The geographic range of the Pacific walrus is confined to the shallow continental shelf of the Bering and Chukchi seas (Fig 1), with coastal haul-outs spanning many locations along the coasts of Russia and the United States (Alaska) [58]. Walrus samples in this study originated throughout their geographic range from several subsistence hunting communities. The villages of Gambell and Savoonga on St. Lawrence Island are positioned at the entrance of the Bering Strait, close to large wintering and breeding grounds in the NBS and along the passageway to the Chukchi Sea during the annual northern spring migration route [18,59]. Subsistence hunts occur here in the spring, typically around May of each year [60]. Walrus samples from the Chukchi Sea were provided from harvests that occurred near Peard Bay, Wainwright and Utqiaġvik while the animals were occupying their summer foraging grounds or in the autumn on their return south (Fig 1). The walruses harvested by hunters from the North Slope communities typically forage further offshore, on the shallow flanks of Hanna Shoal, but also near the coast when sea ice is low [4,11,12,22]. Point Lay is the location of large coastal haul-outs in the North American portion of the Chukchi Sea, particularly during low sea ice years [11,12]. Walrus liver tissues from Point Lay were collected as part of an earlier study, rather than subsistence hunting activities [61].

We used the Distributed Biological Observatory (DBO) sampling program to provide ecological context for the walrus tissue sampling. The DBO regions include five areas of high benthic biomass in the northern Bering and Chukchi sea, which in turn support higher trophic positions [9]. Several of these DBO regions are hotspots for walrus foraging activities owing to the high benthic biomass [4]. Our study focuses on DBO regions 1 and 4 (Fig 1). DBO 1 is located south of the St. Lawrence Island polynya, which is a key area for Pacific walrus wintering and a prominent breeding ground [18]. DBO 4 is located on the southeastern flanks of Hanna Shoal, which as described above is a shallow region on the northeast Chukchi shelf with substantial bivalve populations and the location of a prominent summer foraging ground for walruses [4,11,18].

Walrus liver tissues were collected and donated from subsistence hunting activities in the NBS during the start of the spring migration (April/May) in 2012, 2014 and 2016 (Table 1). These harvests occurred from the villages of Gambell and Savoonga on St. Lawrence Island and from Diomede Village (Inalik) on Little Diomede Island (Fig 1). The NBS samples were stored frozen at -80°C at the University of Alaska’s Museum of the North (UAM). UAM provided 0.5 g tissue plugs, shipped on dry ice, which were freeze dried immediately upon arrival. The collection date, harvest location, sex, and age class of the walruses were documented in the UAM Mammal Collection database (Arctos; https://arctos.database.museum/home.cfm). Age class data were listed as unknown for all walruses sampled in 2012.

Walruses harvested from the summer (June/July) foraging grounds in the Chukchi Sea were collected by subsistence hunters from Peard Bay, Wainwright, and Utqiaġvik in 2012 and 2014 (Fig 1). Additional liver samples from the Chukchi Sea were collected near Point Lay, which were necropsied as part of a large haul-out mortality investigation during October 2014. This sampling effort was jointly conducted by the community of Point Lay (Warren Harding and Leo Ferriera), the United States Fish and Wildlife Survey (Joel Garlich Miller and Jonathan Snyder), the North Slope Borough Department of Wildlife Management (NSB DWM; Raphaela Stimmelmayr and Isaac Levitt), and the Alaska Department of Fish and Game (Lori Quakenbush and Anna Bryan). These walruses were considered recently deceased (<1 week) based on external appearance and the state of internal organ preservation. The ambient temperature during the field investigation was cool (i.e. below and around freezing) with snow on the ground. These conditions resulted in carcasses that were naturally chilled, limiting organ decomposition. Therefore, degradation of HBIs and stable isotopes of nitrogen were concluded to be negligible and appropriate for this analysis [62,63]. The Chukchi Sea walrus samples were stored at -80°C by the NSB DWM (Table 1). Frozen walrus livers were shipped to the Chesapeake Biological Laboratory where 0.5–1 g of tissue was extracted using sterilized stainless steel scalpels followed by freeze drying over 24 hours. The collection date, location, sex, and age class of the walruses were provided from NSB DWM records.

