Ontogenetic δ15N Trends and Multidecadal Variability in Shells of the Bivalve Mollusk, Arctica islandica

Bulk stable nitrogen isotope values of the carbonate-bound organic matrix in bivalve shells (δ15NCBOM) are increasingly used to assess past food web dynamics, track anthropogenic nitrogen pollution and reconstruct hydrographic changes. However, it remains unresolved if the δ15NCBOM values are also affected by directed ontogenetic trends which can bias ecological and environmental interpretations. This very aspect is tested here with modern and fossil specimens of the long-lived ocean quahog, Arctica islandica, collected from different sites and water depths in the NE Atlantic Ocean. As demonstrated, δ15NCBOM values from the long chronologies show a general decrease through lifetime by −0.006‰ per year. The most likely reason for the observed δ15NCBOM decline is a change in the type of proteins synthesized at different stages of life, i.e., a gradual shift from proteins rich in strongly fractionating, trophic amino acids during youth toward proteins rich in source amino acids during adulthood. Aside from this ontogenetic trend, distinct seasonal to multidecadal δ15NCBOM variations (ca. 50 to 60 years; up to 2.90‰) were identified. Presumably, the latter were governed by fluctuations in nutrient supply mediated by the Atlantic Multidecadal Variation (AMV) and Atlantic Meridional Overturning Circulation (AMOC) combined with changes in nitrate utilization by photoautotrophs and associated Rayleigh fractionation processes. Findings underline the outstanding potential of bivalve shells in studies of trophic ecology, oceanography and pollution, but also highlight the need for compound-specific isotope analyses.


INTRODUCTION
The stable nitrogen isotope composition of organic materials serves as a powerful tool in trophic ecology, hydrography and pollution studies (e.g., Fry, 1988;Hobson and Welch, 1992;Ohkouchi et al., 2015;Doherty et al., 2021). For example, as the bulk δ 15 N value of tissues increases from prey to predator (Miyake and Wada, 1967;DeNiro and Epstein, 1981), on average, by 3.4 (Minagawa and Wada, 1984), the trophic position of consumers can be determined, and thus the food web structure and food chain length, provided that the nitrogen isotope baseline (=δ 15 N value of the primary producers) is known (Fry, 1988;Thompson et al., 1995;Post, 2002;Perkins et al., 2014). Tissue δ 15 N data can also help to unravel trophic interactions within food webs (Ishikawa, 2018) and to compute the proportion of different food items in the diet of a consumer, specifically when used in conjunction with other isotope systems (Phillips and Koch, 2002;O'Donnell et al., 2003). Furthermore, tissues register nitrogen isotope changes of total dissolved nitrogen in the water (TDN; δ 15 N TDN ) which enables reconstructions of biogeochemical cycles, water mass sources and anthropogenic nitrogen pollution etc. (e.g., Altabet et al., 1995;Voss et al., 2005;Ellis, 2012;Geeraert et al., 2020). With suitable δ 15 N archives it becomes possible to place all such ecological and environmental changes into precise temporal context, determine the pristine state of ecosystems prior to human perturbations and quantify the anthropogenic impact on ecosystems (e.g., Sherwood et al., 2011;Lueders-Dumont et al., 2018;Wang et al., 2018;Duprey et al., 2020), and test respective ecological models (Duce et al., 2008;Serna et al., 2010).
According to recent sclerochronological studies, shells of bivalve mollusks serve as suitable δ 15 N archives, because they provide high-resolution (annually and seasonally), temporally well-constrained records of past ecological and environmental changes Whitney et al., 2019;Das et al., 2021). Both bulk δ 15 N values of the periostracum (δ 15 N P ) and the carbonate-bound organic matrix of their shells (δ 15 N CBOM ) typically correlate strongly with δ 15 N values of contemporaneous soft tissues (LeBlanc, 1989;O'Donnell et al., 2003;Carmichael et al., 2008;Delong and Thorp, 2009;Watanabe et al., 2009;Kovacs et al., 2010;Versteegh et al., 2011;Graniero et al., 2016Graniero et al., , 2021Darrow et al., 2017), specifically the adductor muscles , and thus also with the diet of the bivalves (δ 15 N Diet ) (O'Donnell et al., 2003;Carmichael et al., 2004;Graniero et al., 2021). However, it still remains untested if ontogenetic δ 15 N trends exist that are linked to changes of the physiology of the bivalve. Such trends could superimpose environmental and ecological signals and bias interpretations. The reason for this assumption is that the protein-amino acid composition of the shell organic matrix varies through lifetime (Goodfriend and Weidman, 2001). Since the different amino acids exhibit vastly different nitrogen isotope fractionation patterns (McClelland and Montoya, 2002;Chikaraishi et al., 2007), a compositional change would inevitably be associated with a shift in bulk δ 15 N values of shell organics and periostracum.
Here, the hypothesis is tested that bulk δ 15 N CBOM chronologies of the long-lived bivalve mollusk, Arctica islandica, from the Northeast Atlantic are affected by ontogenetic trends. This species can attain a lifespan of several decades to hundreds of years (Thompson et al., 1980;Schöne et al., 2005a;Wanamaker et al., 2008;Butler et al., 2013). Since ecological and ontogenetic shifts can eliminate or reinforce each other, specimens were used that lived during different time intervals in different habitats. Possible reasons for seasonal to multidecadal δ 15 N CBOM variations are discussed. The δ 15 N values of shell organics were also compared to such of periostracum. Results of this study can help to improve bivalve shell δ 15 N-based ecological and environmental reconstructions. Findings highlight the need for compound-specific isotope analysis of shell organic matrices which so far has only rarely been explored (Ellis, 2012;Misarti et al., 2017).

