Integrative assessment of amino acid nitrogen isotopic composition in biological tissue samples determined by GC/C/IRMS, LC × EA/IRMS, and LC × GC/C/IRMS

Compound‐specific isotope analysis of nitrogen (δ15N) in amino acids (CSIA‐AA) has significantly contributed to environmental sciences such as anthropology, biogeochemistry, and ecology. Several methods exist for determining δ15N of amino acids (AAs). Although these methods have their own strength and weakness, they have not been intercalibrated yet, especially for biological samples with matrices. To address this issue, we systematically compared AA δ15N values among three methods using gas chromatography/combustion/isotope ratio mass spectrometry (GC/C/IRMS), preparative liquid chromatography (LC) separation followed by elemental analyzer/IRMS (LC × EA/IRMS), and LC separation followed by GC/C/IRMS (LC × GC/C/IRMS). The δ15N values of glutamic acid (δ15NGlu) and phenylalanine (δ15NPhe) in fish muscle, two crucial AAs for estimating the trophic positions (TPs) of organisms, were compared among methods. Although a significant difference in fish muscle δ15NGlu values was found among the three analytical methods, their δ15NGlu and δ15NPhe values were fairly consistent between all pairs of methods (n = 8, R2 = 0.9968 for GC/C/IRMS vs. LC × GC/C/IRMS; 0.9936 for LC × EA/IRMS vs. LC × GC/C/IRMS; and 0.9912 for GC/C/IRMS vs. LC × EA/IRMS), which resulted in similar TP estimates among the methods. Thus, the results provide empirical validation that the CSIA‐AA is comparable among different methods in interdisciplinary research fields. We also highlighted some critical features of each of the three analytical methods that can be used as a guideline for future CSIA‐AA research.

where β is the initial difference (‰) between δ 15 N Glu and δ 15 N Phe found in the primary producers (i.e., À3.4 AE 0.9‰ for aquatic algae and cyanobacteria). TDF Glu and TDF Phe are the offset of the trophic discrimination factors (‰) for δ 15 N Glu (+8.0 AE 1.2‰) and δ 15 N Phe (+0.4 AE 0.5‰), respectively (Chikaraishi et al. 2009a). In previous studies, gas chromatography (GC) (Chikaraishi et al. 2009a) or high-performance liquid chromatography (HPLC) (Broek and McCarthy 2014) are typically used to separate individual AAs extracted from environmental specimens for subsequent δ 15 N measurements in each AA using isotope ratio mass spectrometry (IRMS).
In the former method, hydrolyzed AAs are derivatized by esterification of carboxyl groups and acylation of amino groups before being introduced into GC/combustion/IRMS (hereafter GC/C/IRMS). GC/C/IRMS can determine δ 15 N Glu and δ 15 N Phe and δ 15 N values for several other AAs per single injection. However, retention times of glutamic acid (Glu) and phenylalanine (Phe) derivatives (e.g., N-trifluoroacetyl/isopropyl or N-pivaloyl/isopropyl derivatives) on GC chromatogram (e.g., Agilent DB-5 or Ultra 2 columns) are relatively close to each other (Metges et al. 1996, although they might be more separable using other derivatization methods and/or other GC columns; Corr et al. 2007). Suppose the tail of a large Glu peak overlapped with a small Phe peak with a lower δ 15 N value using the N-pivaloyl/isopropyl derivatization and the Agilent Ultra 2 column. In that case, δ 15 N Phe may be overestimated due to partial coelution of Glu with a higher δ 15 N value, eventually resulting in underestimated TP values (Eq. 1). Therefore, the accurate determination of δ 15 N Glu and δ 15 N Phe values, whose peaks are completely separated at the chromatogram baseline, is critical to estimate the TPs of the focal organisms. This is particularly true for organisms in aquatic food webs where tertiary consumers (i.e., TP = 4) see a difference of more than 30‰ between the values of δ 15 N Glu and δ 15 N Phe (Chikaraishi et al. 2009a).