Annual sea ice persistence data were calculated using sea ice concentration data from the National Snow and Ice Data Center (NSIDC, https://nsidc.org) Defense Meteorological Satellite Program (DMSP) Special Sensor Microwave Imager/Sounder (SSMIS), using a 15% threshold to determine the presence vs. absence of sea ice cover, following methods of Frey et al. [64]. Persistence data were parsed annually from September to September for each year to encompass the full seasonal sea ice advance and retreat cycle (e.g. 15 September 2011 through 14 September 2012). Mean annual sea ice persistence values were calculated for DBO 1 and DBO 4 for each year (Table 1). Sea ice persistence anomalies were calculated relative to the 1981–2010 average (i.e., the average of all years between 15 September 1980/14 September 1981 through 15 September 2009/14 September 2010) to assess broader spatio-temporal changes in sea ice by year. Anomalies were not reported for 2016 owing to the lack of associated walrus data from the Chukchi Sea.

Monthly averaged sea ice concentrations were retrieved at a 12.5-km resolution using the DMSP SSMIS passive microwave data (S1 Table). We calculated the mean monthly sea ice concentrations within the boundaries of DBO 1 or 4 associated with the time and location of the walrus harvest and their likely foraging grounds two weeks prior to determine a sea ice index (S1 Table). For example, a walrus harvested from St. Lawrence Island in early May was assumed to be foraging at DBO 1 in late April. DBO 1 was assigned as the foraging area for the NBS walruses harvested in the spring months and DBO 4 for the walruses harvested from the Chukchi Sea in the summer and fall months. However, we note that these assignments may not always accurately represent walruses from more southeastern foraging aggregations in the St. Lawrence Island/Diomede harvest or walruses recently migrating north in the early summer (June) harvests from Utqiaġvik/Wainwright [11]. To account for the potential influence of variable distributions, and to meet the assumptions of independent data, we calculated the mean of the sea ice organic carbon values associated with sea ice concentrations or sea ice persistence for additional analyses.

After freeze drying, HBIs were extracted following established methods [45,65]. Samples were saponified in a methanolic KOH solution and heated at 70°C for one hour. Hexane (4 mL) was added to the saponified solution, vortexed, and centrifuged for three minutes at 2500 RPM, three times. The supernatant with the non-saponifiable lipids (NSLs) was transferred to clean glass vials and dried under a gentle N₂ stream. The initial extracts were re-suspended in hexane and fractionated using open column silica gel chromatography. The non-polar lipids containing the HBIs were eluted while the polar compounds were retained on the column. The eluted compounds were dried under N₂. 50 μL of hexane was added twice to the dried purified extract and transferred to amber chromatography vials.

The extracts were analyzed using an Agilent 7890A gas chromatograph (GC) coupled with a 5975 series mass selective detector (MSD) using an Agilent HP-5ms column (30 m x 0.25 mm x 0.25 μm), following established methods [65]. The oven temperature was programmed to ramp up from 40°C to 300°C at 10°C/minute with a 10-minute isothermal period at 300°C. HBIs were identified using selective ion monitoring (SIM) techniques. The SIM chromatograms were used to quantify the HBI abundances by peak integration with ChemStation software (Agilent Technologies, California, USA). A purified standard of known IP₂₅ concentration was used to confirm the mass spectra, retention time and retention index (RI). The HBIs were identified by their mass ions and RI including IP₂₅ (m/z 350.3), HBI II (m/z 348.3) and HBI III (m/z 346.3). A procedural blank was run every 9^th sample.

We used the H-Print index (Eq 1) to provide an estimate of the relative organic carbon contributions of phytoplankton to sea ice algae [45], using the abundances of IP₂₅, HBI II and HBI III: H‐Print%=HBIIIIIP25+HBIII+HBIIIIx100 (1)

Lower values are indicative of proportionally greater sympagic organic carbon and higher values are indicative of proportionally greater pelagic organic carbon. Sea ice organic carbon (iPOC), as a proportion of marine-origin carbon within samples, was estimated using a prior H-Print calibration determined from feeding experiments with known algal species (R² = 0.97, p <0.01, df = 23 [66]).

