bone from killer whales that stranded in Greenland (n = 2) and Denmark (n = 3) were acquired from government and museum collections (Table 1). Of the 10 ECA whales, nine either live-stranded or were found frozen in minimal state of decomposition; one had no stranding information. Of these 10 animals, two groups of three whales stranded together, while the other four were alone (Table 1). Of the seven NWA whales, five either live-stranded or were found in minimal state of decomposition, one was found in an advanced state of decomposition, and one had no information. Of these seven animals, two groups of two whales stranded together, while the other three were alone (Table 1).

Dentine was collected for δ18OP analysis by micromilling across one half of a longitudinally bisected tooth. Relatively small sample requirements of δ18OSC analysis also allowed for sampling of individual annual growth layer groups (GLGs) from a subset of ECA (n = 2) and all NWA (n = 7) whales to assess within-tooth δ18OSC variation. Mandibular bone was sampled from skulls lacking teeth using a handheld rotary tool. Bone should be directly comparable to dentine because the δ18OP of both tissues varies linearly with δ18O of environmental water with similar intercepts and slopes21,22, and any effect of cooler ambient temperatures of narrow jaws on 18O fractionation should be similar for teeth and mandibles29. Both whole-tooth and bone samples also integrate long-term isotopic composition, with the caveat that bone is subject to remodelling and dentine generally is not34.

### Oxygen and carbon isotope analysis of dentine and bone

All stable isotope analyses were performed at the Laboratory for Stable Isotope Science (LSIS) at the University of Western Ontario, London, Ontario, Canada. Dentine and bone δ18OP and δ18OSC were analyzed following procedures described in Matthews et al.22. Briefly, for isotopic analysis of phosphate oxygen, powdered samples (25 to 35 mg) were dissolved in 3 M acetic acid, and silver phosphate (Ag3PO4) was then precipitated following the ammonia volatilization method35,36. Aliquots of powdered Ag3PO4 (~ 0.2 mg) were then weighed into silver capsules and introduced into a Thermo Scientific™ High Temperature Conversion Elemental Analyzer (TC/EA). The resulting carbon monoxide (CO) gas was purified on a GC column packed with a 5 Å molecular sieve and swept in a continuous flow of helium to a Thermo Scientific™ DeltaPLUSXL™ isotope-ratio mass spectrometer (IRMS) for isotopic analysis.

Phosphate oxygen isotope ratios are reported in δ-notation relative to Vienna Standard Mean Ocean Water (VSMOW), calibrated using accepted values for standards IAEA-CH-6 (+ 36.40‰)37 and Aldrich Silver Phosphate–98%, Batch 03610EH (+ 11.2‰)38. The precision (SD) of these standard analyses ranged from 0.33‰ (IAEA-CH-6, n = 15) to 0.42‰ (Aldrich, n = 40), respectively. The accuracy of this calibration curve was tested using NBS 120c phosphate rock; 11 analyses returned an average δ18OP =  + 21.57 ± 0.29‰, which compares well with its accepted value of + 21.7‰39. The average reproducibility for replicate δ18OP analyses of samples was ± 0.22‰ (n = 13).

Structural carbonate oxygen and stable carbon isotopes were measured in untreated dentine since the effects of treatment to remove secondary carbonate and organic matter can be inconsistent22,40,41. Powdered samples (~ 0.8 mg) were dried overnight at 80 °C in a reaction vial, which was then septa-sealed, capped, and evacuated using an automated sampler (Micromass MultiPrep). Ortho-phosphoric acid was then added to generate carbon dioxide gas (CO2) by reaction at 90 °C for 20 min. The evolved CO2 was cryogenically scrubbed and transferred to an IRMS (Fisons Optima) for analysis in dual-inlet mode.

