Sampling was conducted during the RV “Polarstern” expedition PS81 (14 August–16 October 2013) in the Weddell-Scotia Confluence Zone (Figure 1, Table 1). Detailed information on the sampling during PS81 can be found in David et al. (2016).

The ice algae community was sampled by taking ice cores with a 9 cm interior diameter ice corer (Kovacs Enterprises) at two ice camps (n = 28) (Figure 1, Table 1). The bottom 10 cm of the ice cores were melted on board. Particulate Organic Matter (POM) from sea ice was concentrated by filtering the water from the melted ice cores via a vacuum pump through pre-combusted GF/F filters (1.0–1.5 L). The phytoplankton community (n = 5) was sampled at the chlorophyll a maximum depth (between 20 and 50 m) by a carousel water sampler connected to a CTD probe (Seabird SBE9+) with attached fluorometer (Table 1). Depending on the biomass concentration, POM from the water column was obtained by filtering between 3 and 21 L of water through pre-combusted GF/F filters. All filters were stored at –80°C until further processing.

Krill was caught in the uppermost 2 m from directly underneath the sea ice by a Surface and Under-Ice Trawl (SUIT) (van Franeker et al., 2009). Larval krill were staged based on the number of terminal spines on the telson (Kirkwood, 1982). All analyzed larvae were staged as furcilia VI krill. Krill which already lost one pair of post-lateral spines from their telson (Fraser, 1937), but had not yet developed external sexual characteristics, were classified as juveniles (Makorov and Denys, 1981). Krill bearing external sexual characteristics were categorized into females and males (Makorov and Denys, 1981), but grouped together as “adults” since no statistical differences in biomarker parameters were apparent. After determination of the body length and developmental stage (Table 2), samples were immediately frozen at –80°C in pre-combusted and pre-weighed sample vials (Wheaton, 6 h, 500°C). Analytical work took place at the Alfred Wegener Institute, Bremerhaven, Germany.

To determine source areas and pathways of sampled sea ice, the individual sample positions were tracked backward in time using a combination of low resolution ice drift and concentration products from satellites. Sea ice concentration data used in this study were obtained from the National Snow and Ice Data Center (NSIDC). The applied sea ice motion information was obtained from NSIDC (prior 2013) and from OSISAF (Ocean and Sea Ice Satellite Application Facility, Norway, from 2013). Sea ice motion information is available on a daily basis and provided on a 25 km (NSIDC, Fowler et al., 2013) and 62.5 km grid (OSISAF), respectively. The tracking algorithm works as follows: Using motion and concentration data, a specific ice area is tracked backwards until: (a) the ice reaches a position next to a coastline, or (b) the ice concentration at a specific location reaches a threshold value of <40% when ice parcels are considered lost. For a detailed description of the applied algorithm, we refer to Krumpen et al. (2016).

Prior to analysis, POM filters and whole krill individuals were freeze-dried for 24 h. Dry masses of krill were determined gravimetrically (Table 2). After homogenization, lipids were extracted with dichloromethane/methanol (2:1, v/v) (Folch et al., 1957). The lipid class composition of krill was investigated by High Performance Liquid Chromatography (HPLC) (Graeve and Janssen, 2009) in order to determine ontogenetic differences in the proportions of storage and membrane lipids. Extracted lipids were converted into fatty acid methyl esters (FAMEs) and free fatty alcohols by transesterification with methanol, containing 3% concentrated sulfuric acid. The proportional mass contributions of the diatom-associated marker FAs 16:1n-7 and 20:5n-3 (Graeve et al., 1994, 1997; Falk-Petersen et al., 1998; Volkman et al., 1998; Scott et al., 1999), the dinoflagellate-associated marker FAs 18:4n-3 and 22:6n-3 (Viso and Marty, 1993; Graeve et al., 1994; Phleger et al., 1998), and the copepod-associated marker FA 20:1n-9 and 22:1n-11 (Dahl et al., 2000; Virtue et al., 2016) were analyzed via gas chromatography. The percentage of each individual fatty acid of the total fatty acid mass was determined by an internal standard (23:0) added prior to lipid extraction. The total fatty acid content was determined gravimetrically by the internal standard (Table 2). 16:1n-7 and 18:4n-3 are mainly incorporated in storage lipids, and 20:5n-3 and 22:6n-3 mainly serve as membrane FAs. Storage FAs are assumed to reflect the more recent carbon source composition, whereas membrane FAs are more conserved (Stübing et al., 2003). To distinguish the short- and long-term importance of diatom- vs. dinoflagellate-derived carbon in the krill diet, the fatty acid ratios 16:1n-7/18:4n-3 and 20:5n-3/22:6n-3 were investigated. For details on sample preparation and measurements as well as analytical equipment see Kohlbach et al. (2016).