In contrast to the H-Print index, higher iPOC values reflect greater proportions of organic carbon derived from ice algae. We define iPOC values exceeding 50% as indication of elevated sea ice organic carbon in the animal’s diet.

Freeze-dried samples were homogenized with a mortar and pestle. 0.4–0.6 mg of tissue was weighed and introduced into tin capsules. The nitrogen stable isotope ratios were measured at the stable isotope facilities at the Chesapeake Biological Laboratory using a Costech elemental analyzer coupled to a ThermoFisher Delta V Isotope Ratio Mass Spectrometer operated in a continuous flow mode. Nitrogen isotope ratios are expressed relative to atmospheric nitrogen (N₂) using the following equation: δX=[(RsampleRstandard)−1]x1000 (3) where R is the corresponding ratio of ¹⁵N/¹⁴N. Accuracy of the data was assured by simultaneous combustion and analysis of the samples with an internal standard of acetanilide, run every 9 samples, which has a well-characterized isotopic composition. Precision was also assessed by analysis of this internal standard. Samples were analyzed on two separate dates and the precision of the acetanilide standard run simultaneously was ±0.21‰ on the first date (n = 5) and ±0.08‰ on the second date (n = 6). The mean δ¹⁵N value determined for acetanilide for both dates were within 0.12‰ of each other.

All analyses were conducted using R version 3.6.1 (www.r-project.org). Data were normally distributed with equal variances. A significance level of α = 0.05 was used for all tests. We explored the regional influence of sympagic and pelagic HBIs graphically through non-metric multidimensional scaling (NMDS) ordination plots using the R package ‘vegan’ [67] to look for distinctions in carbon sources associated with the walrus harvest location and associated sea ice conditions. Distances were based on Bray-Curtis dissimilarities of the three measured HBIs (IP₂₅, HBI II and HBI III). We assigned sea ice organic carbon (iPOC), H-Print, and sea ice persistence as the environmental vectors (noting that iPOC and H-Print are calculated from the HBIs and the direction of the vectors suggest stronger associations with either ice algae or phytoplankton, respectively). We conducted a two-way ANOVA test with sex and region as factors. Both factors and the interaction of the two were significant (p < 0.01). To account for this interaction, the means were adjusted by grouping sexes among each year and comparing iPOC values by region using the R package ‘lsmeans’, which were reported with standard error (SE). We then carried out planned comparisons using the Bonferroni correction for multiple comparisons to detect differences in mean iPOC values between regions, sex and years. Given the limited age class data associated with the walruses harvested in the NBS, the effect of age class on sea ice organic carbon estimates could only be adequately examined for the Chukchi Sea. The age class analysis was therefore conducted using a Welch two sample t-test to examine the difference between adults and calves.

Trophic positions were estimated with a Bayesian model approach from the nitrogen stable isotope values using the R package ‘tRophicPosition’, which uses Markov Chain Monte Carlo simulations [68]. This Bayesian approach uses relevant statistical distributions around the consumer and baseline observations to allow these values to be treated as random variables. We used the trophic enrichment factor (TEF) estimate for each trophic position that is based on Post’s widely utilized meta-analysis (3.4 ± 0.98 [69]). While there are also species- and tissue-specific TEFs to consider [70–72], the modelling approach incorporates uncertainty in the TEFs around our estimation of relative trophic position (hereafter referred to as ‘trophic position’ or ‘TP’). We selected one baseline for the model, which links the δ¹⁵N enrichment per trophic level with the trophic positions of the baseline (in this case, a primary consumer) [68]. For the baseline, we utilized previously published δ¹⁵N values for Macoma calcarea collected from Hanna Shoal in 2012 [73] and from DBO 1 and DBO 4 in 2015 [74] to serve as our baseline consumer in the model (S2 Table). These baseline values were temporally and spatially consistent with the Pacific walrus harvests in this study. The M. calcarea values were comparable between the two studies (δ¹⁵N = 10.1 ± 0.9, n = 36 [73]; δ¹⁵N_{DBO 1} = 9.6 ± 0.5 and δ¹⁵N_{DBO 4} 9.3 ± 0.8 [74]) and allowed us to determine a reasonable estimate of the variability. The median trophic position estimates from the model were reported along with the 95% confidence levels. Pairwise comparisons of posterior trophic positions estimates were assessed to determine significant differences between regions and sexes. This model allows for the input of δ¹³C values for running mixing models (particularly when using two baselines) to generate carbon versus nitrogen plots to examine basal food sources and consumers spatially. The δ¹³C values were excluded from our study owing to the influence of lipids in our samples and inability to correct for their presence [75]. However, for the purpose of exploring associations sea ice organic carbon contributions, iPOC values were substituted for δ¹³C in the model, which is not used in the calculation of the TP estimates with one baseline and does not impact these values (S1 Table). To do this, we used mean iPOC estimates for M. calcarea from a prior investigation of HBIs in benthic invertebrates from DBO 1 (6.2 ± 5.4%, n = 4) and DBO 4 (35.6 ± 6.8, n = 7) [42]. Based on the results of that study, we assumed that there would be comparable iPOC values from year to year based on similar annual distributions of HBIs in the surface sediments and similar HBI values between M. calcarea and surface sediments. We also employed a one-way ANOVA test to explore the relationship between sex, region and nitrogen stable isotope values.

