Highlights

  • Amino acid isotopes show the utilization of marine subsidies by resident char.

  • Glacial meltwater increases the dependence on marine subsidies.

  • Glacier melting can influence the transfer of subsidies.

Introduction

Glacial ice covers approximately 11% of the Earth’s land surface but is threatened due to global warming, resulting in a significant loss of glaciers worldwide (Zemp and others 2019). For instance, predictions indicate that 70% of glaciers in western Canada will disappear by the end of the twenty-first century (Clarke and others 2015). As a consequence, glacier meltwater contribution to water bodies is predicted to change drastically in the near future (Kaser and others 2010). The retreat of glaciers creates depressions where postglacial lakes form, while proglacial rivers are formed by the melting of glaciers. Glacial meltwater is characterized by low temperatures, seasonal high discharge, and high turbidity caused by glacial sediments (Slemmons and others 2013). The input of glacial meltwater typically diminishes local productivity (Milner and others 2001; Dunkle and others 2023), and proglacial ecosystems are usually dependent on allochthonous resources, such as marine resources (Whitney and others 2018), terrestrial carbon (Lafon and others 2014) or dissolved organic carbon from glaciers themselves (Fellman and others 2015). Such spatially-coupled ecosystems, conceptualized in the meta-ecosystem framework, are often dependent on cross-ecosystem linkages, and exhibit intricate and dynamic patterns of species interactions (Loreau and others 2003; Leroux and Loreau 2008). Therefore, to better understand the nature of threatened meta-ecosystems in a warming world, integrating cross-ecosystem linkages into the study of proglacial ecosystems is a necessary step (Anderson and others 2017).

A pivotal concept in the meta-ecosystem framework is that the significance of cross-ecosystem linkages, also defined as subsidies, is contingent upon the nutrient availability within the recipient ecosystem (Polis and others 1997; Subalusky and Post 2019). For example, the reliance of lakes on terrestrial organic carbon predominantly hinges on nutrient availability (Carpenter and others 2005), thereby resulting in small alpine lakes relying heavily on terrestrial prey (Milardi and others 2016). In this context, we posit that glacial meltwater, with its potential to diminish autochthonous productivity, could increase the reliance on subsidies within freshwater ecosystems, consequently shaping the dynamics of the meta-ecosystem. Despite the apparent simplicity of this hypothesis, the interplay between glacier meltwater and resource subsidy in aquatic ecosystems remains unexplored in the field. The lack of empirical validation is understandable due to challenges in setting up experiments that simultaneously test glacier meltwater input, resource subsidy, and their interaction in a natural environment. For instance, glacial meltwater input is linked to different watershed environments, including differences in vegetation cover, which also influences subsidies (Clitherow and others 2013; Sejr and others 2022).

In freshwater ecosystems, the movement of animals, including fish, plays a key role in mediating subsidies (Flecker and others 2010). In arctic ecosystems, anadromous salmonids are known to spawn in proglacial streams (Pitman and others 2020), and their migration between marine and freshwater habitats is a significant vector of marine-derived subsidies to freshwater ecosystems (Wipfli and others 2003; Denton and others 2010; Collins and others 2016). Salmonid spawning events create pulsed subsidies, transferring marine organic matter and nutrients that profoundly impact meta-ecosystem functioning (Moore and others 2011). For instance, up to 75% of consumers in streams can rely on salmon-derived marine nitrogen (Naiman and others 2002). As marine subsidies do not originate from watershed sources, investigating the utilization of marine subsidies becomes relevant when exploring the impact of glacial meltwater on subsidy utilization. Here, we specifically test the hypothesis that glacial meltwater increases the utilization of marine subsidies in freshwater ecosystems.

To test this prediction, we analyzed the utilization of marine subsidies by freshwater arctic char, Salvelinus alpinus, hereby referred to as char, in four sites from a river system in Southern Greenland, which shows low levels of human development. This Greenlandic river meta-ecosystem configuration is particularly useful for our study because rivers of the same drainage can originate from either glacial meltwater (proglacial river) or not (tundra river). Within the river catchments, resident char is widely distributed, while anadromous char faces migration barriers, preventing them from ascending beyond waterfalls. Our study site was chosen to encompass each of four factorial combinations of the presence and absence of glacial inflow and the potential for marine subsidies (for example, due to barriers to anadromous fish migration).

We employed bulk and amino acid stable isotopes of muscle tissues from arctic char, acting as the top predator in the ecosystem. The use of amino acid isotopes in apex predators proves particularly valuable as they inform on the source of primary production sustaining top consumers (Larsen and others 2013; Manlick and Newsome 2022), and the trophic position of organisms in the food chain (Chikaraishi and others 2014). Based on previous work about the transfer mechanism of marine subsidies (Svenning and Borgstrom 2005; Jaecks and Quinn 2014; Frye and others 2021), we hypothesized that resident freshwater char individuals, in partially migratory populations, could effectively utilize marine subsidies by cannibalizing eggs from returning anadromous char. We hypothesized that this utilization of marine subsidies will increase in proglacial environments, following the predictions of meta-ecosystem theory. More specifically, we asked if subsidy utilization increase was achieved through an individual dietary increase of the subsidy utilization, and/or more individuals turning cannibals. As cannibalism represents a niche shift through ontogeny (Pereira and others 2017), we analyzed the effect of fish size on the utilization of subsidies, and its interaction with environmental conditions.

