Rationale: Concerns exist over observed shifts in value and variance of nitrogen isotopes following physicochemical extraction of lipids from organic matter. The mechanisms behind these apparent changes in bulk tissue δ15N values are not fully understood yet have major implications for analytical costs and integrity of data interpretations.

Methods: Changes in proximate analysis, amino acid composition, C:N ratios, bulk tissue and amino acid δ13C and δ15N values, and resulting isotope‐based food web metrics were compared between lipid‐intact and lipid‐extracted muscle tissue of fishes spanning <1% to >20% muscle fat content to identify mechanisms of nitrogen isotope fractionation associated with physicochemical lipid extraction.

Results: Bulk δ13C and δ15N values increased and %N, C:N ratios and crude protein content decreased following lipid extraction. Resulting bulk isotope niche spacing and overlap varied significantly between lipid‐intact and lipid‐extracted tissues. While amino acid composition significantly changed during lipid extraction, particularly for lipid‐associated amino acids (e.g., Glu, Lys, Ser), individual amino acid δ13C and δ15N values, and their associated compound‐specific isotope analysis of amino acids (CSIA‐AA)‐based food web metrics, did not.

Conclusions: Physicochemical lipid extraction caused significant tissue composition changes (e.g., leaching of amino acids and 15N‐deplete nitrogenous waste) that affected δ13C and δ15N values and tissue %C and %N beyond simply removing lipids. However, lipid extraction did not alter individual amino acid δ13C or δ15N values or their associated CSIA‐AA‐based food web metrics.

Four species of marine fishes with varying levels of muscle total lipid content (Table 1) were purchased from Rhode Island fish markets: wild-caught Atlantic cod (Gadus morhua, hereafter “cod”) from southern George’s Bank, wild-caught bluefish (Pomatomus saltatrix, hereafter “bluefish”) from southern George’s Bank, wild-caught Atlantic herring (Clupea harengus, hereafter “herring”) from “the Northwest Atlantic”, and Norwegian farmed “organic” Atlantic salmon (Salmo salar, hereafter “salmon”). Dorsal muscle tissue was isolated from each fish fillet, frozen at -20 °C, and then lyophilized for 72 hrs prior to homogenization with mortar and pestle. Freeze-dried, homogenized samples were subdivided into two aliquots, one with original lipid content intact and one with lipids extracted. For the latter, lipids were removed using a modified version of the classic Bligh and Dyer method. Briefly, freeze-dried, homogenized samples were soaked for 30 min at room temperature in a chloroform:methanol solution (2:1v/v ratio) with solvent volume 5x sample volume, vortexed for 30 s, left undisturbed to settle for 30 min, and centrifuged for 10 min at 3400 rpm. The supernatant containing solvents and lipids was removed and the process was repeated (at least 3 full cycles) until the supernatant was clear and colorless following centrifugation. Samples were re-dried at 60 °C for 24 h to remove any remaining solvent.

Proximate analysis of crude protein, crude fat, crude fiber, crude carbohydrate, ash, and moisture and amino acid compositions (16 individual amino acids) were conducted on both lipid intact and lipid extracted muscle samples from all four species (n = 5 replicates per treatment per species) at the New Jersey Feed Laboratory, Trenton, New Jersey, USA.

For bulk stable isotope analysis, 1.0 ± 0.1 mg of each sample was analyzed for bulk δ¹³C and δ¹⁵N values using a PDZ Europa ANCA-GSL elemental analyzer interfaced to a PDZ Europa 20-20 isotope ratio mass spectrometer (IRMS; Sercon Ltd., Cheshire, UK) at the UC Davis Stable Isotope Facility, Davis, CA, USA. Data are reported with standard d notation in per mil (‰) units relative to Vienna PeeDee Belemnite and Air for carbon and nitrogen, respectively. Long term standard deviation is ± 0.2‰ for d¹³C and ± 0.3‰ for d¹⁵N analyses. For CSIA-AA, 2.0 ± 0.1 mg of each sample and a long-term laboratory working standard of Atlantic cod muscle were acid hydrolyzed to isolate free amino acids in 6N HCl at 110°C for 20 hrs in tubes under N₂ headspace to prevent oxidation during hydrolysis. Acid hydrolyzed samples, the working laboratory standard, and a laboratory amino acid reference standard containing all target amino acids with known isotopic values (Sigma-Aldrich) were spiked with an internal Nor-Leucine standard and then derivatized to N-trifluoroacetic acid isopropyl esters prior to CSIA-AA according to established protocols in McMahon et al. (2018).