The other method separates and isolates underivatized AAs in preparative HPLC using a reversed-phase or mixed-mode column. The recovered AAs are then introduced into elemental analyzer/IRMS (hereafter LC Â EA/IRMS) (Tripp et al. 2006;Broek et al. 2013;Broek and McCarthy 2014;Swalethorp et al. 2020). Broek et al. (2013) compared δ 15 N Phe values determined by LC Â EA/IRMS and GC/C/IRMS and reported that the former yielded a smaller analytical error (1σ AE 0.16‰) than the latter (1σ AE 0.64‰). This conclusion was based on the conventional EA/IRMS system with sample amounts in the range of 7-10 μgN (Broek et al. 2013). However, when a marine apex predator (Harbor Seal) was examined using a small-scale (hereafter nano-scale, < 1.4 μgN) EA/IRMS system, its δ 15 N Phe value considerably differed by 2-3‰ compared to that from GC/C/IRMS. This resulted in a 0.3-unit difference in the TP estimates between the two methods (Broek and McCarthy 2014). Therefore, it is hypothesized that the Glu and Phe fractions recovered from HPLC deliver a significant amount of exogenous nitrogen (e.g., ambient ammonia) from the experimental procedure to the nanoscale EA/IRMS system. It remains unclear on how much the LC Â (nanoscale) EA/IRMS is sensitive to such exogenous nitrogen for CSIA-AA.
Furthermore, AAs hydrolyzed from biological samples often contain organic matrices, including amino sugars or nucleic acids, which are often inseparable from AAs by a single chromatography (Clarke et al. 1999). Swalethorp et al. (2020) reported that chitin coeluting with some AAs on the HPLC chromatogram makes AA purification difficult. These coeluting impurities potentially hamper the precise and accurate measurements of the δ 15 N values in AAs. Takano et al. (2015) found that the δ 15 N values of the authentic AA standard recovered from HPLC and derivatized for GC/C/IRMS measurements (hereafter LC Â GC/C/IRMS) were consistent (R 2 = 0.997) with those without the LC treatment (i.e., GC/C/ IRMS). However, differences in AA δ 15 N values for biological samples with and without the LC treatment have not been systematically compared (but see Ishikawa et al. 2018 that separated methionine from its impurity using LC Â GC/C/IRMS). LC Â GC/C/IRMS could provide more reliable δ 15 N Glu and δ 15 N Phe values than GC/C/IRMS if the coeluting impurities were significant and their δ 15 N were very different from δ 15 N Glu and δ 15 N Phe values. Furthermore, the LC Â GC/C/ IRMS approach is expected to reduce the contamination risk of underivatized blank nitrogen that can potentially affect the LC Â EA/IRMS approach.
Although GC/C/IRMS, LC Â EA/IRMS, and LC Â GC/C/IRMS have their own strength and weakness, which is of critical importance for determination of AA δ 15 N values (Silverman et al. 2022), they have not been intercalibrated yet. To address this issue, we aimed to systematically compare δ 15 N Glu , δ 15 N Phe , and TP values, three major outcomes for biogeochemical CSIA-AA studies, among the three analytical methods using biological samples containing matrices. Although the matrix effect on AAs would be more severe in other environmental specimens such as sediments or dissolved organic matter, the muscle of four marine demersal fish species was used as representative biological samples. This is because consumers with high TP are expected to be sensitive to chromatographic separation of the 15 N-enriched Glu and 15 N-depleted Phe combination. The results would provide fundamental benchmark for the future CSIA-AA studies where the analytical approach is expected to be more diversified.

Solvents and AA standards
Most chemical reagents used in this study (i.e., acetonitrile, ammonia water, dichloromethane, distilled water, hydrochloric acid, isopropanol, methanol, n-hexane, pivaloyl chloride, thionyl chloride) were purchased from Fujifilm Wako Pure Chemicals (Osaka, Japan). Nonafluoropentanoic acid (NFPA) was purchased from Tokyo Chemical Industry (Tokyo, Japan). Seventeen AA standard powders were purchased from Sigma-Aldrich for HPLC analysis. Glutamic acid and eight AA standard powders (with known δ 15 N values as mentioned later) were purchased from Fujifilm Wako Pure Chemicals and Shoko Science (Tokyo, Japan), respectively, for GC analysis.