δ18OSC values were calibrated relative to VSMOW using accepted values for NBS 19 (calcite; + 28.65‰) and NBS 18 (calcite; + 7.20‰), with a precision (SD) of 0.05‰ (n = 51) and 0.12‰ (n = 27), respectively. Accuracy and precision (SD) were assessed using laboratory reference materials not included in the calibration curve: WS-1 (calcite; δ18O measured =  + 26.22 ± 0.22‰, n = 28; accepted =  + 26.23‰), and Suprapur (calcite; δ18O measured =  + 13.26 ± 0.13‰, n = 24; accepted =  + 13.30‰). δ13CSC values were calibrated relative to VPDB using accepted values for NBS 19 (calcite; + 1.95‰) and LSVEC (Li-carbonate, − 46.6‰), with a precision (SD) of 0.05‰ (n = 51) and 0.13‰ (n = 24), respectively. Accuracy and precision (SD) were assessed using international and laboratory reference materials not included in the calibration curve: NBS 18 (calcite; δ13C measured = − 5.08 ± 0.09‰, n = 28; accepted = − 5.01‰); WS-1 (calcite; δ13C measured =  + 0.76 ± 0.09‰, n = 28; accepted =  + 0.76‰), and Suprapur (calcite; δ13C measured = − 35.64 ± 0.11‰, n = 24; accepted = − 35.55‰). The average reproducibility for replicate analyses of samples was ± 0.14‰ (n = 16) for δ18OSC and ± 0.07‰ (n = 16) for δ13CSC.

### Data analysis

We first tested for temporal baseline isotope shifts (e.g., refs.42,43) since specimens were collected over an ~ 70-year timespan. Whale ages were estimated from counts of GLGs observed under reflected light, and the median of three counts conducted by one reader (CJDM) over several weeks was taken as the age estimate (Table 1). Calendar year representing the mid-point of tooth deposition was then estimated by subtracting half the whale age from year of death. Linear regression of δ18OP against calendar year showed no evidence of temporal baseline shifts in isotopic composition over the timeframe (Figure S1). Values of δ13CSC, however, exhibited an apparent linear decline over the same period, although this relationship was not significant (Figure S2). However, the estimated slope (− 0.023‰ year−1) was similar in direction and magnitude to declines in North Atlantic δ13C attributed to the Suess effect42,43. δ13CSC values were therefore adjusted to the most recent sample year (2013) by subtracting 0.023‰ year−1 to remove temporal bias.

Differences in δ18OP among groups based on stranding location were assessed using the non-parametric Kruskal–Wallis rank sum test, since residuals from one-way ANOVAs violated assumptions of normality and heteroscedasticity. In addition to the ECA and NWA killer whales, we included data from ECA beluga whales and Gulf of Mexico (GoM) dolphins22 (Table S1) as δ18O reference ‘endpoints’, as they are from regions representing extreme low and high ends, respectively, of the δ18O range across the North Atlantic. We assume any biases stemming from different body sizes or diets of the species are minimal, given the relatively small deviations (< 1‰) of dentine/bone δ18O of a large number of cetacean species from linear relationships with environmental δ18O21,22. Although the two GoM dolphin species, common bottlenose dolphins (Tursiops truncatus) and Atlantic spotted dolphins (Stenella frontalis), exhibit nearshore-offshore habitat partitioning along a gradient in surface water δ18O values23, their δ18OP were statistically indistinguishable22 and therefore grouped in our analyses. Greenland and Denmark killer whales were excluded from these analyses given their small sample sizes. Significant Kruskal–Wallis tests were followed by Dunn’s multiple comparison post-hoc test with Benjamini–Hochberg adjustment of p-values44. All analyses were done using R version 4.0.032 and its associated package ‘FSA’45.

Hierarchical cluster analysis of δ18OP based on a Euclidean distances and average linkage was performed on all killer whales (including Greenland and Denmark samples) using the R packages ‘cluster’46 and ‘dendextend’47.