Length, dry weight, lipid content, and relative composition of storage and membrane lipids of krill used in BSIA and CSIA measurements were in the same range as those for lipid class and fatty acid composition (Table 2). Bulk nitrogen and carbon isotopic compositions of POM and krill were determined from freeze-dried bulk sample material (BSIA). Lipids were not removed prior to BSIA in order to avoid inducing changes in the isotopic composition of the krill samples (Mintenbeck et al., 2008). The FA-specific stable isotope composition of carbon was measured in FAME derivatives of the diatom-associated FAs 16:1n-7 and 20:5n-3, and the dinoflagellate-associated FAs 18:4n-3 and 22:6n-3 in POM and krill (CSIA; Kohlbach et al., 2016). During transesterification, a methyl group is added to the FA, which can result in a difference in δ¹³C values between the free FA and the corresponding FAME (e.g., Budge et al., 2011). In a preparatory experiment for this study, however, we did not find significant differences between the δ¹³C values of the free FA and the FAME (e.g., 16:0 FA: −28.56 ± 0.12‰, 16:0 FAME: −28.57 ± 0.16‰; C. Albers unpubl.). Therefore, we did not correct for these potential differences.

All isotopic ratios were expressed in the δ notation as parts per thousand (‰) differences from the primary (calibration) standards Vienna Pee Dee Belemnite (VPDB) for carbon measurements (δ¹³C), and atmospheric nitrogen for nitrogen measurements (δ¹⁵N).

Verification of accuracy and precision of BSIA measurements was done by measuring the secondary reference material USGS41 (δ¹⁵N = 47.6 ± 0.2‰, δ¹³C = 37.6 ± 0.1‰, measured: δ¹⁵N = 46.8‰, δ¹³C = 36.8‰), provided by the International Atomic Energy Agency (IAEA, Vienna). Measurement errors were indicated as ±0.8‰ for nitrogen and ±0.5‰ for stable carbon measurements, respectively (representing ±1 standard deviation of 17 analyses). Furthermore, the laboratory standards isoleucine (δ¹⁵N = −11.9‰, δ¹³C = −3.1‰) and peptone (δ¹⁵N = 8.0‰, δ¹³C = −15.7‰) were analyzed every five samples (Sigma Aldrich). Measurement errors were ±0.3‰ for nitrogen and ±0.6‰ for carbon isotope ratios of isoleucine (representing ±1 standard deviation of 16 analyses) and ±0.3‰ for both peptone measurements (representing ±1 standard deviation of 8 analyses). For CSIA measurements, quality assurance and analytical precision of the carbon stable isotope ratios were established by analyzing the certified standard FAME 14:0 (certified: δ¹³C = −30.0‰, measured: δ¹³C = 29.8‰), 16:0 (certified: δ¹³C = −30.7‰, measured: δ¹³C = −30.2‰), and 18:0 (certified: δ¹³C = −23.2‰, measured: δ¹³C = −23.8‰), supplied by Indiana University, every 5 samples. Analytical error was ±0.2‰ for 14:0 and ±0.3‰ for both 16:0 for 18:0 (representing ±1 standard deviation of 8 analyses). The samples were analyzed in duplicates, and true δ-values were obtained after two-point linear normalization (Paul et al., 2007). For details on sample preparation, measurements and analytical equipment for BSIA and CSIA see Kohlbach et al. (2016).