Sea ice persistence progressively declined in the NBS throughout the study period from 170 days/year in 2012, to 126 days/year in 2014, and 109 days/year in 2016; an overall decline of 61 days (36%) (Fig 2A). In contrast, sea ice persistence was similar between 2012 (269 days/year) and 2014 (261 days/year) in the Chukchi Sea (Fig 2A). Sea ice persistence anomalies were calculated to look at broad regional trends in 2012 and 2014, revealing a divergent pattern throughout the Pacific Arctic in 2012 (Fig 2B). Sea ice persistence at DBO 1 (NBS) was approximately 21 days above the 1981–2010 average, while sea ice persistence was approximately 34 days below average at DBO 4 (Chukchi Sea; Fig 2B). In contrast, sea ice persistence was uniformly below average at DBO 1 (-21 days) and DBO 4 (-31 days) in 2014 (Fig 2C).

Assigned monthly mean sea ice concentration values ranged from 0–98% overall (S2 Table). In 2012, walruses harvested in the NBS in May were associated with a 98% (April) monthly mean sea ice concentration. Walruses harvested in July from the Chukchi Sea were associated with a 72% (June) monthly mean sea ice concentration. For 2014 in the Chukchi Sea, mean sea ice concentrations associated with the walrus harvest dates were 0% (September) for walruses sampled in October and 92% (June) for walruses harvested in July. For the NBS in 2014, walruses harvested in early May were associated with a 74% monthly mean sea ice concentration in April and dropped to 10% in May for walruses harvested in late May. During April of 2016, the month prior to walrus harvests in the NBS, the mean sea ice concentration was 48%. There were no significant relationships between either satellite-derived sea ice parameter and sea ice organic carbon based on linear regressions (Fig 3). As stated previously, linear regressions were based on the mean iPOC values associated with specific sea ice concentration or sea ice persistence values, rather than the individual data points.

The NMDS ordination plot indicated there were distinct groupings in sea ice organic carbon content between walruses in the NBS and Chukchi Seas (Fig 4A). The environmental vectors (iPOC, H-Print and sea ice persistence) indicate a stronger association with ice algae contributions (i.e. iPOC) and sea ice persistence on the Chukchi Sea walruses while there were greater associations with phytoplankton contributions (i.e. H-Print) on the NBS walruses. Two-way ANOVA tests indicated that region was a significant factor (DF = 1, F = 8.32, p<0.01), as well as sex (DF = 1, F = 10.51, p<0.01). The adjusted iPOC values (mean± SE) in livers were highest overall in the Chukchi Sea with mean female values of 46 ± 6% and male values 58 ± 6%. In the NBS, mean iPOC values for females were 45 ± 3% and 30 ± 3% for males (Fig 4B, Table 1).