Methods

Site and Sample Collection

We collected char during three field campaigns carried out in July–September of 2018, 2019, and 2021. The study was conducted in the Qinngua valley, which is situated north of Narsarsuaq, southwest Greenland (Figure 1A). The study area comprises a main drainage where the primary river, Qinnguata Kuua, discharges glacial meltwater into the fjord (Figure 1B). Glacier positions in the drainage have been generally stable over the last ~ 11,000 years, following a period of rapid glacial retreat (Larsen and others 2011). The drainage is mostly ice-free (Figure 1A), and the main river exhibits low conductivity (14 μS/cm) and nutrient concentrations (total dissolved P = 0.010 μmol/L) at discharge to the estuary, indicative of oligotrophic conditions (Appendix 1). Upstream of their confluence, we sampled two tributaries: a proglacial river and a tundra river (Figure 1B). These tributaries display contrasting characteristics in turbidity (Figure 1C, D) and temperatures, with mean summer temperatures of 8.8 and 10 °C in the proglacial and tundra rivers, respectively. A study from the same area showed that dissolved organic carbon is twice higher in tundra streams than proglacial ones, supporting more oligotrophic conditions in proglacial water (Martin and others 2020). To characterize the local changes in glacial influence, we used modeled land and ice discharge data based on field and satellite observations from (Mankoff and others 2020). Qinnguata drainage data (Mankoff and others 2020) were compiled to determine total annual land and glacial runoff, along with relative glacial input since 1970. While total discharge did not vary over time (linear regression, P = 0.5), the total glacier discharge increased by ~ 13% per decade (P < 0.01, R2 = 0.20), driving an increase of glacier meltwater input by ~ 1% per decade (P < 0.01, R2 = 0.12, Figure 1E), indicating a trend of increasing glacier meltwater input.

Figure 1
figure 1

Map of the study system A, Satellite image of the study site area in Southwest Greenland, with dark lines denoting drainages. An arrow on the map of Greenland (bottom) indicates the location of the sites. The red rectangle shows the location of the study sites shown in B. B, Map illustrating the study sites. Triangles denote migration barriers, while crossed circles represent the four sample sites (refer to legend). P1 and P2 indicate the locations of photographs (C and D), with symbols indicating the viewing direction. C, depicts the confluence of proglacial and tundra streams. D, highlights the main proglacial river. E, Discharge (blue) and relative proportion of glacier meltwater (red) in the Quinnguata drainage since 1970.

Full size image

The rivers were inhabited by the same partially migratory population, consisting of individuals that either stay resident or undergo marine migration and return to spawn (Chapman and others 2012). Char density was notably high, with observations of char spawning throughout the entire stretch up to the first migratory barrier, approximately 5 km upstream from the mouth (Figure 1B). To account for migratory influence on the population, we sampled sites above migratory barriers (waterfalls) references for a fully freshwater diet. We sampled two different reference populations: one directly above the proglacial river and another in a tundra stream that has similar characteristics as the previously mentioned tributary, as it lacked migration barrier (Figure 1B). Although the two tundra rivers are from different drainages (Figure 1A), their drainage is located in the same granitic geological formation (Sørensen 2016), and they share similar physical characteristics (mean width: 3.3 and 4.7 m, mean depth: 22 and 26 cm).

Kick sampling in 2022 revealed that the proglacial control stream had a total invertebrate count approximately five times lower than both tundra streams (Figure S1), in line with a previous survey in the sites with migration (Grønlands Fiskeri- og Miljøundersøgelser 1984). Therefore, the available environmental data indicate that the study site offered a natural contrast of rivers under proglacial influence, depicting differences in resource availability with or without migration.

We collected fish by electrofishing and hand nets. Resident fish were captured in a semi-quantitative manner, meaning that fish of all sizes were collected. Although fishing was made difficult in turbid water, we assume that the capture method is comparable between sites, enabling a comparison of size distribution. After capture, we euthanized the fish with clove oil (as carcasses were dedicated to other projects), photographed their left lateral side, and measured standard body length and total length to the nearest mm, weighed wet-mass, and took a biopsy of the dorsal muscle. In the field, or later based on pictures, we visually assessed life history forms (that is, anadromous or resident life history forms) based on their color and morphology (see Figure 2A for example), as described in (Takatsu and others 2022). In brief, migratory char can be visually assessed mainly based on four traits (Doenz and others 2019; Grenier and Tallman 2021): body coloration (silvery hue in young migrants or adults outside of the reproductive season), fin shape (more rounded in river residents), fin length (longer in residents), and relative head size (larger in residents).

Figure 2
figure 2

Bulk isotope analysis A, Char pictures of typical representative of the six cohorts. B, Bulk stable isotopes of carbon (x-axis) and nitrogen (y-axis). Crosses show the fish selected for amino acid stable isotope analysis. Ellipses show 50% confidence interval bimodal distribution.

Full size image

Analysis of Bulk Isotopes

We sampled a total of 404 fish that we analyzed for bulk stable isotopes δ15N and δ13C (Table S1, see Appendix 2 for full description of bulk isotope preparation). Because some fish had high lipid content (~ 14% had CN molar ratio > 4.5, mean 4.1), we corrected for 13C-depletion (Δδ13Clipid) due to lipid content, following the model by (Hoffman and others 2015). This approach assumes a mass balance of lipids (carbon) and proteins (carbon and nitrogen), such as the proportion of lipids (flipid) is given by: flipid = 1 – C:Nlipid free/C:Nbulk. Here, C:Nbulk represents the C to N molar ratio in muscle sample, and C:Nlipid free the C to N molar ratio after lipid extraction (here we used 3.4, which we empirically found, see Appendix 2).