We estimated the proportional contribution of ice algae-produced carbon α_Ice to the body carbon of krill from the bulk and FA-specific carbon stable isotope compositions of POM from sea ice (simplified to ice algae) and the water column (simplified to phytoplankton), and krill by applying Bayesian multi-source stable isotope mixing models (MixSIAR) (Parnell et al., 2013; Stock and Semmens, 2015). The mixing models can account for different isotopic turnover rates in the consumers. However, krill-specific trophic enrichment of ¹³C is unknown, and was thus assumed to be zero for both BSIA and CSIA models (Budge et al., 2011; Graham et al., 2014; Wang et al., 2015). For the BSIA models, trophic enrichment of ¹⁵N was assumed to be 3.4‰ per trophic level (Minagawa and Wada, 1984). For the FA-specific modeling, α_Ice was calculated separately for (1) FAs associated with storage lipids (16:1n-7 + 18:4n-3) and for (2) FAs associated with biomembranes (20:5n-3 + 22:6n-3). Additionally, α_Ice was calculated by weighting the average mass proportions of storage and membrane lipids in each developmental stage, to account for differences in the contribution of the two lipid fractions to the total lipid content.

Sea ice drift data suggested a different origin of the ice sampled at ice camp 1 and ice sampled at ice camp 2 (Figure 2). To assess the potential influence of spatio-temporal variability of ice algae samples on the results, we investigated algal FA profiles and isotopic values separately for the two ice camps. Conversely, we assessed the potential effect of the spatial variability of the krill population structure by comparing FA profiles and isotopic values between geographically separated cohort groups identified by Schaafsma et al. (2016) for larval and juvenile krill (Figure 1, Table 1). Cohort 1 comprised predominantly juvenile krill from station 555-47. Cohort 2a was dominated by larval krill from stations 557-2 to 562-5. Krill from cohort 2b had similar length distribution and developmental stages to krill from cohort 2a, but were sampled considerably later in the season and farther north (stations 571-1 to 579-2). Cohort 3 krill (stations 565-5 to 567-2) differed considerably in the composition of krill developmental stages to the other cohorts (Schaafsma et al., 2016). Because the overall krill population was dominated by larval and juvenile krill in the sampling area, the cohort grouping was the basis of grouping the krill population for this study into four groups (Groups 1, 2a, 2b, and 3). Krill from stations 549-1 and 551-1 were classified as group 1 krill due to the close spatio-temporal proximity to station 555-47.

Based on the proportional α_Ice quantified for young krill (larvae + juveniles) derived from the CSIA model incorporating the stable isotope values of the storage FAs, ingestion rates (Pakhomov et al., 2004) and species abundances in the upper 2 m of the water column and the first 500 m of the water column during the sampling period (Schaafsma et al., 2016), we estimated the ice algae-produced carbon demand of young krill.

Variations in lipid class, FA and stable isotope compositions were tested using 1-way ANOVA followed by Tukey HSD post-hoc tests. For testing between ice algae and phytoplankton and to assess variability between the two ice camps, Student's t-tests were applied. A 2-way ANOVA was applied to determine if there was an interaction between the two independent variables (factors), carbon source (ice algae vs. phytoplankton), and sampling area (ice camp 1 vs. ice camp 2), on the dependent variable (stable isotope values). A Principal Component Analysis (PCA) was applied to visualize differences in krill FA proportions between the different groups. A distance-based analysis of similarities (ANOSIM) was applied to test for significant differences in FA proportions between krill from group 3 compared to all other groups, using a Euclidean distance matrix (Clarke and Ainsworth, 1993). FA data were transformed applying an arcsine square root function in order to achieve the near-normal distribution assumed by these parametric tests. In all statistical tests, we considered results with a statistical threshold value of α = 0.05 as significant. All performed data analyses were conducted using Software R, version 3.2.3 (R Core Team, 2015).

In the ice algae samples, the diatom-associated marker FAs 16:1n-7 and 20:5n-3 were the most abundant marker FAs, altogether contributing over 30% to the total fatty acid mass. Conversely, the dinoflagellate-associated marker FAs 18:4n-3 and 22:6n-3 were only present in small amounts (mean < 5%). In the phytoplankton samples, the mean proportional contributions of both diatom- and dinoflagellate-associated marker FAs ranged between 5 and 7%.