Pairwise planned comparisons confirmed sea ice organic carbon values were significantly higher in NBS females when compared to NBS males for all 3 years of this study (p < 0.001; Fig 5). In contrast, sea ice organic carbon values were similar (p>0.05) between males and females in the Chukchi Sea in both 2012 [35 ± 6% (males) and 36 ± 8% (females); Table 1] and 2014 [71 ± 7% (males) and 63± 9% (females); Table 1]. However, mean iPOC values in walruses were significantly different between 2012 (36 ± 6%) and 2014 (68 ± 6; p < 0.001) (Table 1) in the Chukchi Sea, while values were comparable between 2012, 2014 and 2016 in the NBS. In 2014, the mean iPOC values were significantly different between the Chukchi Sea (68 ± 6%) and NBS (44 ± 3%), but were comparable in 2012 (36 ± 6% and 35 ± 3%, respectively) (Table 1). When factoring in both sex and region, the differences were only observed among the males and not the females. In 2012, Chukchi Sea males (35 ± 6%) had higher values than NBS males (17 ± 3%; p < 0.05; Table 1). In 2014, Chukchi Sea males (71 ± 7%) were again significantly greater than NBS males (40 ± 3%, p < 0.001).

A Welch two-sample t-test indicated the difference in iPOC values between calves and adults in the Chukchi Sea was significant (t₈ = -2.8, p<0.05). Sea ice organic carbon values in calves was 65% ± 25% (mean ± SD, n = 7) versus 42% ± 11% (n = 9) in adults (Fig 6). Six of the seven walrus calves had liver iPOC values that exceeded 50%, the threshold that suggests proportionally greater contributions of sea ice algae at the base of their diet.

Overall, δ¹⁵N values ranged from 11.9‰ to 16.9‰. In the NBS, δ¹⁵N values for females were 13.3 ± 0.7‰ in 2012, 13.6 ± 0.6‰ in 2014 and 14.3 ± 0.9‰ in 2016 (mean ± SD, Table 1). For males, δ¹⁵N values were 13.2 ± 1.2‰ in 2012, 12.8 ± 0.4‰ in 2014 and 12.8 ± 0.5‰ in 2016 (Table 1). In the Chukchi Sea, mean δ¹⁵N values for females were 15.0 ± 0.9‰ in 2012 and 14.7 ± 1.1‰ in 2014 (Table 1). For males, δ¹⁵N values were 14.1 ± 0.8‰ in 2012 and 15.5 ± 1.2‰ in 2014 (Table 1). Sex and region were both significant (p < 0.001) factors for δ¹⁵N values as determined by a two-way ANOVA test. The interaction of the two factors was not significant (p > 0.05). The pairwise comparisons show each group was significantly different, with the exception of Chukchi males versus females.

The TP estimates were similar between regions and sexes. Overall, TPs were minimally but statistically higher in the Chukchi Sea at 3.5 compared to 3.2 in the NBS (Table 2). In the NBS, TPs were similar for females and males (3.2 and 3.1, respectively; Table 2). Median TP estimates were the same for males and females in the Chukchi Sea (3.5; Table 2). Posterior pairwise comparisons indicated males in the NBS and Chukchi Sea were significantly different (p<0.05).

The higher contribution of sea ice organic carbon in the Pacific walrus tissues from the Chukchi Sea compared to the NBS (Figs 4 and 5) are consistent with sea ice persistence and HBI distributions previously determined from surface sediments [76]. Those HBI measurements were assumed to reflect differences in ice algae production and deposition throughout the Pacific Arctic region [76], largely driven by sea ice persistence, where sea ice is typically present 2–4 months longer in the Chukchi Sea [64]. A recent study of particle fluxes in the northeast Chukchi Sea identified year-round export of diatoms (including ice-associated diatoms) on the shallow shelf, which suggested ice algae could be a sustaining resource for benthic communities in the less productive months [77]. Additionally, subsurface chlorophyll measurements alongside disassociated ice algae also suggest that ice algae are likely contributing to prolonged periods of productivity on the Chukchi shelf [78]. These data imply that sympagic HBI levels in the northeast Chukchi Sea would be higher than those found in the NBS, and therefore the sea ice organic carbon would be higher in the prey items utilized by northerly walrus populations in the summer and autumn. The sea ice organic carbon values in Pacific walrus livers presented here likely reflect the broader HBI distributions observed in the environment and the location of primary foraging activities rather than differences in resource selection.