Sample Material
For the current study, shells of four specimens of Arctica islandica from different localities and water depths in the North Atlantic Ocean were selected (Figure 1 and Table 1). Two specimens were collected alive from surface waters, one from NE Iceland (þistilfjörður, þórshöfn) and one from the Irish Sea (Figure 1 and Table 1) in August 2012 and July 1868, respectively. The modern specimen (ICE12-05-12 A-L) was eviscerated immediately after collection and both valves then stored in a dry and dark environment. The historical ocean quahog (M071868-A1L) belongs to the Karl August Moebius collection that is curated by the Zoological Museum of the University of Kiel. Details on the exact sample locality, water depth and treatment after collection are not available, but the valves were stored in dry condition when obtained from the museum collection in 2004. In the historical shell, large portions of the periostracum were missing and flaked off from the shell. In addition, single valves of two radiocarbon dated subfossil specimens came from the Fladen Ground (Late Holocene, specimen WH241-604-BoxL-D11R) and the East Farn Deeps (Early Holocene, specimen WH241-677-BoxB-D1R), North Sea (Figure 1 and Table 1). These specimens were collected by bottom trawling from 113 and 79 m water depth, respectively, during cruise no. 20020219 with the R/V "Walther Herwig III" in summer 2002. Uncalibrated radiocarbon ages (Libby years) were 2,840 ± 40 year BP (ontogenetic years 5-9 in specimen WH241-604-BoxL-D11R) and 8,770 ± 50 years BP (years 7-9 of WH241-677-BoxB-D1R). Calibrated 14 C ages (cal yr BCE, 2σ) were calculated with CALIB 8.2 ( 1 ; last access: 28 June 2021; Table 1) using the calibration set marine20.14c (Heaton et al., 2020) and R values of −140 ± 35 and −146 ± 68 year, respectively.