The lipid corrected δ13C is thus given as: δ13Clipid free = δ13Cbulk – flipid. Δδ13Clipid, with δ13Cbulk the measured δ13C in untreated muscle sample. Here, we used Δδ13Clipid = 7 ‰, which is a common value reported in fish and salmonid (Post and others 2007; Hoffman and others 2015; Mumby and others 2018). Our approach was supported by data from another char collection for which we extracted lipids from frozen samples, more details can be found in Appendix 2.

Analysis of Amino Acid Isotope

For stable isotopes of amino acids (AAs), we took a subset of 73 char (Figure 2B). We weighed an amount of ~ 4 mg fish muscle, hydrolyzed the samples with 1 mL 6 M HCl at 110 °C for 20 h, and derivatized the free AAs to N-Acetyl Methyl Ester (NACME) derivatives following previous protocols (Corr and others 2007; Larsen and others 2013). Briefly, we first extracted the lipophilic compounds using 2 ml n-hexane/DCM. We added 0.1–1 mL (depending on the sample concentration) norLeucine 4.7 μmol.L-1 as an internal standard. Prior to derivatization, the samples were evaporated at 110 °C under N2. First, we methyl-esterified carboxylic groups using an excess of acidified methanol (6:1 anhydrous methanol:acetyl chloride) for 60 min at 75 °C. Then, amine groups were acetylated using an excess of acetylation mixture (1:2:5 acetic anhydride:triethylamine:acetone) for 10 min at 60 °C. Before and after each reaction, we flushed the vials with N2 and dried excess solvent under N2 at 60 °C, respectively. We dissolved the samples in 2 mL ethyl acetate and removed precipitates and salt using 1 mL of saturated NaCl solution. We dried the ethyl acetate solution of NACME-AA under N2 to a final volume of ~ 300 μL that we transferred to GC vials. We also prepared an in-house AA standard containing 15 AAs of interest, whose isotopic composition was measured by EA-IRMS: Ala, Gly, Val, Leu, Ile, Pro, Asp, Ser, Thr, Glu, Met, Phe, Lys, Tyr, and Met. Individual AAs were purchased at Sigma-Aldrich, purity > 99.5%.

We ran every sample in duplicate or triplicate for C and N, respectively. For δ13C, to account for the isotopic kinetic effect of the carbon added during derivatization, we calculated correction factors using in-house standard measurements following the formula in (Corr and others 2007). For nitrogen, we reported measurements for Glx because during hydrolysis, glutamine (Gln) is deaminated to glutamic acid (Glu), and Glx δ15N corresponds to the δ15N value of the α-amine of Glu and Gln.

We separated AAs on an Agilent 6890 Series Gas-Chromatography System equipped with a Gerstel MultiPurposeSampler MPS2 and an InertCap 35 GC column (GL Sciences, 60 m X 0.32 mm X 0.50 mm), using the same temperature program as in Larsen and others (2013). We ran the samples in four different batches, for which we independently derivatized the in-house standard. We determined the isotopic ratio using a Delta V/ Delta V + Advantage, Thermo coupled to a combustion tube. The standard measurements did not statistically differ for the reported values of the four batches (ANOVA analysis of variance for each reported AA δ15N and δ13C, P > 0.05). For carbon, we had a mean standard deviation of 0.5‰, including propagation of uncertainty due to added carbon. For nitrogen the standard deviation was 0.4‰ and 0.5‰ for Glx and Phenylalanine (Phe), respectively. The standard deviation Δ15NGlx-Phe was 0.65‰, which corresponded to the propagation of uncertainty and indicated no correlation of measurements of Glx and Phe within sequences.

EAA Fingerprinting and Bulk δ13C to Characterize Freshwater Diet and Marine Subsidy

Freshwater and marine primary producers, such as algae, are usually associated with distinct δ15N and δ13C values (Peterson and Fry 1987). However, quantifying the consumption of a marine subsidy into freshwater consumers is difficult as consumers might discriminate 13C, and because δ13C of freshwater prey might largely vary between species, locations and seasons (Sertić Perić and others 2020). This holds particularly true for arctic systems that are characterized by strong seasonal variations (Benstead and Huryn 2011).

To alleviate those difficulties, we characterized basal production using the essential amino acid (EAA) fingerprinting, which represents the relative δ13C difference between EAAs. The EAA fingerprinting, here “fingerprinting,” arises from the distinct anabolic pathways of primary producers for the different EAAs, which are minimally influenced by variations in carbon sources and growth rates (Scott and others 2006; Larsen and others 2009; Stahl and others 2023). For instance, Phenylalanine shows more positive δ13C values (relative to other EAAs) in plants than in algae (Liew and others 2019; Saboret and others 2023). As EAAs show little δ13C fractionation up the food chains (Whiteman and others 2019), consumers that feed on one resource can be used to infer the resource fingerprinting (Skinner and others 2021). As consumers integrate proteins over a long period of time, their fingerprinting reflects the resource average over seasons (Matley and others 2016). To use the fingerprinting approach, we mean-normalized δ13C values (δ13CEAA.norm), given by: δ13CEAA.norm = δ13CEAA—\({\sum }_{i}^{EAAs}{\delta }^{13}{C}_{i}\) / NEAAS, which represents the relative difference between an EAA δ13C value and the mean δ13C value of all EAAs, with NEAAs the amount of EAAs considered.