As stated previously, Pacific walruses are central place foragers that primarily consume bivalves [18], which often dominate benthic macrofaunal biomass across the region [4]. The sea ice organic carbon values in benthic invertebrates across the Pacific Arctic, specifically in Tellinid bivalves [42], are consistent with this hypothesis. These bivalves are ubiquitous throughout the entire region and generally matched the HBI distribution patterns in the surrounding surface sediments, with elevated sea ice organic carbon relative to the surface sediments in a few locations [42]. Tellinid biomass has proven to be a significant predictor of Pacific walrus resource selection models in both the northern Bering [26] and Chukchi [32] seas and therefore M. calcarea is a useful representative species for tracking sea ice organic carbon contributions to the Pacific walrus diet. While iPOC data in Tellinid clams are currently unavailable to correspond with the timing of the walrus harvests and samples used in this study, mean iPOC values from both regions in 2018 suggest a significant difference in their sea ice organic carbon content (~6% in the Bering Sea and ~40% in the Chukchi Sea) [42]. Given the similarities between males and females in the Chukchi Sea walrus samples, the prey HBI data suggest these walruses likely have similar foraging behaviors. The study assessing sea ice biomarkers in benthic fauna also revealed elevated sea ice organic carbon use by another common walrus prey item, sipunculid worms [42], which are also more prevalent in the Chukchi Sea than in the NBS [79]. If the availability of certain preferred prey items in the Chukchi Sea declines in the future because of this apparent dependence on ice algae, there will likely be consequences for Pacific walruses [79]. However, it remains to be seen how benthic organisms respond to shifting carbon sources.

We found no significant relationship between the two sea ice parameters tested (monthly mean sea ice concentration and sea ice persistence) with iPOC values in Pacific walrus tissue. However, this may be unsurprising owing to the fact that this species migrates across a gradient of sub-Arctic to Arctic habitat with distinct sea ice conditions, while also coupling the benthic and pelagic oceanic realms along the way. These large-scale movements likely result in both horizontal and vertical habitat integration. This integration across the entire Pacific Arctic region over the course of a few months during the ice melt season is likely why our sea ice index did not robustly match iPOC values in the walrus liver tissues. The elevated sea ice organic carbon values measured in the Chukchi Sea walruses collected during open water periods (October) exemplify the nuances of analyzing associations with satellite-derived sea ice data. These walruses had a longer period of time to forage in the Chukchi Sea, consuming more of the prey that had previously assimilated sea ice organic carbon than those harvested earlier in the summer as the ice edge retreated. We conclude that direct associations between these biomarkers and sea ice data (e.g. concentration, persistence, or break-up date) in migratory consumers in this region may not always be possible and require further interpretation and perhaps paired with additional trophic markers (e.g. fatty acids, compound-specific stable isotopes).

Despite the lack of association between the sea ice parameters and sea ice organic carbon, there were interesting patterns within a broader context when considering the sea ice persistence anomalies from 2012 and 2014. For 2012, the NBS was a cold year in which sea ice persistence was well above average (Fig 2B). However, there was also a geographically divergent pattern in 2012, as sea ice in the Chukchi Sea was a record low year [64] (Fig 2B). This geographically divergent sea ice pattern would potentially explain higher than average sea ice algae trophic markers in the Bering Sea in 2012, potentially reflected in comparable values for the two regions (Fig 5B). However, in 2014, sea ice persistence anomalies indicate similarly negative values throughout the region (Fig 2B). Therefore, the default HBI distribution pattern observed for the region (increasing sympagic HBIs from south to north [76]) was likely reflected in consumer tissues. This broad assessment supports the use of HBIs as fingerprints of where large marine predators were foraging, similar to the way that carbon isotope distributions have been used to create isoscapes that map differences in the intensity of primary production throughout a given area [80]. Additionally, tracking sea ice organic carbon in these organisms over time may still be useful in monitoring responses to declining sea ice. HBI measurements of other marine mammals and seabirds in this region with differing foraging ecologies will be potentially useful in further assessing changes in food webs as seasonal sea ice declines.