Samples
This study used four fish species: Jelly eelpout Bothrocara tanakae; Pacific cod Gadus macrocephalus; Longarm grenadier Abyssicola macrochir; and Snubnosed eel Simenchelys parasitica collected from the western North Pacific (Ohkouchi et al. 2016;Ishikawa et al. 2021). One individual per species was used for analysis. A small piece of muscle near the dorsal fin was excised, freeze-dried, and defatted with methanol and dichloromethane (Ohkouchi et al. 1997). The samples (12-13 mg dry weight) were ground into homogenous powders using a tube mill (Tube Mill control, IKA, Staufen, Germany) and were stored at À20 C until further processing.

Separation and isolation by HPLC
The other AA subsample was dissolved into 0.5 mL of 0.1 mol L À1 HCl and injected into an HPLC system (1260 series, Agilent Technologies, CA, USA), and separated by the modified method of Takano et al. (2015) and Furota et al. (2018). We used a porous graphite carbon column (Hypercarb, 4.6 Â 150 mm, particle size 5 μm; Thermo Fisher Scientific, MA, USA) and a guard column (4.6 Â 10 mm, 5 μm; Thermo Fisher Scientific) with a column cooler (Cool Pocket, Thermo Fisher Scientific) stabilized at 10.0 C. Mobile phases were distilled water with 20 mmol L À1 NFPA (ion pair reagent) (solvent A) and acetonitrile (solvent B). The injection volume of the autosampler was set at 80 μL per run, and five injections were performed per sample. The solvent gradient for each run was linearly programmed as follows: 0 min (A: 100%, B: 0%) to 60 min (A: 40%, B: 60%), with a flushing time of 60-70 min using 60% B, followed by equilibration with 100% A for 30 min (Takano et al. 2015). The flow rate was maintained constant at 0.2 mL min À1 . The separation of Glu and Phe was monitored using a charged aerosol detector (Corona CAD Veo, Thermo Fisher Scientific). After screening the AA profile of hydrolyzed fish samples with reference to the AA standard mixture (Supporting Information Fig. S1; Fig. 2a), a fraction collector (Agilent Technologies) with a time-based trigger mode was used to separate and isolate the target Glu (retention time: 30.0-32.0 min) and Phe (retention time: 50.0-52.0 min). In addition, 2 min before and after the targets were also collected if there was a drift in the retention time. All of these fractions collected were freeze-dried to evaporate solvents, dissolved into 1 mL of 0.1 mol L À1 HCl, and 1 μL injected into the HPLC to confirm that only target AAs were detected (Fraction ii for Glu and Fraction iv for Phe, Fig. 2b and c).
The isolated Glu and Phe fractions for all species were each split into two subfractions. Half of the Glu subfraction was combined with the Phe subfraction for each species. The combined Glu and Phe samples (n = 4) were passed through cation exchange chromatography (CEX) using a resin (AG 50W-X8, 200-400 mesh, Bio-rad, CA, USA) following the method of Takano et al. (2010). It is reported that CEX does not affect AA δ 15 N measurements using GC/C/IRMS ). We used a 10% ammonium solution to recover AAs. The samples were then dried under N 2 gas flow and derivatized as mentioned above. To test the effect of CEX on GC/C/IRMS, an aliquot of the combined Glu and Phe fractions of G. macrocephalus was derivatized without the CEX treatment. We expected that method optimization of the entire wet chemistry pretreatments could gain the benefits of analysis after CEX procedure Takano et al. 2018;Blattmann et al. 2020). The other half of the Glu and Phe subfractions were not combined, freezedried to evaporate solvents, and independently measured δ 15 N values using nanoscale EA/IRMS as described later. δ 15 N measurement by GC/C/IRMS The δ 15 N Glu and δ 15 N Phe values of derivatized samples were determined using a Delta Plus XP isotope ratio mass spectrometer connected to a gas chromatograph (6890N, Agilent Technologies) via an interface (Conflo III, Thermo Finnigan) with combustion (CuO, NiO, and Pt wires in a microvolume ceramic tube) and reduction (reduced Cu wires in a microvolume ceramic tube) furnaces (Thermo