Geographic distributions were assigned by first estimating source δ18Omarine from each individual killer whale’s dentine δ18OP by rearranging the following published calibration:

$$\delta^{18} {\text{O}}_{{\text{P}}} = \,\upbeta _{0} +\upbeta _{1} \times \delta^{18} {\text{O}}_{{{\text{marine}}}} ,$$
(1)

where the intercept (β0, 18.73 ± 0.30‰) and slope (β1, 0.81 ± 0.23‰) estimates are based largely on dentine δ18OP measurements of cetaceans from the North Atlantic and associated basins with heterogenous δ18Omarine22. Intercept and slope standard deviations and δ18OP measurement error were propagated through to final δ18Omarine estimates using first-order Taylor series expansion in the R package ‘propagate’48. δ18Omarine values encompassed by the estimated 95% CI for each individual killer whale were then extracted from modelled North Atlantic gridded surface δ18O data23 downloaded from Global Seawater Oxygen-18 Database49. Marine δ13C isoscapes20 were not used in this manner because confounding dietary and metabolic influences on dentine δ13CSC values50 could not be constrained to allow for direct comparison with baseline DIC or phytoplankton values.

Finally, the standard deviation of individual GLG δ18OSC measurements was calculated as an index of within-tooth variation for each of the 9 teeth that were measured.

### Ethics approval

Ethics approval was not required for museum specimens used in this study. Destructive sampling was formally approved through museum protocols.

## Results

Individual killer whale dentine δ18OP values ranged from + 15.94 to + 22.40‰ (Table 2, Fig. 2). Mean δ18OP values differed among species/regions (Kruskal–Wallis, Chi square = 16.2, p = 0.0010, df = 3), although pairwise comparisons indicated whales that stranded in the ECA (+ 17.21 ± 1.11‰) and NWA (+ 18.45 ± 1.81‰) were not significantly different (p = 0.087). ECA killer whale δ18OP did not differ from ECA belugas (+ 17.38 ± 0.69‰, p = 0.48) but was significantly lower than GoM dolphins (+ 18.82 ± 0.86‰, p = 0.0027) (Table S1). The δ18OP of NWA whales did not differ from ECA belugas (p = 0.14) or GoM dolphins (p = 0.23). Mean bone δ18OP values of the Greenland and Denmark killer whales were + 17.4 ± 0.3‰ and + 17.4 ± 0.4‰, respectively (Table 2, Fig. 2). Suess-adjusted δ13CSC values ranged from − 15.1 to − 11.5‰ across all whales (Fig. 2).

Cluster analysis of δ18OP produced groupings that were not consistent with stranding location. The most distant clusters comprised five killer whales from both the ECA and NWA with the highest δ18OP values (Fig. 3). The most distant whale, NWA-BP-1998, had a δ18O value that was 2 to 6‰ higher than the rest of the sampled whales (Fig. 3). The remaining clusters comprised killer whales that stranded in the ECA, NWA, Greenland, and Denmark, with no clear patterns or divisions based on stranding region within them (Fig. 3). Most whales that stranded together clustered together with similar δ18OP (whales NWA-SC-1975-1 and NWA-SC-1975-2; whales ECA-SQ-2016-1 through 3; and whales NWA-CB-1971-1 and NWA-CB-1971-2); the exception was the three whales that stranded together in Cumberland Sound in 1977 (the ECA-CS-1977 series) (Fig. 3).

Estimated source δ18Omarine values of ECA/NWA whales ranged from − 3.44‰ to + 4.53‰ (Table 2). The estimated 95% CI of most whales (n = 13) corresponded to high-latitude 18O-depleted waters characteristic of coastal ECA, NL, and Greenland. Several killer whales (n = 5), however, had inferred distributions that largely excluded high-latitude coastal areas and encompassed a broader band of the North Atlantic, including the relatively 18O-enriched waters south of the Gulf Stream (Fig. 4). The 95% CI around predicted source δ18Omarine values of whale NWA-BP-1998 completely exceeded predicted isoscape values (Fig. 4). Estimated source δ18Omarine values of the Greenland and Denmark killer whales (Table 2) also corresponded to high-latitude 18O-depleted waters (Fig. 4).