We explored whether char fingerprinting originated from algae, bacteria and/or terrestrial plants. Our motivation was twofold: first, to test for dietary reliance on terrestrial consumption, as this is needed for estimating trophic position (see below); and second, to assess whether freshwater and marine productions in our system had distinct phyla foundations. Here, we characterized freshwater production and marine subsidy based on the resident above barriers (as they feed exclusively on freshwater resources), and migratory char (as they feed in the marine environment and transfer the subsidy). To predict the relative contribution on algae, bacteria and plants, we used the same method as in (Saboret and others 2023). In brief, the method is based upon literature δ13C values for algae, plants and bacteria, of Ile, Leu, Phe, Thr and Val. We employed bootstrapped linear discriminant analysis with 1000 iterations, resampling 80% of the source dataset and modeling consumer data to account for uncertainty in analytical precision. By doing so, the method approximates the posterior probability of dietary reliance assignment.

Finally, we assessed the fingerprinting differences in our system between freshwater and marine sources of production. We tested for differences in EAA δ13C normalized values between resident controls, marine feeders and residents with migration, using pairwise Kruskal Wallis tests with post-hoc comparisons using the Fisher’s least significant difference, in the package agricolae (De Mendiburu and Simon 2015). To maximize the separation between marine and freshwater sources, we performed a linear discriminant analysis (LDA) on normalized δ13C values of Ile, Leu, Met, Phe, Thr and Val. To test if there was a predictive relationship between the fingerprinting and bulk δ13C, we performed a linear regression of the LD score (LD1) against δ13C for char in subsidized context. By doing so, we aimed at confirming that the variations in char δ13C in a subsidized context were truly representative of the assimilation of EAAs (and so proteins) from marine subsidies. Because tundra and proglacial rivers differed in their freshwater δ13C baselines, we performed two separate regressions.

Trophic Position Estimates

The stable isotope ratio of nitrogen, δ15N, has been widely applied in ecology to estimate trophic position (TP) as consumers are on average enriched by ~ 3 ‰ more than their prey (DeNiro and Epstein 1976), albeit the trophic discrimination factor (TDF) might be largely influenced by physiological factors, such as the nutritional quality of dietary sources (Stephens and others 2023). In our study, assessing the TP of char proved challenging due to its potential reliance on two sources of primary production—freshwater and marine—with distinct δ15N baselines and macromolecule compositions, thus differing in nutritional quality. Changes in δ15N could result from shifts in TP within the habitat, alterations in source reliance, variations in TDF, or changes in the δ15N baseline. For instance, the excretions of returning migratory fish are known to enrich freshwater ecosystems with 15N-enriched marine nitrogen (Feddern and others 2019).

Instead, we used the AA method of TP estimate, based on the δ15N difference between trophic AA and source AA (Bowes and Thorp 2015). Specifically, some AAs defined as trophic, such as Glx, are involved in nitrogen metabolism and show consistent and large TDF with deamination flux, reflecting trophic steps (Goto and others 2018). Other AAs, defined as source, such as Phe, show little trophic enrichment and represent the δ15N value of primary producers (Chikaraishi and others 2014; O’Connell 2017). We calculated TP as follows (Chikaraishi and others 2009, 2014):

$$ {\text{TP}} = (\Delta^{{15}} {\text{N}}_{{\text{Glx}} - {\text{Phe}}} - \beta_{{\text{Glx}} - {\text{Phe}}} )/{\text{TDF}}_{{\text{Glx}} - {\text{Phe}}} + {1} $$

where TDFGlx-Phe and βGlx-Phe represent the trophic discrimination factor (TDF) of Glx over Phe in consumers, and the isotopic offset in primary producers, respectively. We used a TDFGlx-Phe of 5.7‰, which is a mean value inferred empirically from marine fish that showed high predictability with fish stomach content (Bradley and others 2015), and βGlx-Phe of 3.6‰, which is a mean value for aquatic producers (Ramirez and others 2021). We calculated TP standard deviation only considering the propagation of uncertainty of measurement error because we were mostly interested in relative TP among char rather than uncertainty regarding βGlx-Phe and TDFGlx-Phe values.

Mixing Models of Marine Subsidy Reliance

We made all analyses in R (R Core Team 2020). We used Bayesian mixing models in MixSIAR (Stock and others 2018) to estimate fish reliance on either freshwater sources or marine subsidies. We implemented a main model based on bulk δ13C for which we had the most data. As eggs (the marine subsidies) are rich in fat (mean C/N molar: 5.2), the correction for lipid 13C -depletion might bias the results as some dietary lipids can be routed for the synthesis of amino acids (Arostegui and others 2019). To test for such bias, we implemented two models, with and without lipid corrections. Finally, we implemented a model based on the EAA fingerprinting to test the robustness of our conclusions and to verify whether char truly assimilated proteins derived from marine sources.

In brief, we used the residents above barriers and migratory char as end-members for freshwater and marine production. To account for δ13C fractionation in eggs, we extrapolated our fish dataset of mature individuals by examining the relationship between maternal muscle and egg δ13C, for both lipid-corrected and raw data (Appendix 3). For the fingerprinting model, we used LD1 as a marker that differentiated between marine and freshwater sources (see description in the EAA fingerprinting section). This method was proven successful to maximize source separation, and to provide source reliance estimates from fingerprinting (Fox and others 2019; Manlick and Newsome 2022). We considered individuals as “random” factors because we hypothesized that individuals randomly deviate from the mean (Semmens and others 2009) and we were interested in the individual component to infer trophic diversity. We did not include other factors (prior information), as we wanted to compare the model estimates (which originate from two different datasets), and as individual resource utilization is closely reflected in the fingerprinting.

We set the models using “long” run, meaning 300,000 iterations (burn-in = 200,000) on three parallel Monte Carlo Markov chains with a thinning interval of 300 using non-informative priors. We assessed model convergence using Gelman–Rubin (Gelman and others 2014) and Geweke diagnostics (Geweke 1991).