Finnigan) . The Pv/iPr-derivatized AAs were injected with a programmable-temperature vaporizing (PTV) injector (Gerstel, Mülheim an der Ruhr, Germany). The PTV temperature program was as follows: 50 C (initial temperature) for 0.3 min, heated from 50 C to 350 C at a rate of 600 C min À1 , and held at 350 C for 10 min. The carrier gas (He) flow rate was controlled at a constant flow mode of 1.4 mL min À1 . The gas chromatograph oven temperature was programmed as follows: 40 C (initial temperature) for 3.0 min, heated to 110 C at a rate of 15 C min À1 , heated to 150 C at a rate of 3 C min À1 , heated to 220 C at a rate of 6 C min À1 , held at 260 C for 18 min, heated to 280 C at a rate of 60 C min À1 , and held at the final temperature for 10 min. The AAs were separated on a low polar column (Ultra 2, 0.32 mm Â 50 m, film thickness 0.52 μm; Agilent Technologies) before introducing them into the IRMS through combustion (950 C) and reduction (550 C) furnaces ). An isotopic reference mixture of nine AAs (i.e., alanine, glycine, leucine, norleucine, aspartic acid, methionine, glutamic acid, phenylalanine, and hydroxyproline), with δ 15 N values ranging from À26.6‰ to +45.6‰ (Indiana University; Shoko Science), was analyzed every 5-6 injections to confirm the reproducibility of isotope measurements. Three and two pulses of the reference N 2 gas were measured for calibration at the beginning and the end of each run, respectively (Supporting Information Fig. S7). Using Isodat 3.0 software (Thermo Finnigan), peak detection parameters were set to the following: start slope of 0.2 mV s À1 , end slope of 0.4 mV s À1 , minimum peak height of 50 mV, peak resolution of 95%, and maximum peak width of 500 s. The δ 15 N values of all samples were corrected using the regression line between the published δ 15 N values and the measured δ 15 N values for the aforementioned nine AA standards (Ohkouchi et al. 2017). The standard measurement analytical errors (1σ) were lower than AE0.8 ‰ and covered the analyte size range of > 70 ngN. Triplicate δ 15 N measurements were performed per sample.
The PTV temperature program was as follows: 50 C (initial temperature) for 0.3 min, heated from 50 C to 350 C at a rate of 600 C min À1 , and held at 350 C for 10 min. The carrier gas (He) flow rate was controlled at a constant flow mode of 1.4 mL min À1 . The gas chromatograph oven temperature was programmed as follows: 40 C (initial temperature) for 3.0 min, heated to 110 C at a rate of 15 C min À1 , heated to 150 C at a rate of 3 C min À1 , heated to 220 C at a rate of 6 C min À1 , held at 260 C for 18 min, heated to 280 C at a rate of 60 C min À1 , and held at the final temperature for 10 min. The AAs were separated on a low polar column (Ultra 2, 0.32 mm Â 50 m, film thickness 0.52 μm; Agilent Technologies) before introducing them into the IRMS through combustion (950 C) and reduction (550 C) furnaces ). An isotopic reference mixture of nine AAs (i.e., alanine [Ala], glycine, leucine, norleucine, aspartic acid [Asp], methionine, glutamic acid, phenylalanine, and hydroxyproline), with δ 15 N values ranging from À26.6‰ to +45.6‰ (Indiana University; Shoko Science), was analyzed every 5-6 injections to confirm the reproducibility of isotope measurements. Three and two pulses of the reference N 2 gas were measured for calibration at the beginning and the end of each run, respectively (Supporting Information Fig. S7). Using Isodat 3.0 software (Thermo Finnigan), peak detection parameters were set to the following: start slope of 0.2 mV s À1 , end slope of 0.4 mV s À1 , minimum peak height of 50 mV, peak resolution of 95%, and maximum peak width of 500 s. The δ 15 N values of all samples were corrected using the regression line between the published δ 15 N values and the measured δ 15 N values for the aforementioned nine AA standards (Ohkouchi et al. 2017). The standard measurement analytical errors (1σ) were lower than AE0.8 ‰ and covered the analyte   size range of > 70 ngN. Triplicate δ 15 N measurements were performed per sample.