The range of dentine δ18OSC values within the subset of nine teeth averaged 1.36‰, and was lowest in whale NWA-BB-2002 (0.20‰—although this whale’s profile comprised only two data points) and highest in whale NWA-SC-1975-1 (2.61‰) (Fig. 5). Within-tooth δ18OSC variation was generally greater in NWA than ECA killer whales (Fig. 5). δ18OSC variation was typically accompanied by concurrent, often inverse, variation in Suess-adjusted δ13CSC (Fig. 5).

## Discussion

Distribution patterns have not been fully described for even the best-studied killer whale populations (e.g., ref.58), as observational effort has been limited primarily to nearshore areas when killer whales are seasonally abundant (e.g., ref.59), and photo-ID catalogues have not been developed for most regions. Our assessment of killer whale distributions using dentine oxygen isotopes integrated over long timeframes has revealed patterns in the species’ distribution in the North Atlantic that have been suggested, but not resolved, by previous sightings compilations across the region2,3,4,6,14,60. The ranges of most whales were restricted to high-latitude 18O-depleted waters; however, the broader ranges of a smaller number of whales encompassing lower latitude 18O-enriched waters in the North Atlantic indicates a considerable degree of spatial structure among killer whales in the ECA and NWA that has not been reported previously.

Isoscapes have been used routinely to infer distributions and movements of terrestrial animals18, but their use in marine systems has been hindered by limited baseline measurements over appropriate spatiotemporal scales20,61. LeGrande and Schmidt’s23 predicted gridded marine δ18O isoscape was constructed using a 50-year dataset combined with modeling of 18O-salinity relationships, and the spatiotemporal coverage of samples used to inform and validate their model is considered to be relatively good in the Arctic and North Atlantic Oceans. However, there are areas of sparse coverage and the authors caution that their use of multi-seasonal, multi-decadal data would obscure variation in Arctic surface water δ18O, which is temporally and spatially more variable than ice-free waters due to seasonal cycles of ice formation and melt62,63,64. Discrepancy between estimated source δ18Omarine values and modeled isoscape values could also reflect the fact that only variance in predicted source δ18Omarine values was incorporated in our assignment of distributions (via 95% CIs), while variance in predicted isoscape values was not. The much greater discrepancy observed for whale NWA-BP-1998, for which only the lower limits of the 99% CI of its predicted source δ18Omarine corresponded to isoscape values in lower latitudes of the North Atlantic (not shown), could reflect extremely sparse sample coverage informing modeled values in that region23.

The most obvious pattern in our data is the spread in δ18OP values and estimated δ18Omarine values, which encompass both sides of the sharp boundary in marine δ18O values demarcated by the Gulf Stream23. The inferred high-latitude distributions of most killer whales above this boundary is consistent with global patterns that show killer whale density increases by 1–2 orders of magnitude from tropical to polar regions1, including in the NWA, where comparatively few killer whale records exist from the eastern seaboard of the U.S.A. south to the equator4. However, inferred distributions of several killer whales that stranded in both the ECA and NWA encompassed relatively 18O-enriched, lower latitude waters across a broader area of the North Atlantic. We can rule out passive transport of carcasses from distant locations via currents as a source of variation for these five whales, which either live-stranded (ECA-CS-1977, NWA-CB-1971-1 and 2, and NWA-BP-1998) or were found in inland Arctic waters in minimal state of decomposition (ECA-RB-2009). Whales observed seasonally in the ECA have been hypothesized to range possibly as far south as the Caribbean during winter2, and recent studies have demonstrated10 or inferred17 movements of ECA killer whales into temperate and perhaps tropical waters. Killer whales are observed year-round in low numbers throughout the Caribbean4,65, and large numbers of killer whales were observed by whalers in the 1800s in a widespread offshore area known to them as the Western Ground, which stretched from Bermuda to the Azores14. Limited data preclude interpretations of killer whale residence patterns in either region4,65, although Katona et al.4 suggested there could be resident and migratory populations stretching from the Bay of Fundy to the equator.