A full description of the models can be found in Appendix 3.

Generalized Linear Models

First, we asked how marine subsidies were transferred to fish. We used a generalized linear model (glm), fitting trophic position to a set of explanatory factors: marine dependence, river type, and their potential interaction. For this model, we only used the subset of fish for which we had AA isotope data.

Second, we asked how glacial meltwater influenced the resident char population’s reliance on marine subsidies. To compare the distribution of size and marine reliance between the two subsidized tributaries, we used nonparametric Mann–Whitney-Wilcoxon Test. We calculated the overall reliance of char populations as the body mass arithmetic weighted mean of individual reliance.

Lastly, we asked how glacial meltwater affected the size at which char started to rely on marine subsidies. We used a generalized linear model (glm) with a logistic link, fitting marine subsidy utilization (as deduced by the lipid-corrected bulk mixing model) to a set of explanatory factors. We included as factors: body size (standard length), river type (proglacial or tundra), their potential interaction, and day-of-year as fish were collected at different months. To visualize how river type and body size affected the degree of resource specialization, we estimated the 95% confidence interval of a local polynomial regression (Weisberg 2005), fitting marine reliance to body size. To do so, we used the loess.sd() function of the package msir (Scrucca and Scrucca 2016), that we fitted for the two river types. We constrained the estimated confidence interval between 0 and 1.

For the glm analyses, we selected the best approximating model based on the Akaike information criterion (Burnham and others 2002), retaining the model with the lowest AIC and considering all models with ΔAIC < 2. We computed model effects using the package sjPlot (Lüdecke and Lüdecke 2015).

Results

Characterization of Resource Use by Bulk Isotopes and Essential Amino Acid Fingerprinting

All fish from the different habitats had distinct δ13C (Figure 2B, pairwise t-test, all p < 0.001), except for subsidized residents in tundra river and migratory juveniles (p = 1.00). Bulk δ 13C and δ 15N distinguished freshwater residents (Resident above barriers, δ13C = -25.8 ± 1.0 ‰, δ15N = 7.7 ± 1.4 ‰) and marine adult feeders (δ13C = -18.1 ± 0.7, δ15N = 14.4 ± 0.8 ‰). Subsidized residents had intermediate δ13C, overlapping with marine feeders (δ13C up to -17.5‰). Freshwater residents above barriers had distinctive δ15N (pairwise t-test, all p < 0.001). With migration access, residents in tundra river and migrants had similar mean δ15N (pairwise t-test, all p > 0.05), while only residents in proglacial river had higher δ15N than migratory adults (p < 0.05) and juveniles (p < 0.01).

There were nine char in the subsidized tundra river which had very low δ13C (down to -33‰), and lower δ15N, indicative of a distinct primary production source (Figure 1D). Analysis of invertebrates from the stream also revealed some “light” source of carbon in tributaries (simuliidae larvae δ13C ~ -37‰), suggesting that those fish recently moved from a side arm of the main tributary. We further excluded those fish from further analysis, as they were not representative of the sampled tributary.

Comparison between char EAA fingerprinting and literature data supported the reliance on algae (Figure 3A), with predicted reliance > 95%, except for resident controls that showed some bacterial reliance ~ 5–10%, while terrestrial plants were negligible (Figure S2). Further, freshwater feeders were characterized by having relatively 13C-enriched Met, and 13C-depleted Val, Ile and Thr, compared to marine ones (Figure S3, Kruskal–Wallis test, p < 0.05). The linear discriminant analysis distinguished between freshwater and marine sources (Figure 3B, C, Kruskal–Wallis test, p < 0.001, Table S2), and its score scaled linearly with bulk δ13C in subsidized char (Figure 3D).

Figure 3
figure 3

Characterization of source reliance by essential amino acid fingerprinting and bulk carbon isotopes A, Linear discriminant analysis (LDA) of primary producers (algae, bacteria and plants) based on literature values of normalized δ13C of five essential amino acids (EAAs): Ile, Leu, Thr, Phe and Val. EAA positions show their relative contribution to LD axis. Circles show char (see legend for group assignment). Ellipses show the 50% confidence interval of group estimates (A and B). B, LDA of char end-members (fully freshwater and marine feeders) based on normalized δ13C of six EAAs (Ile, Leu, Met, Thr, Phe, Val). The bar plot illustrates the EAA coefficients in LD1. C, LD scores in char. Letters show significant differences based on Kruskal–Wallis test, with post-hoc comparison using the criterion Fisher’s least significant difference. D, Relationship between bulk carbon isotopes, corrected for lipid depletion (y-axis), and LD1 (x-axis, see B). The lines show linear regression for tundra and proglacial rivers, in a subsidized context.

Full size image

Transfer of Marine Subsidy by Cannibalism

The three mixing models, based on raw δ13C, lipid-corrected δ13C, and the EAA fingerprinting gave consistent outputs of marine reliance for residents with migration (linear regressions, p < 0.001, Figure S4). The largest offset was observed between the raw and lipid-corrected δ13C models for fish with high marine reliance (from the raw value model). Both bulk models were consistent with the fingerprinting model, with mean offset of -5% and 6% of marine reliance for the lipid-corrected and raw model, respectively. For modeling marine dependence, we used the lipid-corrected model.

Based on AA δ15N, we found a wide range of TP, from ~ 3 to ~ 5.5. All resident fish above barriers had TPs between 2.9 and 3.8. However, in a subsidized context, resident char had higher TPs with increasing size (Figure 4A). The largest resident char in the partially migrating population had the highest TP of ~ 5.6, in both tributaries, which corresponded to approximately one TP above the large migratory adults.