GC/MS measurements
To characterize the chemical structure of impurities found on the GC chromatogram, we used the G. macrocephalus sample and injected its (1) AA derivatives; (2) Glu and Phe derivatives after LC isolation with the CEX treatment; and (3) Glu and Phe derivatives after LC isolation without the CEX treatment to a gas chromatograph (6890N, Agilent Technologies). The gas chromatograph was coupled to a mass spectrometry detector (5973 MSD, Agilent Technologies) with a PTV injector (Gerstel) and a low polar column (Ultra 2, 0.32 mm Â 25 m, film thickness 0.52 μm; Agilent Technologies) using the modified method of Chikaraishi et al. (2009b). The PTV temperature program was as follows: 50 C (initial temperature) for 0.3 min, heated from 50 C to 350 C at a rate of 600 C min À1 , and held at 350 C for 10 min. The carrier gas (He) flow rate was controlled at a constant flow mode of 1. mL min À1 . The gas chromatograph oven temperature was programmed as follows: 40 C (initial temperature) for 4.0 min, heated from 40 C to 220 C at a rate of 8 C min À1 , heated to 260 C at a rate of 30 C min À1 , and held at the final temperature for 2.17 min. Ionization was performed based on the electron impact method to detect m/z in the range of 40-600 at 70 eV. By comparing the mass fragmentation patterns with the standard pattern (Chikaraishi et al. 2009b), we examined whether the impurities were by-products of the Glu and Phe derivatives or whether other molecules were contaminated during the experimental process. The δ 15 N Glu and δ 15 N Phe values of underivatized samples (i.e., Glu and Phe fractions separated by HPLC) were determined with a Flash EA1112 elemental analyzer connected to a Delta Plus XP isotope ratio mass spectrometer with a Conflo III interface (Thermo Finnigan) modified for nanoscale measurements (>0.08 μgN, Ogawa et al. 2010;Isaji et al. 2020). Samples were dissolved in approximately 20 μL of 0.1 mol L À1 HCl (final concentration: 0.05-0.1 μgN μL À1 ), placed in a smooth-wall tin capsule (pre-cleaned with methanol and dichloromethane), and completely dried on a hot plate (90 C). Triplicate or more measurements were performed per sample (n = 8, 0.5-2.7 μgN, Glu and Phe from four fish species). The dried capsule was injected by an autosampler to the oxidation (1050 C) and reduction (750 C) furnaces with a helium carrier gas (100 mL min À1 ). Two pulses of the reference N 2 gas were discharged at the beginning of each run. The δ 15 N data were calibrated using five inter-laboratory determined standards in the range of À5.70‰ to +61.3‰ (BG-T, BG-P, CERKU-01, L-glutamine, and L-valine; Tayasu et al. 2011, Indiana University, Shoko Science). The analytical errors (1σ) of δ 15 N values obtained by the repeated measurement of BG-T were lower than AE0.3‰ (n = 12, 0.6-2.5 μgN).

Trophic position calculations and analyses
To systematically assess uncertainties among GC/C/IRMS, LC Â GC/C/IRMS, and LC Â EA/IRMS, we calculated the standard deviations (SDs, 1σ) of repeated measurements of δ 15 N Glu and δ 15 N Phe values for the four fish species (n = 8). TPs were then calculated using mean δ 15 N Glu and δ 15 N Phe values determined by each of the three analytical methods using Eq. 1. Finally, the mean and SD of TP values obtained from δ 15 N Glu and δ 15 N Phe repeated measurements were used to calculate the propagation of error (1σ TP ) using the following equation (Bradley et al. 2015): where x i corresponds to δ 15 N Glu , δ 15 N Phe , β, TDF Glu , and TDF Phe . One-way ANOVA was used to test differences in δ 15 N Glu and δ 15 N Phe values determined by GC/C/IRMS, LC Â GC/C/IRMS, and LC Â EA/IRMS. [Correction added on August 30, 2022, after first online publication: The word "analysi(" has been deleted.] A Tukey HSD multiple comparison was applied when ANOVA showed significant (p < 0.05) difference. The statistical significance (p < 0.05) with and without CEX treatment was examined using the t-test. All statistics were run and the graphic was created using Matlab 2020a (Mathworks).