The largely similar inferred distributions of most of the killer whales in the high-latitude waters surrounding the ECA, NL, and Greenland could reflect a continuous distribution across the region, as has been hypothesized2. However, the ~ 2‰ range of individual δ18OP values among these whales exceeds the degree of variability expected for cetacean populations with similar distributions within waters with homogenous δ18O. Clementz and Koch66 reported δ18O ranges of ~ 0.5 to 1‰ in tooth enamel of cetacean populations from the isotopically homogenous Californian coast. For comparison with other North Atlantic cetaceans, bone carbonate δ18O differed by just 0.4‰ between fin whales (Balaenoptera physalus) from western Iceland and northwestern Spain31, and dentine carbonate δ18O differed by a similar amount between sperm whales (Physeter macrocephalus) from northwestern Spain and Denmark30. While groupings based on stranding location did not differ significantly, spatial differences among individuals inferred from δ18OP are largely consistent with previous bulk and amino acid-specific δ15N analyses that suggested some degree of spatial separation among these same killer whales51.

Geographic assignments can be refined by using multiple isotopes to narrow potential distributions to areas where tissue values correspond with intersecting isoscape values67,68. Carbonate in biogenic apatite precipitates from blood bicarbonate and δ13CSC thus reflects the bulk isotopic composition of respired carbohydrates, lipids and proteins49,69,70. Despite advances in modelling spatial variation in DIC and phytoplankton δ13C across the North Atlantic20, calibration of killer whale dentine δ13CSC values to the predicted δ13C isoscape would therefore require knowledge not only of trophic position, but also the fraction of carbon from dietary lipids and proteins routed to δ13CSC (see ref.66). However, the ~ 4‰ range in δ13CSC among killer whales exceeds that expected from trophic factors alone, particularly since most of the killer whales in our sample fed at a similar tropic position51,52. With more detailed dietary information to allow for partitioning of trophic and spatial influences on δ13CSC, relative δ13CSC differences among killer whales with similar δ18OP values could offer additional discriminatory power, particularly in areas of uniformly 18O-depleted coastal, high latitude waters.

Myrick et al.71 demonstrated a constant rate of dentine deposition across all months in a captive killer whale, which we assume is also true of wild killer whales. However, dentine deposition rates vary seasonally in many odontocete species in response to seasonal food availability or environmental variation, with more material typically laid down from spring to fall72,73. If this deposition pattern occurs in wild killer whales, then our analysis of whole-tooth samples and subsequent inferences about distributions would be biased towards summer months. Bone, however, is slowly and continually remodelled34; as such, bone δ18OP of the five whales from Greenland and Denmark reflect a long-term average distribution within high-latitude, 18O-depleted waters. That being said, temporally crude whole-tooth and bone samples would obscure any seasonal movements10 or interannual variation in distribution. Within-tooth variation in δ18OSC on the order of 1–2‰ suggests interannual variation in ECA and NWA killer whale distributions. By comparison, dentine carbonate δ18O declines of ~ 2‰ across male North Atlantic sperm whale teeth were linked possibly to large-scale latitudinal migrations at sexual maturity30. Large-scale movements of killer whales have now been documented globally, including the North Atlantic10. However, their ranging patterns remain poorly understood, and fine-scale spatial resolution (i.e., µm) afforded by in situ microbeam sampling techniques74 could help resolve seasonal and interannual variation in movements and distribution.

Spatial structure inferred here from oxygen isotopes is supported, at least generally, by previous studies indicating genetic differentiation among ECA and NWA killer whales. Foote et al.75, using samples not included in this study, showed NWA whales formed a different clade than other North Atlantic killer whales. In a more recent analysis that sequenced whole genomes, Lefort76 differentiated two clusters among ECA and NWA killer whales: one that comprised animals from northern Baffin Island and Newfoundland, and another that comprised animals from Hudson Bay and Baffin Island. There was also high covariance among killer whales from the low Arctic (Hudson Bay) and those from the eastern North Atlantic (Greenland, Iceland, and Norway)76, which is consistent with isotope clusters that comprised animals from the ECA and Greenland and Denmark. We note, however, that the Hudson Bay whale included in Lefort76, ECA-RB-2009, was relatively distant from the five Greenland and Denmark specimens based on δ18OP values. Of the four animals included in both studies, low genetic variation was observed between whales ECA-AB-1948 and ECA-RB-2009 and between whales NWA-CB-1971-2 and NWA-BB-200276. Neither of these pairings, however, clustered tightly together based on δ18OP (Fig. 3).