Derivatized samples and standards were injected (240°C) in split mode (15:1 split, d¹³C analyses) or splitless mode (d¹⁵N analyses) onto a BPX5 column (60m length, 0.32mm ID, 1µm film thickness) for separation in a Thermo Trace 1310 GC, combusted into CO₂ or N₂ gas via a GC-IsoLink II oxidation/reduction furnace (1000°C), passed through a liquid nitrogen cold trap (d¹⁵N analyses only), and analyzed with a Thermo Scientific Finnigan Delta V Plus IRMS at the University of Rhode Island - Graduate School of Oceanography.

Standardization of runs was achieved using intermittent pulses of a CO₂ or N₂ reference gas of known isotopic composition. All CSIA-AA samples were analyzed in triplicate bracketed by the laboratory amino acid reference standard and the laboratory working standard to correct for the introduction of exogenous carbon (δ¹³C only) and kinetic fractionation associated with derivatization (δ¹³C and δ¹⁵N) following Yarnes & Herszage (2017). The long-term reproducibility of the laboratory working standard, δ¹³C ± 0.4‰ and δ¹⁵N ± 0.5‰, (calculated as the long-term SD across >100 separate full analyses, averaged across all individual AAs), provides an estimate of full protocol reproducibility (hydrolysis, derivatization, and GC-C-IRMS analysis). For δ¹³C analyses, 13 individual amino acids were measured and classified as essential (minimal trophic discrimination): isoleucine (Ile), leucine (Leu), lysine (Lys), methionine (Met), phenylalanine (Phe), threonine (Thr), valine (Val) or non-essential (variable trophic discrimination): alanine (Ala), aspartic acid + asparagine (Asx), glutamic acid + glutamine (Glx), glycine (Gly), proline (Pro), serine (Ser). For δ¹⁵N analyses, 14 individual amino acids were measured and classified as source (minimal trophic discrimination): arginine (Arg), Lys, Met, Phe; trophic (large trophic discrimination): Ala, Asx, Glx, Ile, Leu, Pro, Val; and metabolic (variable trophic discrimination): Gly, Ser, Thr.

Offsets in tissue %C and %N, C:N ratios, proximate analysis (protein, fat, fiber, carbohydrate, ash, moisture), amino acid composition, bulk d¹³C and d¹⁵N values, and amino acid d¹³C and d¹⁵N values were compared between lipid intact and lipid extracted muscle samples of the four fish species spanning a large range of tissue fat compositions. Total hydrolysable amino acid (THAA) d¹³C and d¹⁵N values, as proxies for the carbon and nitrogen isotope bulk protein pool, were calculated as the sum of individual amino acid d¹³C or d¹⁵N values, weighted by the relative composition of each individual amino acid (Ala, Asx, Glx, Gly, Ile, Leu, Lys, Phe, Pro, Ser, Thr, Val representing 64.6 ± 4.5% of the total protein pool; Table S1). d¹³C_THAA and d¹⁵N_THAA values were compared to bulk tissue d¹³C and d¹⁵N values, respectively, for lipid intact and lipid extracted tissues for each species via linear regression to assess patterns of changing protein vs. bulk tissue isotope pools.

To examine how changes in bulk and amino acid d¹³C and d¹⁵N values of fish muscle tissue due to lipid extraction impacted ecological interpretation of those data, a series of common isotope food web metrics were calculated from the same individual muscle samples with lipids intact and lipids extracted: 1) isotopic niche width and overlap using bulk tissue d¹³C and d¹⁵N values, 2) amino acid carbon isotope fingerprinting, and 3) trophic position. 