LC chromatograms
Using the analytical protocol established by Takano et al. (2015), the Glu and Phe of the four fish muscle samples were successfully separated from other AAs and unidentified impurities on the LC chromatogram (Fig. 2a). Peaks around 35 and 42 min were consistently observed on chromatograms even for standard and blank injections (Fig. S1). These peaks should be incombustible inorganic ions because the carbon and nitrogen of these peaks were not detected in the EA analysis (Furota et al. 2018). The fractions of Glu (Fraction ii, retention time 30.0-32.0 min) and Phe (Fraction v, retention time 50.0-52.0 min) were collected as single target compounds (Fig. 2b and c). The target peak was not detected by charged aerosol detector (CAD) in the fractions before Glu (Fraction i, retention time 28.0-30.0 min), after Glu (Fraction iii, retention time 32.0-34.0 min), before Phe (Fraction iv, retention time 48.0-50.0 min), and after Phe (Fraction vi, retention time 52.0-54.0 min) (Fig. 2b,c). This is important for the CSIA where the δ 15 N value of the focal compound often drifts from its front to tail within its peak as a consequence of chromatographic isotopic fractionation (Hayes et al. 1990;Hare et al. 1991;Merritt and Hayes 1994;Macko et al. 1997;Filer 1999;Broek et al. 2013;Isaji et al. 2020). Here, we appreciate a pioneering statement by John Hayes and co-workers in the 1990s for the importance of chromatographic baseline resolution in AA carbon isotope analysis: "When peaks overlap, the isotopically light tail of the first component underlies the beginning of the second peak, and the isotopically heavy front of the second peak underlies the end of the first peak" (Hayes et al. 1990). This has promoted AA isotope studies by combining online and offline methods (Hare et al. 1991;Minagawa et al. 1992;Metges and Petzke 1997), as demonstrated in the present study. Our chromatograms indicate that (1) the analytical settings were sufficiently stable with no drift in the retention time of the chromatogram, and (2) the entire target compound was successfully collected in the final Glu and Phe fractions.

GC chromatograms
The IRMS chromatograms around the Glu and Phe peaks were comparable between GC/C/IRMS and LC Â GC/C/IRMS (Fig. 3). Therefore, the GC/C/IRMS approach is likely Pv/iPr derivatized AAs extracted from the Gadus macrocephalus muscle sample. The first three and the last two peaks are reference N 2 gas. Asterisks denote major impurities. adequate, at least for the determination of δ 15 N Glu and δ 15 N Phe . However, isolating Glu and Phe using HPLC before derivatization made the GC/C/IRMS chromatogram much simpler than without HPLC (Fig. 3b). This is because the separation of LC prevents compounds other than Glu and Phe from being injected into the IRMS system. Unknown peaks were found on the GC/C/IRMS and LC Â GC/C/IRMS chromatograms around 25, 33, 36, 37, and 40 min (Fig. 3). Most of these peaks are assigned by GC/MS to the secondary products of the Pv/iPr derivatives of either Glu or Phe (Supporting Information Figs. S2-S7, except for Fig. S3 showing Asp Pv/iPr derivative).

δ 15 N and TP comparisons
A fairly good correlation was detected in δ 15 N Glu and δ 15 N Phe values (n = 8, R 2 = 0.9968) between GC/C/IRMS and LC Â GC/C/IRMS (Fig. 4a). The impurities coeluted with the Glu and Phe peaks that could not be removed by the LC Â GC, if present, are probably not important. The results confirmed that the GC/C/IRMS approach is appropriate for determining δ 15 N Glu and δ 15 N Phe values in biological samples with undesired matrices. The δ 15 N Glu and δ 15 N Phe values determined by LC Â EA/IRMS were also consistent with those by the other two methods (n = 8, R 2 = 0.9936 for LC Â EA/IRMS vs. LC Â GC/C/IRMS and 0.9912 for GC/C/IRMS vs. LC Â EA/ IRMS) (Fig. 4b and c).
One way ANOVA showed that δ 15 N Glu values of the four fish species were significantly different among the three methods (F > 6.84, p < 0.028). However, differences in mean δ 15 N Glu values among methods were not greater than the 2σ analytical error (AE1.6‰) of GC/C/IRMS. On the other hand, δ 15 N Phe values of the four fish species were not significantly different among the three methods (F < 2.63, p > 0.152) ( Table 1). In particular, no significant difference in δ 15 N Phe values between GC/C/IRMS (δ 15 N Glu and δ 15 N Phe determined simultaneously) and LC Â EA/IRMS (δ 15 N Glu and δ 15 N Phe determined separately) suggests that the effect of the Glu tail on the Phe peak is not important in terms of their δ 15 N values. The difference in the estimated fish TP (ΔTP) between each pair of GC/C/IRMS, LC Â GC/C/IRMS, and LC Â EA/IRMS (ΔTP range: 0.1-0.3, mean AE SD: 0.2 AE 0.1, n = 12) was not greater than their respective 1σ TP error propagation (1σ TP range: 0.4-0.7, mean AE SD: 0.5 AE 0.1, n = 12) ( Table 1). The results suggest that the three analytical methods examined in this study return consistent TP output. The TP values are consistent with those expected from their typical diet menu (i.e., large benthos and small fish) and those reported in the previous study (Ishikawa et al. 2021). All evidence suggests that these fish are carnivores. It should be noted that propagation of error is largely affected by the uncertainty of β, TDF Glu , and TDF Phe values rather than the SD of repeated δ 15 N Glu and δ 15 N Phe measurements.