Oxygen isotope data are consistent with purported killer whale ecotypes in the ECA and NWA52. Two of the killer whales with inferred broader, lower latitude distributions (ECA-RB-2009 and NWA-BP-1998) have amino acid-specific δ15N patterns indicative of dietary differences from the other whales, coupled with morphological differences in tooth wear51,52. These whales appear to be ecologically and morphologically similar to the North Atlantic Type 1 killer whales that range from NL to Norway75,77, and morphologically similar to a single killer whale specimen from the Caribbean65. We assume that distribution, and not dietary, differences are the primary driver of observed δ18OP variation among these whales, as ingestion of water via food and metabolic water formed via oxidation of food are thought to be relatively minor oxygen fluxes compared to diffusion of water across cetacean skin24,25. This assumption is supported by similar dentine δ18OP values of fish-eating and mammal-eating killer whale ecotypes from the eastern North Pacific (16.75 vs. 16.87‰, respectively)22. It is also supported more generally by relatively small deviations (< 1‰) of dentine/bone δ18O of whale species with varied diets from linear relationships with environmental δ18O21,22. Vertical variation in δ18Omarine could potentially confound our interpretations, if, for example, higher killer whale oxygen isotope compositions were acquired while spending significant amounts of time foraging in deep-water on species such as Greenland shark (Somniosus microcephalus). Greenland sharks from northern Baffin Island have relatively 18O-enriched tooth phosphate that corresponds to δ18Omarine at depths exceeding 300 to 500 m78. While this depth is within the dive range of killer whales79, the ~ 6‰ range in δ18OP separating the sampled killer whales (Fig. 2) exceeds the ~ 1‰ vertical increase in δ18Omarine from 0–250 m (~ − 1.5‰) to > 250 m depths (~ − 0.5‰) in Baffin Bay62.

The oxygen isotope compositions of groups of killer whales that stranded together also allow for some inference on the nature of their associations. Similar δ18OP values among individuals of three of the four groups of animals that stranded together indicate they likely comprised stable, long-term associations with similar distribution and movement history, consistent with research indicating offspring generally do not disperse from killer whale matrilines80. However, the close to 4‰ spread in δ18OP among the group of three killer whales that stranded together in Cumberland Sound in 1977 suggests they were not part of a cohesive group. The different ages (and thus sizes) of these whales does not appear to be a factor. While there is some suggestion of a decline in δ18OP with age in some of the older NWA whales, no such pattern was observed for the ECA whales (Fig. 5). Moreover, the group of three killer whales that stranded together in southeast Hudson Bay (ECA-SQ series) had almost identical δ18OP despite having a similar age range (Tables 1, 2). The three killer whales from Cumberland Sound were sampled from a group of 14 animals that became entrapped in a saltwater lake while hunting belugas2. Temporary associations between small, stable groups of killer whales have been observed in other locations where prey resources are highly localized81, a possibility in Cumberland Sound where belugas are seasonally aggregated in a relatively small area.

The high-latitude distributions inferred for most of the sampled whales is relevant for understanding potential killer whale range expansions and/or population increases with reductions in sea ice extent and duration. However, with half of the ECA specimens and most of the NWA specimens in our sample coming from the 1970s or earlier, much of our sample does not encompass the approximately five-decade period over which sightings have increased dramatically in both the ECA and off NL5,6,11,82. Distribution shifts in response to changing environmental variables such as declining sea ice extent and duration can be rapid, as exhibited across various taxa in Arctic regions83. We therefore recommend further study of factors that drive current killer whale distribution and population structure throughout the North Atlantic, using a range of complementary approaches such as long-term, coarse-scale data reported here versus short-term high-resolution telemetry data8,9,10.