In the ice algae samples, the mean δ¹³C values of the bulk algal material and the diatom-associated marker FAs (16:1n-7, 20:5n-3) were on average significantly higher by 5–6‰ compared to the phytoplankton samples (Table 3). In the two dinoflagellate-associated marker FAs, significant differences in the mean δ¹³C values ranged between 9‰ (22:6n-3) and 14‰ (18:4n-3; Table 3). Only for 16:1n-7 δ¹³C values (dependent variable), the 2-way ANOVA indicated an interaction between the two carbon sources (ice algae and phytoplankton) and the ice camps (p < 0.05).

Between the two ice camps, the bulk δ¹³C values were significantly higher in the ice algal samples from ice camp 1 vs. ice camp 2. Contrastingly, the δ¹³C values in the FAs 16:1n-7 and 22:6n-3 were significantly higher in the ice algae samples from ice camp 2 vs. ice camp 1 (Supplementary A). In phytoplankton, there was no significant spatial variability in the δ¹³C values.

The different drift history of the two ice camps in combination with differences in the isotopic composition of ice algae between the two ice camps indicated that the variability of ice algae parameters needed to be accounted for in the computation of the proportional contribution of ice algae-produced carbon α_Ice of the analyzed krill individuals. Therefore, we calculated α_Ice using ice algae stable isotope values from ice camp 1 for krill from groups 1 and 2a due to the close spatio-temporal proximity between the sampling, and the similarities in FA compositions. Krill from cohort 2b were also included in this group due to their similarity in fatty acid and stable isotope parameters with groups 1 and 2a. The trophic dependency on ice algal carbon in krill from group 3 was calculated with the ice algae stable isotope values from ice camp 2, based on the close spatio-temporal proximity between the sampling, the similarity in isotopic values between krill from group 3 and ice algae from ice camp 2, and the different drift trajectory of the sea ice region where group 3 krill were sampled compared to groups 1 and 2 (Figure 2).

Both BSIA and CSIA indicated higher mean α_Ice estimates in larval and juvenile krill compared to adult krill. BSIA-based mean proportions of ice algae-produced carbon α_Ice, considering the mean estimates of the different groups in each developmental stage, were 44% in larvae, 29% in juveniles, and 20% in adults (Figure 5A). CSIA-based estimates of α_Ice were calculated separately for storage lipids (FAs 16:1n-7 and 18:4n-3) and membrane lipids (FAs 20:5n-3 and 22:6n-3) in each individual krill. This procedure ensured integrating the trophic signal of both diatom- and dinoflagellate-associated marker FAs, respectively in each of the two lipid pools. In the CSIA-based mixing model using storage-associated FAs, mean α_Ice-values were between 69% in larvae and 65% in juveniles, and 53% in adults (Figure 5B). Mean α_Ice-values in membrane-associated FAs were considerably lower, with values between 26 and 33% in larvae and juveniles, and 17% in adults (Figure 5C). To account for the differences between storage lipids and membrane lipids in their relative contribution to the lipid mass, we estimated the overall CSIA-based α_Ice, weighted by the proportional mass of the two lipid fractions in each developmental stage. As a result, the mean CSIA-based relative dependency on ice algae-produced carbon was 39% in larvae, 46% in juveniles, and 33% in adults (Figure 5D).

The variability in α_Ice estimates between the groups was higher in larval and juvenile krill than in adults (Table 7). Variability in young krill was particularly evident in the α_Ice estimates based on the storage-associated FAs 16:1n-7 and 18:4n-3. Mean α_Ice estimates using these two storage-associated FAs indicated the highest trophic dependency on ice algae-produced carbon in larvae from group 2a (88%) and the lowest contribution of ice algal carbon to larval krill from group 3 (52%). In juveniles, mean α_Ice estimates were somewhat higher in krill from group 2b vs. groups 1 and 2a, based on the storage-associated FAs. Furthermore, BSIA results indicated a higher contribution of ice algae-produced carbon to juveniles from groups 2a and 2b vs. group 1. The weighted total CSIA estimates were similar among the groups in each developmental stage. Only the estimate for larvae from group 2a was considerably higher compared to the estimates from groups 2b and 3 (Table 7).