The effects of declining sea ice on Pacific walruses may already be evident in the Bering Sea. Observations from the southwestern Bering Sea indicate that occupation of land haul-outs by males in this region may be shifting to northward locations in the Chukchi Sea on the Chukotka Peninsula [81]. There have also been more mixed-sex herds observed in the Chukchi Sea [25]. Coincident shifts with their prey base are also occurring as indicated by a northward shift in bivalve populations at DBO 1 [28], which is near a prominent walrus breeding ground and a prey base for these walruses [26]. In addition to declining sea ice, the northward shift in bivalve biomass will also likely influence the Pacific walrus haul-out locations to move north. Negative associations with sea ice declines have also been reported in body condition observations for females and juveniles of other pinniped species in this region [82]. These observations together suggest a broader ecosystem shift is underway in response to warming waters, reduced sea ice coverage and timing of ice retreat, and changes in the quality, quantity and timing of primary production [5,29].

In a recent similar study using HBIs in Atlantic walruses (O. r. rosmarus) in the Canadian Arctic, coupled with measurements of δ¹³C_phe (phenylalanine–a source amino acid), suggested nearly exclusive sea ice-derived carbon as the base of the walrus diet at high-latitudes (89–98% iPOC) [83]. These values were significantly higher than Atlantic walruses examined from lower latitude locations, suggesting differences in carbon energy sources based on location and changing ice algal phenology at lower latitudes. This study also observed a 75% decrease in sea ice organic carbon over two decades in the lower latitude walruses that was strongly associated with declining sea ice, which was not observed with the similar decline in sea ice at high latitudes. While these two subspecies differ in their migratory behavior (Pacific walruses follow the ice-edge and Atlantic walruses do not), their findings regarding regional distinctions in sea ice-derived carbon are similar to what we observed in this study. However, the iPOC values in the Atlantic walruses were notably higher than most of the values observed in the Pacific walruses of this study, which is consistent with the higher persistence of sea ice in the Canadian Arctic. The differences in migratory behavior are also probably why associations between the Atlantic walrus iPOC values and satellite-based sea ice observations were stronger. Additionally, they concluded that shifts in ice algal phenology are likely leading to a decoupling of sympagic-benthic systems and impacting benthic consumers, which is also a likely explanation for the changes in dominant benthic organisms (biomass and abundance) and northward migrations underway in the northern Bering Sea ecosystem [28,29].

The contribution of sea ice derived organic carbon was higher in females than in males in the NBS, whereas there was no difference between sexes in the Chukchi Sea (Fig 4). This can be explained by several factors. First, the foraging behavior of walruses may be different between males and females in winter and spring when walruses are in the NBS, but similar in summer when they are in the Chukchi Sea. Early studies of walrus stomachs suggested that females and juveniles tended towards smaller bivalves compared to that of males [18]. Reported sea ice organic carbon values among various feeding strategies of benthic invertebrates throughout the Pacific walrus range suggested elevated ice algae consumption by subsurface deposit feeders and elevated phytoplankton consumption by suspension feeders and predator/scavengers [42]. If female walruses spend greater efforts foraging on infaunal organisms deep in the sediments rather than on motile epifauna, including predatory gastropods or crustaceans, this may be one explanation for differences in their sea ice organic carbon levels compared to males. Differences in male and female prey selection have been previously reported. For example, male Pacific walruses may opportunistically forage on higher trophic level prey, including seals and seabirds, if there is a lack of preferred prey owing to availability or requirements to forage in nearshore environments [18,84]. Indications of seals and other higher trophic level prey in walrus diets have been suggested through stable isotopes and is supported to some extent with observational data [85]. However, this behavior, and whether it differs between males and females, remains inconclusive but is believed to be atypical [18,84]. The similar trophic position estimates between males and females observed in this study do not indicate this distinct behavior occurs by one sex (Table 2). Owing to the primarily female migration during the summer and autumn seasons, general differences in diet by sex can be anticipated based on the prey availability in different foraging areas, particularly with males foraging from coastal haul-outs and females foraging from offshore sea ice. It had previously been concluded that males and females generally eat the same prey while occupying the same area [13,86,87].