Figure 4
figure 4

Transfer of marine subsidies through cannibalism on migratory fish eggs A, Body size-based (x-axis) relationship to trophic position (y-axis), as inferred from the amino acid δ.15N method (Glx-Phe). Lines show locally weighted scatterplot smoothing (loess) for resident control (dotted line), and in a subsidized context (full line). B, Relationship between marine reliance (x-axis, fingerprinting model) and trophic position (y-axis), as inferred from the amino acid method. The line shows the linear regression between the two. Vertical bars show standard deviation of trophic position estimates. Horizontal bars in (B) show the 95% confidence estimates of mixing models. The colors show the fish cohorts (see legend)

Full size image

This increase in TP in subsidized fish was best fitted to marine subsidy utilization, (β = 1.59, 95% CI [1.26, 1.92], t(28) = 9.47, p < 0.001, R2 = 0.76), (Figure 4B, Table S3). The model with interactive effect of tributaries (tundra or proglacial) was supported by an AIC lower by 1.2 (Table S3), although this effect was not significant (p = 0.13). Altogether, the data strongly supports a direct assimilation of marine subsidies through egg cannibalism, with some fish showing a strong reliance > 95% on marine subsidies.

Effect of Glacial Water on the Magnitude of Marine Subsidy Reliance

We found a disparity in the distribution of marine subsidy utilization between the tundra and proglacial tributaries (Mann–Whitney-Wilcoxon Test, p < 0.001, Figure 5A), while the overall size distribution was not different (p = 0.37, Figure S5). We quantified that the overall char populations derived 83% and 48% of their proteins from marine subsidies in the proglacial and tundra tributaries, respectively.

Figure 5
figure 5

Effect of glacial meltwater on total and size-based dependence of fish on marine subsidies A, Distribution of marine reliance estimates in the two tributaries. Colors refer to the relative reliance on marine subsidy (blue) and freshwater (red) resources. B, The bottom panel shows the relationship between marine subsidy reliance (y-axis) and char body size (x-axis), in tundra (dark green) or proglacial (light green) streams. Dots show individual estimates from bulk mixing model. Lines with shaded area show the mean estimates and associated 95% confidence intervals from logistic regression. The top panel shows the degree of individual specialization resources as depicted as the 95% range fit of the regression. C, Conceptual diagram of the size-based dependence of char on freshwater resources or marine subsidies. In both rivers, char start relying on freshwater production only. With increasing size, some individuals start relying on marine subsidies, following a logistic model (white line). There is a specific body size for which specialists on either freshwater or marine co-exist, as illustrated by the char pictures. The pictures show examples of char of similar size, which exclusively utilize (> 95%) either freshwater or marine resource. In proglacial river, char start relying on marine subsidies at a smaller size (see B and pictures in C). As a result, most fish growth is derived from marine subsidies (see A).

Full size image

The utilization of marine subsidies was best predicted by fish body size (β = 1.79, 95% CI [0.97, 2.69], p < 0.001), along with the additive effect of river type (β = 0.02, 95% CI [0.01, 0.03], p < 0.001), (Tab. S3). The model choice was not affected upon which mixing model outputs we chose (that is, raw or lipid-corrected δ13C, Figure S6). When estimating the range of resource use, we found that two distinct strategies—fully marine or freshwater dependent—coexisted only at specific size thresholds, approximately 90 mm and 170 mm standard length in the proglacial and tundra tributaries, respectively (Figure 5B). Despite distinct ecological strategies, char of similar sizes did not exhibit observable differences in phenotype (see example in Fig. 5C).

Discussion

Our results support a strong marine subsidy in the form of eggs, which sustains resident char individuals in partially migrating populations. This subsidy is particularly important in rivers originating from glacier melting, confirming the prediction that subsidies are of higher magnitude in oligotrophic systems (Polis and others 1997; Flecker and others 2010). Consequently, the importance of subsidies can be strengthened by climate change if it lowers the productivity of the recipient ecosystem (that is, the freshwater) and strengthens the asymmetry, defined as an imbalanced distribution of resources, in comparison to the donor ecosystem (that is, the marine).

Isotopes of Amino Acids Characterize the Utilization of Marine Subsidies

Estimating resource use and trophic position (TP) through bulk stable isotopes alone proved challenging due to the intricate effects of true trophic shift and/or resource shift. For instance, it was not clear for the resident char individuals with access to marine-derived nutrients if the bulk δ13C increase reflected feeding in the marine environment (for example, intertidal zone), if the bulk δ15N increase reflected nutrient enrichment and/or cannibalism, or the use of terrestrial prey (Davidsen and others 2023). These challenges are exacerbated in arctic environments with strong seasonality, leading to substantial variations in isotopic baselines. In addition, by the time of sampling in summer, it was technically difficult to sample invertebrates in the turbid and fast flowing proglacial stream, and during the rest of the year, the site is relatively isolated and harder to access for sampling.

In light of difficulties sampling diets directly throughout the year, we employed an essential amino acid (EAA) fingerprinting approach to characterize resource use of individuals (Larsen and others 2013). As the EAA fingerprinting is conserved up the food chain (Whiteman and others 2019; Manlick and Newsome 2022), we could also use fish themselves as integrative baselines to characterize resource use, instead of using all potential prey across the year. This method has proven successful in environments where isolation and characterization of baselines is difficult, such as in reef ecosystems (Skinner and others 2021). Here, we could use fish above barriers and migratory ones as baselines for freshwater production and marine subsidy, respectively. Using literature values, we confirmed the high resemblance with algae fingerprinting, although there was some potential minor ~ 5–10% influence of bacterial de novo synthesis in freshwater fish. Crucially, the predicted algal origin supported the use of discrimination factors in AAs for comparing trophic positions (Ramirez and others 2021).