Lipids provide two essential functions in living organisms: (1) they constitute structural components in membranes, and (2) they are a component for cellular energy storage (Lee et al., 2006). In most Antarctic euphausiids, energy is mainly stored in wax esters, providing long-term energy compared to triacylglycerols (TAGs), which are faster metabolized (Fricke et al., 1984; Stübing et al., 2003; Ko et al., 2015). Krill, however, mainly rely on TAGs and phosphatidylcholines (PCs) as energy stores for the survival of food-limited periods (Hagen et al., 1996). The observed ontogenetic differences in lipid class compositions among the different developmental stages were consistent with the known higher energy storage capacity of adults compared to younger stages (Meyer, 2012).

The distribution of marker FAs in the ice algae samples confirmed the well-known dominance of diatoms in the ice algal community (Garrison and Close, 1993; Lizotte, 2001). The evenly distributed proportions of diatom- and dinoflagellate-associated FAs in the phytoplankton samples indicated a co-dominance of both algal taxa (Buck and Garrison, 1983; Garrison et al., 1991). The high proportions of the long-chain FAs 20:5n-3 and 22:6n-3 in all three krill developmental stages were in agreement with previous studies (Falk-Petersen et al., 2000; Virtue et al., 2016). Hagen et al. (2001) also found higher levels of 20:5n-3 and 22:6n-3 accompanied by lower levels of 16:1n-7 in krill furcilia larvae compared to adults, suggesting that the fatty acid signatures of larvae might provide a more evident dietary input of primary synthesized FAs compared to the FA content of the adult krill. The diatom-associated marker FA 16:1n-7 and the dinoflagellate-associated marker FA 18:4n-3 are mainly incorporated into storage lipids, whereas the diatom-associated marker FA 20:5n-3 and the dinoflagellate-associated marker FA 22:6n-3 are predominantly incorporated in membrane lipids (Stübing et al., 2003). Hence, in this study we considered the two pairs of FAs separately in their two corresponding lipid fractions. The two storage-associated marker FAs are metabolically more dynamic, and hence more likely to reflect dietary patterns during the sampling periods than the highly conserved membrane-associated marker FAs, which possibly still reflected in parts the trophic signal from autumn grazing (Stübing et al., 2003). The higher ratio of diatom-associated vs. dinoflagellate-associated FAs in storage vs. membrane lipids in all three developmental stages indicated that diatoms were a more important carbon source for krill during the weeks of our sampling than integrated over the months before.

Several Antarctic copepod species, including Oithona spp., Metridia spp., Calanus propinquus, and Calanoides acutus, produce the long-chain FA isomers 20:1 and 22:1 de novo in large amounts, which can be used as carnivory marker FAs in omnivorous or carnivorous predators (Kattner et al., 1994; Cripps and Hill, 1998; Ju and Harvey, 2004). The low levels of these FAs in all krill samples suggested that copepods did not serve as major food source during our sampling period. It has to be considered, however, that the FAs 20:1n-9 and 22:1n-11 are largely egested with the krill's fecal pellets (Stübing et al., 2003; Stübing, 2004). Therefore, the estimation of the absolute copepod-derived carbon uptake based on relative proportions of these FAs in the consumers can be difficult when they are metabolized quickly. Results from stomach analysis indicated that other copepod genera (such as, Stephos longipes) were an important diet component at least during the hours and days of our sampling (Schaafsma et al., 2017).

Estimates of the proportional contribution of ice algal carbon α_Ice based on BSIA can be influenced by the varying metabolic fractionations of δ¹³C in the different organic compounds, i.e. lipids, proteins, and carbohydrates, of the bulk material (DeNiro and Epstein, 1978; Kohlbach et al., 2016, 2017). In contrast, δ¹³C values of trophic marker FAs are assumed to be unchanged by metabolic processes, and independent from the chemical composition of organisms. CSIA allows estimating δ¹³C in individual marker FAs to accurately trace and quantify sea ice-produced carbon in food webs (Wang et al., 2015; Kohlbach et al., 2016). Here, we applied for the first time both stable isotope methods to quantify the relative dependency of overwintering krill on ice algae-produced carbon. Both BSIA- and weighted CSIA-based estimates of α_Ice consistently showed a high dependency on ice algae-produced carbon by larval krill, and a lower dependency in adults (Figure 5).