The blank of the EA capsule is estimated to be 0.009 AE 0.003 μgN (Isaji et al. 2020). This is two orders of magnitude smaller than the nitrogen amount of Glu (0.9-2.7 μgN) and Phe (0.5-2.2 μgN) samples. Furthermore, the stationary phase of the HPLC column used in this study (i.e., Hypercarb: 100% porous graphite carbon) does not contain nitrogen. Therefore, the δ 15 N Glu and δ 15 N Phe values determined by LC Â EA/IRMS are rarely affected by blank nitrogen derived from the HPLC column and the tin capsule. However, the Phe fractions of G. macrocephalus and A. macrochir may contain considerable amounts of excess nitrogen derived from exogenous sources (e.g., ammonia and ammonium salt) because their reproducibility (i.e., 1σ of repeated measurements, AE3.0‰) exceeded the 2σ analytical error of EA/IRMS (i.e., AE0.6‰). The excess nitrogen can significantly impact the Phe fraction, the concentration of which is generally lower than the concentration of Glu in biological samples (typically less than half ). In addition, the final solution used in this study (i.e., 0.1 mol L À1 HCl)

Discussion
Most previous CSIA-AA studies have been implemented by using GC/C/IRMS (after McClelland and Montoya 2002;Chikaraishi et al. 2007;McCarthy et al. 2007;Popp et al. 2007) or LC Â EA/IRMS (Broek et al. 2013;Broek and McCarthy 2014;Swalethorp et al. 2020). The GC/C/IRMS has been well established as a fast and easy technique for CSIA research (Table 2). Also, its advantages include a small sample size (down to 20 ngN) and simultaneous δ 15 N measurements of multiple AAs. At the same time, its drawbacks are derivatization-related artifacts that can potentially generate by-products and destabilize end-products, and targets limited to derivatizable AAs (Table 2). Although the present study confirmed that it was of minor significance for biological samples, GC/C/IRMS is inherently sensitive to the effects of matrices and to a wide range in individual AA concentrations within a single sample on the chromatographic separation (Table 2). LC Â (conventional) EA/IRMS with a sample amount of over 7 μgN offers a smaller analytical error (1σ AE 0.16‰) than GC/C/IRMS (1σ AE 0.64‰) (Broek et al. 2013). The analytical error of LC Â (nano-scale) EA/IRMS that was used in this study (1σ AE 0.3‰ with 0.6-2.5 μgN) is still better than GC/C/IRMS. LC Â EA/IRMS is free from complex derivatization and can be applied to AAs such as arginine (Arg) and histidine (His) because their Pv/iPr derivatives are nearly below detection (Ohkouchi et al. 2017, but N-acetyl/isopropyl derivatization is amenable for Arg, see Kendall and Evershed 2020) (Table 3). However, LC Â EA/IRMS requires a large sample size (>50 μgN on conventional EA and >0.08 μgN on nanoscale EA) and holds the contamination risk of exogenous nitrogen (e.g., ambient ammonia), increasing with the decreasing sample size ( Table 2). The exogenous nitrogen would affect δ 15 N of the final target AAs, as demonstrated in low reproducibility of δ 15 N Phe values for G. macrocephalus and A. macrochir determined by our LC Â (nanoscale) EA/IRMS. Compared to the two methods above, LC Â GC/C/IRMS is more labor-intensive and time-consuming ( Fig. 1; Table 2). Similar to GC/C/IRMS, LC Â GC/C/IRMS is not applicable to Pv/iPr-derivatized AAs such as Arg and His. However, multiple AAs isolated from LC can be easily recombined with arbitrary concentrations and compositions (Table 2). This is particularly useful for biological samples such as protein, where Glu is originally much more abundant than Phe. This imbalance potentially causes the Glu peak tail to overlap with the following Phe peak. LC Â GC/C/IRMS has less contamination of exogenic nitrogen (e.g., ambient ammonia) than LC Â EA/ IRMS because the derivatization procedure eliminates all underivatized nitrogen compounds (Table 2). Although this depends on the columns, solvents, ramp programs, and derivatization methods, target AAs, which cannot be distinguished from other compounds on the LC chromatogram, may be separated on the GC chromatogram. For example, methionine and isoleucine coelute on LC but their Pv/iPr-derivatives are separable on the posterior GC (Ishikawa et al. 2018) (Figs. 2,  3). The reverse is also true: serine and threonine Pv/iPr  Fig. 3; Supporting Information Fig. S1). This is an excellent advantage over GC/C/IRMS and LC Â EA/IRMS, where the complete separation of the target peaks from others on the GC or LC chromatogram is required (Tables 2 and 3).