Freshwater and marine sources exhibited significant differences in fingerprinting for four essential amino acids: Isoleucine, Threonine, Methionine, and Valine. Notably, the freshwater diet was characterized by a relative 13C-enrichment in Methionine, a pattern that was consistent in proglacial and tundra rivers, suggesting that glacial meltwater did not affect the backbone of freshwater production. Methionine is often overlooked in the literature (McMahon and others 2010). Here, we achieved sufficient peak intensity and separation, attributing this success to its high abundance in fish and the use of NACME derivatives (Corr 2007). As we found that both fingerprinting were similar to those of algae, this pattern suggests a main difference in taxonomic composition between marine phytoplankton and stream biofilms, from which the fingerprinting arises. Although the underlying cause of this fingerprinting is unknown, it holds promise for distinguishing between marine and freshwater algae sources. In this context, our method proved particularly relevant for the goal of estimating reliance on either freshwater or marine resources, thus providing support for the framework of using fish as integrated baselines.

This fingerprinting was strongly correlated with bulk δ13C in resident char individuals, confirming that the change in bulk δ13C was truly indicative of the assimilation of proteins of marine origin. Furthermore, we showed that mixing models based on the fingerprinting and bulk stable isotopes yield similar outputs, providing some of the first empirical support that these two methods can be comparable. It is essential to note that these two metrics are not necessarily correlated. On the one hand, the fingerprinting represents EAAs that are directly routed from primary producers. On the other hand, bulk δ13C includes carbohydrates and non-EAAs that can be of lipidic origins (Newsome and others 2014; Takizawa and others 2020).

This dietary routing of lipids becomes particularly significant to consider when the putative source contains a high lipid content, such as eggs, as fish can utilize these lipids to synthesize certain non-EAAs (Arostegui and others 2019). To address this concern, we employed two bulk isotope models, one incorporating lipids and the other without. Both models provided estimates, but the model incorporating lipids tended to overestimate marine reliance compared to the one without lipids. This suggests that some marine lipids are not assimilated into the non-lipidic fraction of char. Instead, they may serve as an energy reserve, enhancing survival during periods of starvation (Ruyter and others 2000). Crucially, we confirmed that the choice between the two models did not alter the main conclusion regarding the differences between tundra and proglacial river systems. To further address the metabolic routing of lipids in fish tissue, the use of stable isotopes in fatty acids (Twining and others 2020) or other AAs, notably glycolytic AAs, hold promise (Wang and others 2019).

Overall, our findings support the use of EAA fingerprinting to trace organic matter flows in ecosystems, and offer a useful tool for meta-ecosystem research where baseline assessments of organic matter fluxes are difficult (Elliott Smith and others 2021; Arsenault and others 2022; Vane and others 2023). In our case, the EAA fingerprinting showed that our estimates of marine reliance, as inferred by bulk δ13C, authentically reflected the assimilation of marine subsidies.

Egg Cannibalism is an Effective Mechanism for the Transfer of Marine Subsidies

Based on the AA method, our investigation revealed a robust correlation between marine subsidy reliance and an elevation in trophic position (TP), pointing to cannibalism on migratory fish eggs. TP estimates were derived using a widely accepted discrimination factor of 5.7 ‰ (Bradley and others 2015), which yielded ecologically sound values. Resident control char and large migratory char were secondary and ternary-quaternary consumers, consistent with diet on aquatic larvae and marine fish (Dennert and others 2016). Accordingly, we found that resident cannibalistic char had exceptionally high TPs (up to 5.6), among the highest for aquatic ecosystems, especially in streams (Jake Vander Zanden and Fetzer 2007). Those TP estimates were derived from high values of Δ15NGlx-Phe, up to 29.8 ± 0.3 ‰, similar to those of large predatory fish in the ocean, such as swordfish (Choy and others 2015). Such high TP is not expected for a small and oligotrophic ecosystem (Ward and McCann 2017), except in the context of subsidies in meta-ecosystems (Perkins and others 2021).

The mechanism governing the transfer of subsidies plays a pivotal role in determining their impact (Flecker and others 2010). For instance, although migratory salmon can bring a large amount of nutrients in the form of ammonium excretions, they were shown to decrease nutrient availability in the recipient ecosystem, due to bioturbation (Holtgrieve and Schindler 2011). This transfer of ammonium can be challenging to quantify using isotopes (Feddern and others 2019), but the enrichment in bulk and source δ15N for freshwater production with subsidies compared to controls above migration barriers suggest some use of marine nitrogen by stream algae. Although invertebrates have the ability to directly assimilate marine subsidies, particularly through scavenging on carcasses or eggs (Chaloner and Wipfli 2002), our data did not support the efficient transfer of these pathways up the food chains. In fish, we found that the freshwater energy channel overlapped in production, as indicated by δ13C and EAA fingerprinting.

Here, we found that egg cannibalism was a strong mechanism for the transfer of marine subsidies, consistent with previous work (Childress and McIntyre 2015). Arctic char are known to be generalists in their diet (Klemetsen and others 2003; Klemetsen 2013; Moccetti and others 2019), and egg cannibalism is a common behavior (Pereira and others 2017; Frye and others 2021), illustrated in the paradigm that “all char can be cannibalistic” (Svenning and Borgstrom 2005). This ability to benefit directly from an abundant subsidy makes arctic char a keystone species, transferring nutrients into freshwater ecosystems.