The isotopic turnover rate in an organism depends on its feeding rate and the incorporation of new dietary carbon when switching to a different food source, e.g., from pelagic algae to ice algae. Particularly in winter, krill larvae have higher feeding rates compared to adult krill, and a large proportion of the ingested carbon is utilized for growth (Atkinson et al., 2002; Meyer et al., 2002). Contrastingly, in adults the signal of more recent ice algae-derived carbon uptake may have been buffered by their older and more modified lipid pool (Stübing et al., 2003). Because of the varying carbon turnover rates between the different developmental stages, the lower α_Ice estimates from both the short-term and the long-term isotopic compositions in adults may have been negatively biased compared to larvae and juveniles. Overall, CSIA-based estimates of α_Ice in membrane FAs probably retained in parts the trophic signal of pre-winter feeding, because these FAs are highly conserved and accumulated over long time periods (Stübing et al., 2003). Conversely, CSIA-based estimates of α_Ice in storage-associated FAs rather represented carbon sources during our sampling period, indicating that about two-thirds of the carbon demand in larval and juvenile krill originated from ice algae in late winter. Our weighted CSIA-based estimate of α_Ice should therefore be considered as a conservative estimate.

δ¹³C values in POM can vary interanually, seasonally and regionally, ranging from −16 to −28‰ in ice-associated POM and from −25 to −31‰ in pelagic POM from the Weddell Sea (Rau et al., 1991). The large range of ice algae δ¹³C values in our study suggests that the sampled ice algae were representative for the range of values that krill have encountered throughout the research area. Thus, we are confident that our mixing models of the ice algae-derived carbon uptake in krill were not biased by the specific temporal and spatial framework of our sampling. In the areas around both ice camps, δ¹³C from bulk material and marker FAs could be used to confidently discriminate between ice algae—and pelagic algae-associated carbon sources. However, the FA and stable isotope compositions varied considerably between ice algae from the two ice camps (Supplementary A). Such differences can occur due to variations in the isotopic baseline during ice formation, variations in ice thickness, snow cover, and light intensity, subsequently affecting growth conditions of the ice algae. Our backward-projected sea ice drift data indicated that such variations could have been related to differences in the origin of ice formation, ice age, and drift trajectory of the sea ice in the areas of the two ice camps (Figure 2). The sea ice tracking indicated that much of the sea ice near ice camp 1 fostered the development of microbial in-ice communities over a longer period than the sea ice near ice camp 2, where the drift possibly only allowed for the development of microbial in-ice communities with lower food quality, coinciding with differences in the FA composition and a reduced physiological condition of krill larvae of group 3 (Schaafsma et al., 2016; Figure 2). Passive-microwave retrieved ice drift products are provided by different institutions and have been widely used in sea ice studies and for model assimilation (e.g., Spreen et al., 2011; Sumata et al., 2014; Krumpen et al., 2016). According to Sumata et al. (2015), the uncertainty of satellite-based drift products in the Arctic range from 0.9 to 2.0 cm s⁻¹, depending on the type of product, drift speed, ice concentration and season. Uncertainty estimates are based on a comparison of buoy derived sea ice motion with satellite derived drift. Unfortunately, the uncertainty of motion products in the Antarctic has not been studied in detail. Therefore, we assume the uncertainty to be within the range of estimates carried out in the Arctic.

Accounting for the spatio-temporal variability in ice algal δ¹³C values between the two ice camps, we determined the utilization of ice algal carbon in krill grouped into four different groups (Schaafsma et al., 2016). Accordingly, ice algae-produced carbon contributed up to 88% to the carbon budget of larval krill at the earlier sampled locations near ice camp 1 in the closed pack-ice (group 2a), whereas the dependency on ice algal carbon was approximately 20% less in individuals sampled near the ice edge (group 2b), based on the combined isotopic information of the short-term FAs 16:1n-7 and 18:4n-3 and the weighted total CSIA results (Table 7). Supporting these results, the impact of diatom-vs. dinoflagellate-associated FAs in the storage lipid fraction was higher in young krill from group 2a vs. group 2b, possibly reflecting the higher impact of ice algae—vs. pelagic algae-produced FAs in the weeks before the sampling. The juvenile krill from group 1 mainly comprised individuals in their second year, whereas the other groups consisted predominantly of juveniles in their first year (Schaafsma et al., 2016). Thus, a decrease in the utilization of ice algae-produced carbon with ontogeny might have been reflected by the considerably lower α_Ice values in juveniles from group 1 compared to groups 2a and 2b based on the bulk stable isotope values and δ¹³C values in storage-associated FAs.