Second, to support lactation and pregnancy, females may show selective foraging on more lipid-rich food sources and increase their foraging effort to support the energetic demands during lactation [36]. Females have been shown to double clam consumption during this time [86]. A higher consumption rate might result in a greater contribution of sea ice derived carbon in the diet of females compared to males. Resource partitioning may also be linked to breeding, which was implicated as the reason for differences in fatty acids measured in Pacific walrus blubber [88]. During the winter breeding season in the NBS, male walruses reduce foraging efforts to focus on mating [89]. The breeding season for the Pacific walrus occurs approximately from January through March [18]. The walrus livers from the NBS included in this study were collected in April and May (Table 1). Based on the assumed turnover rate of HBIs in liver, it is possible that the lower sea ice biomarker values are a reflection of this behavior.

The higher contribution of sea ice carbon in female walrus during the breeding period strongly suggest an elevated reliance of females to sea ice during this period compared to males. The results from Oxtoby et al. [88] suggested an incorporation of lipid-rich resources from the previous season, reflecting turnover rates of several months for blubber. By contrast, the use of liver in this study and a more rapid turnover rate further constrains this possible resource partitioning to the time period close to the breeding season and subsistence hunt. By reducing the time period of metabolic activity, we conclude the dependence on sea ice organic carbon is not derived from a prior year. With ongoing sea ice decline, females are likely to be at greater risk of disturbance than males [22,36,56,90]. The loss of sea ice could impact the females’ ability to successfully reproduce and provide adequate nutrition for their offspring [12,36,86]. Stress and reproductive biomarkers analyzed from recent and archaeological Pacific walrus bones suggested potential resilience to declining sea ice [91], but ongoing monitoring of this species is necessary to track their response to unprecedented change.

δ¹⁵N values of bulk tissue in predators is a useful tool to estimate trophic position but is influenced by changes in δ¹⁵N values at the base of the food web, or baseline. Here, we estimated the trophic position of Pacific walruses taking in account regional values of the δ¹⁵N at the baseline (i.e. primary consumers), allowing comparison between the two regions. Trophic position estimates were similar between regions and sexes, with median values ranging from 3.1 to 3.5, which were in alignment with expected values (~3.3; [83]). The model output indicated the difference was significant between the NBS (3.2) and Chukchi Sea (3.5) overall, but this difference is not particularly meaningful in terms of food web structure (Table 2). Based on previous studies of stomach contents, these TP estimates are in agreement that the summer and autumn diets of Pacific walruses in the Chukchi Sea are not substantially different than those in the NBS in the spring and winter [13]. The distribution and availability of bivalves, their preferred prey item, throughout the Pacific Arctic region supports this conclusion [4]. The trophic position estimates for Pacific walruses were also similar to those reported for Atlantic walruses (2.8–3.5), which also primarily feed on bivalves [83]. Based on comparable trophic positions between the Bering and Chukchi Sea, the observed elevation in sea ice organic carbon values in the Pacific walrus liver tissues from the Chukchi Sea reveals an increasing contribution, and perhaps significance, of sea ice algal production at the base of the food web on a latitudinal gradient. Our estimates do not provide evidence of higher trophic level prey consumption (i.e. seals), which has been reported to occur opportunistically with this species [84].

The δ¹⁵N values of bulk tissue of Pacific walrus liver tissues from this study agreed with previous liver tissue stable isotope measurements made from Pacific walruses harvested from St. Lawrence Island (13.0 ± 1.2 ‰) [84]. We found higher δ¹⁵N values in walruses from Chukchi Sea compared to the Bering Sea, that could be driven by differences at the base of the food web, as we found no regional differences in TP when taking in account the δ¹⁵N baseline. The higher δ¹⁵N values at the base of the Chukchi sea food web can be a result of several processes. The food web in the Chukchi Sea becomes more complex than the NBS as a result of the Pacific water inflow system, where advective processes deliver substantial carbon sources that are in part degraded and deposited as currents move across the shelf [92]. With these inputs, the detrital food web becomes more influential in the northern Chukchi, which could lead to higher δ¹⁵N values in the baseline. The hydrography of the Chukchi Sea around Hanna Shoal has been shown to influence the benthic food web, with increased benthic-pelagic coupling under nutrient rich Anadyr water [37]. As a result, benthic denitrification can impart an enriched δ¹⁵N signal in the water column, a process called “sedimentary isotope effect” [93,94]. This effect has been previously observed in the benthic food web in the Chukchi Sea [95]. The advection of degraded material inputs is also evident in the observed gradation of denitrification in the water column moving from south to north in this region [96].