Changes in isotopic niche widths and overlap between lipid intact and lipid extracted samples of each fish species were calculated using bulk δ¹³C and δ¹⁵N data and the R package SIBER. Isotopic niche, which reflects the isotopic variation of an organism’s resource use (e.g., diet, habitat), is often used as a proxy for ecological niche width. Standard ellipse area (SEA), used as a metric of isotopic niche width, was calculated using both frequentist (SEA_C, sample size corrected) and Bayesian approaches (SEA_B) within SIBER. The Bayesian approach yields probability distributions of isotopic niche width that reflect SEA uncertainty by treating isotopic niche region as a probability density function with posterior estimates calculated by iteratively fitting multivariate normal distributions to each dataset via Markov chain Monte Carlo simulation. The model ran two chains of 20,000 iterations, burning the first 1,000 draws and thinning every 10 draws. To test for differences in isotopic niche size between lipid intact and lipid extracted bulk δ¹³C and δ¹⁵N data, I compared each pair of SEA_B posterior draws and quantified the proportion that were smaller (or larger) in magnitude. Differences in SEA_B were considered significant if ≥ 95% of posterior draws were smaller or larger than the other (lipid intact vs. lipid extracted for each species). I quantified the extent to which lipid intact and lipid extracted isotopic niches overlapped for each species using the maxLikOverlap function within SIBER. As it was hypothesized that the majority of the variation in bulk tissue isotope values following lipid extraction would come from changes in d¹³C values owing to removal of low ¹³C lipids, I also examined the impact that lipid extraction had on niche overlap using changes in d¹⁵N alone (i.e., d¹³C values held constant at pre-lipid extraction values). This test facilitated an assessment of niche overlap impact owing to changes in the non-target isotope N.

I quantified the relative contributions of potential primary producer end members to individual fishes (n = 5 individuals per treatment per species) using a fully Bayesian mixing model approach within the “Mix” - Stable Isotope Analysis in R (MixSIAR, Stock et al. 2018) package using essential amino acid δ¹³C values (Thr, Ile, Val, Leu, Phe), with species as a fixed factor and treatment as a random factor. I used published water column phytoplankton and benthic macroalgae as end members for cod, bluefish, and herring. For farmed salmon, I also included terrestrial plants as a potential end member given the increased use of plant-based aquaculture feed for farmed salmon. A low trophic discrimination factor of 0.1 ± 0.1‰ was selected to reflect the minimal fractionation that occurs between diet and consumer essential amino acids.⁴³ The model was run with an uninformative prior using three chains for 1,000,000 iterations, with a burn-in period of 500,000 iterations and a thinning interval of 500. Model convergence was verified using Gelman-Rubin (no variables >1.05) and Geweke (<5% of variables outside the 95% confidence interval) criteria. Resulting posterior probability distributions were used to estimate the proportional contributions of the primary producer end members to fishes as a means of testing whether potential alterations to individual amino acid d¹³C values altered resulting amino acid carbon isotope fingerprinting results. Note, this approach identifies the relative contribution of primary producer end members at the base of the food web supporting these upper trophic level fishes and does not reflect direct consumption of these end members.

I compared source amino acid d¹⁵N values, a metric of baseline nitrogen isotope dynamics, from lipid intact and lipid extracted muscle tissue via linear regressions for all species and source amino acids (Arg, Lys, Met, Phe). Similarly, I compared trophic positions of individual fishes from lipid intact and lipid extracted muscle tissue via linear regressions for all species. Trophic positions of individual fishes were calculated using a TP_CSIA-AA approach that is internally indexed to the δ¹⁵N baseline. Based on the results from the amino acid carbon isotope fingerprinting mixing models (Table S1), I used a fully non-vascular autotroph 𝛽 value (3.3 ± 1.8‰)⁵² for cod, bluefish, and herring along with a marine fish food web TDF_Glx-Phe (6.6 ± 1.7‰). For farmed salmon I used a 𝛽_mix value (-1.45 ± 2.6‰) reflecting 52% non-vascular autotroph 𝛽 (3.3 ± 1.8‰) and 48% vascular autotroph 𝛽 (-6.6 ± 3.4‰) and a composite TDF_Glx-Phe (7.1 ± 1.1‰) reflecting 52% marine fish food web TDF_Glx-Phe (6.6 ± 1.7‰) and 48% terrestrial plant TDF_Glx-Phe (7.6 ± 0.4‰) based on their amino acid carbon isotope fingerprinting mixing model data. Measures of error in fish δ¹⁵N_Glx and δ¹⁵N_Phe values from triplicate analytical analyses, 𝛽, and TDF_Glx-Phe were propagated through the trophic position calculation using R package “Propagate v.1.0-6”.