A good body condition of young krill from group 2a was indicated by high proportions of stored lipids in form of TAG, which further supports the assumption of the constant feeding on ice-associated biota. The higher δ¹⁵N values in all three krill developmental stages from group 2b in comparison to the other groups indicated an elevated indirect uptake of ice algae-produced carbon via predatory feeding. These results suggest a somewhat lower trophic association to sea ice when other food sources than ice algal carbon become more abundant in the water column, triggered by enhanced melting and pelagic productivity at the onset of spring.

Krill from group 3 differed considerably in lipid, fatty acid and stable isotope composition compared to the other three cohort groups. The high proportions of the dinoflagellate-associated FA 22:6n-3 were in agreement with the considerably lower α_Ice estimates for larval krill from group 3 vs. groups 2a and 2b (Table 7), and might have reflected the higher proportions of 22:6n-3 in ice algae from ice camp 2 vs. ice camp 1. The low δ¹³C and δ¹⁵N values in larval krill from group 3 suggest that the sea ice conditions and availability of sea ice resources were less favorable for krill in this region compared to the other regions. However, the trophic dependency on ice algal carbon by young krill was overall significant throughout the entire sampling area and sampling period, and still above 60% in krill near the ice edge (group 2b), suggesting a generally high importance of ice algae-produced carbon during winter.

With a good agreement between both isotopic approaches, this study demonstrates for the first time that up to two-thirds of the carbon demand of young krill during their first winter is covered from ice algae production. It has been suggested that high abundances of young developmental stages of krill underneath ice floes indicate that feeding on sea ice biota is important for their winter survival, but this behavior may also be related to other factors, e.g., predator avoidance (Hamner et al., 1989; Frazer et al., 2002; Meyer et al., 2009; Flores et al., 2012b; David et al., 2016). A high importance of ice algae in the diet was supported by analyses of stomach contents and FA compositions in previous studies, finding that diatoms were an important carbon source of young krill (Daly, 1990; Schmidt et al., 2014). Stomach content and FA analyses alone, however, cannot accurately discriminate between the signal of ice algae and phytoplankton, because these two communities overlap significantly in species composition (Garrison and Buck, 1985; Garrison et al., 1987; Melnikov, 1998). Heterotrophic prey species sometimes also account for a significant part of the diet of krill (Perissinotto et al., 2000), but their link to ice algae as a carbon source is not always clear, because the trophic signal is diluted from one trophic level to another (Meyer et al., 2009; Schmidt et al., 2014). Overcoming the uncertainty of these well-established methods to reliably trace the trophic signal of ice algae, our isotopic biomarker approach provides quantitative evidence of a crucial dependency of particularly larval and juvenile krill on ice algae-produced carbon, integrated over several trophic levels.

In a recent study by Jia et al. (2016), bulk stable isotope data of krill collected from East Antarctica during two winter-spring transitions are compared. As found in our study, adult specimens had lower δ¹³C values than larval and juvenile krill, confirming the common notion that our observed ontogenetic trend of a higher utilization of ice algal carbon by younger krill is a rather general pattern in different regions of the Southern Ocean (Meyer, 2012). A lower dependency of adult krill on ice algae-produced carbon (Figure 5) in combination with a higher proportion of storage lipids confirms the paradigm that adult krill survive the winter mainly by reducing metabolism and mobilizing lipid reserves accumulated during summer and autumn (Meyer, 2012). This assumption was supported by previous studies on the feeding behavior of postlarval krill, reporting very low feeding rates (Morris and Priddle, 1984; Quetin and Ross, 1991) and no significant growth during austral winter (Kawaguchi et al., 1986; Siegel, 1987). In our study, mean lipid contents in adults of 7% dry mass were at the low end of the typical range of 5–30% for winter and spring (Hagen et al., 1996; Ju and Harvey, 2004), indicating that energy reserves were almost depleted. For the comparison of the lipid levels with other studies, it has to be considered that instead of the total lipid content, we determined the mass percentage of the total fatty acid content. The total lipid content, containing molecule backbones besides the single FAs, can be assumed to be somewhat higher. The low lipid levels in our study in combination with the high α_Ice estimates based on the storage fatty acids, representing the more recent carbon source, suggest that ice algae-produced carbon could become critical for adult krill to survive the late winter-spring period, as long as phytoplankton production remained low (Meyer, 2012). Winter-feeding on sea ice biota by post-larval krill is corroborated by reported concentrations of juvenile and adult krill at the ice underside during winter and early spring (Marschall, 1988; Flores et al., 2012b). Both in larvae and adults, however, our knowledge about the seasonal variability of their dependency on ice algae-produced carbon is still limited. Therefore, trans-seasonal studies on the role of ice algae in the carbon budget of krill are needed to pinpoint critical time periods for krill survival.