The LC Â GC/C/IRMS approach enables researchers to minimize the matrix effect because the target AAs are isolated by HPLC before derivatization (Table 2), which is also useful for the CSIA-AA of non-biological samples such as geochemical or cosmochemical samples. The sample amount is often limited in such situations, and the matrix effect is presumably more critical than biological samples. Since our research goal in this study was to test whether the TP values are comparable (i.e., not different within propagated uncertainties) among three analytical methods, the complexity of the sample matrix was out of our primary scope. Although we used the fish muscle with a less complex matrix, its δ 15 N Glu and δ 15 N Phe values were not completely identical among the three methods (Table 1). The results suggest that, if the sample matrix was one of the factors making AA δ 15 N values different from one method to another, a material that has a more complex matrix such as sediments, dissolved organic matter, chitinous tissues, or meteorites would require more careful attention to analytical settings, especially when their AA δ 15 N values are compared among different methods.
Finally, the CEX treatment prior to derivatization removes impurities associated with the HPLC process (Supporting Information Fig. S8). Without CEX treatment, the peak balance of Glu and Phe changed significantly, and at least four peaks with intensities sabove the peaks of Glu and Phe were observed (Supporting Information Fig. S8a). The mass fragment patterns obtained from the GC/MS analysis showed that the impurities are the derivatives of their respective parent molecules (i.e., Glu and Phe Pv/iPr) (Supporting Information Figs. S9-S12). Thus, it is speculated that the isopropyl groups of Glu and Phe isolated from HPLC can be replaced with methyl groups without CEX treatment. More importantly, both the δ 15 N Glu and δ 15 N Phe values determined by LC Â GC/ C/IRMS without CEX treatment were significantly (by $6‰) lower than those determined by LC Â GC/C/IRMS with CEX treatment (δ 15 N Glu , t = 14.7, p < 0.001; δ 15 N Phe , t = 7.4, p = 0.002; Supporting Information Fig. S13). The results suggest that these daughter compounds, which stemmed from the Pv/iPr derivatives of Glu and Phe, change their mother δ 15 N values because the isotopic shift was consistent for δ 15 N Glu and δ 15 N Phe values. Therefore, in future LC Â GC/C/ IRMS studies, it is recommended to perform CEX treatment after LC separation to accurately determine δ 15 N Glu and δ 15 N Phe values. It should be noted that the CEX treatment is not applicable to LC Â EA/IRMS because the 10% ammonium solution that recovers AAs is not removable during the dry-up step for EA/IRMS, which delivers a significant amount of exogenous nitrogen to the sample. Gly: glysine; Ser: serine; Ala: alanine; Hyp: hydroxyproline; Thr: threonine; Cys: cysteine; Asn: aspargine; Gln: glutamine; Asp: aspartic acid; Pro: proline; Glu: glutamic acid; Val: valine; Lys: lysine; Leu: leucine; Ile: isoleucine; Met: methionine; His: histidine; Arg: arginine; Phe: phenylalanine; Tyr: tyrosine; Trp: tryptophane. *Peak coelution might be avoided by optimizing GC analytical settings. † Hardly recovered after HCl hydrolysis. ‡ Peak might be detected by using other GC analytical settings (e.g., other derivatization methods and/or other GC columns). § Peak coelution might be avoided by optimizing LC analytical settings.