This transfer of subsidies is important as it signifies the assimilation of nutrients into fish muscle tissues. Examining stomach contents of sampled resident char revealed a consistent diet in summer, primarily comprising aquatic larvae, with a minor presence of terrestrial invertebrates or char juveniles (Sperlich 2020). Egg cannibalism, observed during late sampling (beginning of October), indicates substantial retention of marine subsidies from approximately 10 months prior to sampling, aligning with findings in egg cannibalism by Dolly Varden (Salvelinus malma) (Jaecks and Quinn 2014). Such retention highlights char’s ability to rapidly accumulate biomass over a short period while maintaining a low turnover rate of muscle proteins over an extended duration. Another non-exclusive hypothesis is that char cannibalize emerging char juveniles in the spring, therefore acting as a secondary subsidy pulse, lasting over a longer period of time.

Glacier Melting Influences the Magnitude of Marine Subsidy Utilization

Theory of meta-ecosystem functioning predicts that nutrient availability in the recipient ecosystem determines the magnitude of subsidy use (Subalusky and Post 2019). Given the oligotrophic nature of proglacial environments (Zah and others 2001), we predicted that char would rely more intensively on marine subsidies. Our study system provided a natural experiment to test this prediction within a watershed that has both tundra and proglacial rivers. We found that glacial meltwater decreased three-fold the reliance of char on freshwater production. Notably, the majority of proglacial fish tissue was attributed to marine subsidies (83%), affirming the profound impact of recipient ecosystem characteristics on cross-ecosystem fluxes (Polis and others 1997).

While glacier meltwater increased the total reliance on marine subsidies, this trend was not universally applicable to all individual fish in our study. The statistical support for the additive effect of size and river type on subsidy reliance indicated that the proglacial environment did not alter the slope of the size-based dependence on marine subsidies. Instead, it shifted the body size threshold at which char began relying on marine subsidies (Figure 4C). In particular, there was a specific size at which multiple strategies could exist (that is, almost entirely marine or freshwater resource use), illustrating the size-based metabolic constraints on resource use (Finstad and others 2006). Given that we did not find differences in size structure between the tundra and proglacial rivers, the change in size-based dependence resulted in more individuals relying on marine subsidies in the proglacial river, with estimated 83% of proteins of marine origins. This estimate, however, overlooks juvenile char < 5 cm that could not be sampled, which are also important for population dynamics but whose total biomass is a marginal part of the total population at the time of sampling.

Potential Implications of Glacier Melting for Cross-Ecosystem Linkages

In partially migratory populations, theory suggests that the decision to migrate represents a fundamental fitness trade-off that considers the risk and growth in the resident and migratory habitats (Lundberg 1987). Resource availability in the resident environment crucially influences migration proportions (Olsson and others 2006) and, consequently, the amount of subsidies (Saboret and others 2021), illustrating feedback loops between the causes and consequences of migration (Peller and others 2023).

With glacial inputs, we found a tighter coupling between freshwater and marine ecosystems. First, juveniles depend more on marine resources. As the carrying capacity for juveniles is more coupled to marine subsidies in glacial streams, such population dynamics might be more vulnerable to changes in the migratory pattern (Joy and others 2020). Second, individual char in proglacial environments have a stronger predation effect on anadromous eggs, which ultimately influences the total number of returning fish and the marine subsidies they bring back. Thus, proglacial resident char might be more vulnerable to changes in the marine environment.

The impact of warming on glacier meltwater discharge varies across different regions worldwide, contingent upon whether the system has reached its “peak water” stage, representing the maximum proglacial influence (Huss and Hock 2018). The proglacial drainage in our study shows a net trend toward disconnection from glacial influence, with most glacial retreat occurring during global deglaciation (before 11,000 years ago), suggesting this may represent “peak water,” and with relatively stable glacial positions over the last ~ 11,000 years. However, paleo-reconstructions in the drainage suggest multiple periods of smaller glacial retreat and expansion over the last ~ 10,000 years (Larsen and others 2011), and modeled discharge data for the last 50 years indicate a continued increase in proglacial influence, suggesting repeated “local peak water” states superimposed on broader trends. These observations from the Qinngua valley align with the broader pattern of increased glacial discharge observed throughout Greenland (Hanna and others 2008; Bamber and others 2012). Changes for salmonids are further influenced by additional factors such as changes in forest cover and water temperature, yielding a complex array of predictions for salmonid populations (Pitman and others 2020). Altogether, it emphasizes the need to integrate global changes in glacial influence but also consider local factors.

Overall, we propose that the effects of glacial melting extend beyond the immediate habitats of streams and fjords, with potential cascading impacts on the entire ecosystem. As we look toward the future, the coming century presents new threats that further jeopardize ecosystems. These threats include, among others, intensified fishing pressure in the marine environment (Jackson and others 2001), alterations of species community (for example, the substitution of arctic char by brown trout, (Svenning and others 2022), and climate change (Briner and others 2016). For instance, overfishing in the marine habitat (Thériault and others 2008) or increased stream temperatures (Morita and others 2014) are both predicted to decrease migration propensity. Such effects might act in synergy with glacier melting and threaten the existence of native salmonid populations. Our findings underscore the necessity of integrating water changes in the meta-ecosystem framework.

Ethics

This research was done under research permits G18-013, G19-002, G21-006 and G22-057 from the Naalakkersuisut (Government of Greenland).