Krill are distributed throughout the Southern Ocean, extending over about 27 degrees of latitude, from coastal waters to deep-sea and from year-round ice-covered to virtually ice-free regions, and are adapted to survive a wide range of environmental conditions (Flores et al., 2012a). The dependency of overwintering krill on ice algae-produced carbon is therefore likely to vary regionally, particularly driven by the duration of the sea ice season (Jia et al., 2016). A large part of the circum-Antarctic krill population (Atkinson et al., 2008) as well as major recruitment areas (Meyer, 2012) and fishing grounds (Flores et al., 2012a) are found in the Weddell-Scotia Confluence Zone and adjacent Scotia Sea. Stretching over a distance of approximately 900 km from west to east, our study area represented a major part of this key region of krill distribution and recruitment (Figure 1).

Based on an ingestion rate of 23 μg C ind. d⁻¹ (Pakhomov et al., 2004), and α_Ice values for larval and juvenile krill derived from storage FAs (69 and 65%, respectively), first-year krill took up about 0.016 mg ice algal carbon and 0.007 mg phytoplankton carbon ind. d⁻¹ during our sampling period. The northern Weddell Sea/Scotia Sea has experienced a shortening of the sea ice season of about 10 days per decade between 1979 and 2006, and it may be delayed by up to 40 days by the end of this century (Flores et al., 2012a; Piñones and Fedorov, 2016). Without sea ice, carbon availability in the water column would need to at least triple in order to support the 2012 standing stock of overwintering first-year krill in our study region. In the Weddell-Scotia Confluence Zone, chlorophyll a and pigment concentrations in the water column rarely exceed 0.1 mg m⁻³ between April and August even in ice-free waters (Comiso et al., 1993; David et al., 2016). In contrast, typical Antarctic bottom ice algal chlorophyll a concentrations during these months are in the order of 2–4 mg m⁻² (Meiners et al., 2012). This suggests that the gap created in the krill energy budget by a loss of ice algae is highly unlikely to be filled by phytoplankton in an ice-free winter. More importantly, sea ice preserves carbon in the form of ice algae and heterotrophic sea ice biota throughout the dark period, when even ice algae production is nearly zero (Arrigo and Thomas, 2004). This carbon storage capacity will be reduced when delayed sea ice formation shortens the period for ice algal production before winter darkness (Arrigo and Thomas, 2004).

There is some evidence, however, that in the Southern Ocean climate change-related decline of sea ice and, consequently, ice algae production, may be balanced by an increase of overall phytoplankton productivity due to a longer ice-free growth season, higher temperatures and better iron supply (Arrigo and Thomas, 2004; Bopp et al., 2013; Vancoppenolle et al., 2013). This effect, however, may be counter-acted by deeper mixing due to increasing wind speeds in the future Southern Ocean (Lovenduski and Gruber, 2005; Cai, 2006). Furthermore, climate change could come at the cost of changes in phytoplankton composition toward taxa less supportive of krill, e.g., from diatoms to cryptophytes (Moline et al., 2004). With a comprehensive study comparing multiple models of the distribution of krill spawning habitats at the end of the twenty first century under various scenarios of environmental change, Piñones and Fedorov (2016) concluded that in most scenarios the available spawning habitat in the Southern Ocean declines, even when the models accounted for increased phytoplankton productivity and hence better energy reserves of krill at the onset of winter. Altogether, a prolonged growth season and/or increased primary production of Antarctic phytoplankton seem unlikely to significantly counter-act the detrimental effect of sea ice decline on the winter survival of first-year krill.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.