Sampling
Prior to sampling, one valve of each specimen was gently rinsed with de-ionized water and allowed to air-dry. No oxidative reagents were used to clean the shells prior to shell powder acquisition. As outlined by Gillikin et al. (2017), in contrast to other, porous bioceramics, bivalve shells are very dense and thus do not require oxidative cleaning. In addition, oxidative agents can potentially alter the isotope composition of shells (Schöne et al., 2017;own unpublished data). Likewise, no acids were used to remove adhering calcium carbonate crystals from periostracum, because this treatment modifies the δ 15 N values as well (Schlacher and Connolly, 2014; own unpublished data). Data presented here thus represent the δ 15 N values of both the intercrystalline and intracrystalline organic matrix of the shells. For reliable identification of shell growth patterns and temporal contextualization of the samples, valves were mounted to Perspex cubes with a fast-curing plastic welder (WIKO Multi Power 03) and surfaces along the axis of maximum growth from the hinge to the ventral margin coated with a ca. 8 mm wide thin layer of high-viscosity WIKO metal epoxy resin 05. Along that axis, the valves were cut with a low-speed precision saw (Buehler FIGURE 1 | Map showing sampling localities of Arctica islandica (filled yellow circles) and major currents in the NE Atlantic Ocean (NAC, North Atlantic Current; EIC, East Greenland Current; IC, Irminger Current; SC, Shetland Current). Red colors denote warm surface currents, blue cold currents. See Table 1 for details of specimens.
Annual shell growth increments and lines guided the acquisition of periostracum and shell powder samples and ensured that isotope data of these samples could be placed in precise temporal context. Schöne (2013) reviewed that growth line formation in A. islandica starts ca. 4 weeks after the seasonal temperature maximum and takes ca. 2 months. During this time period, biomineralization rate is extremely slow and may occasionally come to a complete halt (Jones, 1980;Schöne et al., 2005a). Thereafter, growth resumes at slow rate until food availability increases (Thompson et al., 1980;Schöne et al., 2005a). In surface waters, the main growing season of A. islandica covers the time interval from ca. mid-November to mid-September of the following calendar year, while in deeper waters, the main growing season is typically lagged by several months, because the seasonal temperature maximum occurs later than in surface waters (Schöne et al., 2005a,b). The "annual" growth increments of specimens from different water depths thus cover different parts of the year (Schöne et al., 2005a,b).
Sampling was completed with a Rexim Minimo low-speed milling system (ca. 600 rpm) which was firmly attached to a binocular microscope (64 and 160 × magnification). For micromilling, various different drill bits were used: (i) a diamondcoated cylindrical drill bit (1 mm diameter; Gebr. Brasseler GmbH & Co., KG #835 104 010), (ii) a diamond-coated inverted cone (3 mm diameter; 814 104 030) and (iii) a conical SiC drill bit (300 µm at tip; H52 104 003). Periostracum was micromilled from individual "annual" increments in shell portions located posteriorly and anteriorly from the metal epoxy band. In slowgrowing shell portions, each periostracum sample represented an average of up to 3 years, in one case 4 years (Supplementary Material 1). Furthermore, shell powder samples were taken from the inner portion of the outer shell layer (iOSL). For that purpose, the metal epoxy, periostracum and the outer portion of the outer shell layer (oOSL) were physically removed from one half of each valve. Then, powder samples were micromilled in cross-section, guided by internal growth patterns, to obtain "annual" averages. In fast-growing shell portions near the umbo, samples were obtained by surface micromilling (approx. 1 to 2 cm broad swaths). To assess the seasonal δ 15 N variance, subannual samples were obtained in some fast-growing shell portions. In slow growing shell portions near the ventral margin of the oldestgrown subfossil specimen, each shell powder sample contained up to 6 years (Supplementary Material 1).
It should be pointed out that this kind of shell powder sampling was extremely time-consuming (i.e., 30 to 60 min per sample), because micromilling had to be extremely precise to minimize unwanted time-averaging and proceed very gently to avoid too much frictional heat. Depending on column size (largevolume or small-volume method, see further below) approx. 5.0 to 16.7 mg or 2.4 to 4.0 mg of shell powder were required

Nitrogen Stable Isotope Measurements
Powdered samples were weighted into tin capsules and measured on a Thermo Scientific MAT 253 continuous-flow isotope ratio mass spectrometer coupled to an elemental analyzer (Thermo Scientific Flash EA 2000) equipped with a Costech zero blank autosampler at the Institute of Geosciences, University of Mainz, Germany. Samples were combusted in the EA at a temperature of 1,020 • C (reduction column temperature: 650 • C). 85 analyses of samples were completed with the standard column (largevolume method, hereafter referred to as Lvol-EA), the remaining 247 analyses of samples with a small-volume setup (Svol-EA; column outer diameter: 10.5 tapered to 6 mm, column inner diameter: 7.2 tapered to 1.8 mm, #IVA4680090001 distributed by IVA Analysentechnik GmbH & Co., KG), similar to the nano-EA setup described in Polissar et al. (2009). The Svol-EA method required approx. three times less sample material than the Lvol-EA method [shell powder of iOSL of A. islandica: 9.6 ± 3.8 mg (N = 56) vs. 3.8 ± 0.3 mg (N = 193) (average ± 1σ); periostracum: 473 ± 330 (N = 29) vs. 133 ± 79 µg (N = 54)] to attain signal intensities (amplitude m/z 28) of approx. 305 mV (shell powder) and up to 1,155 mV (periostracum); note, even small amounts of adhering shell powder resulted in lower signal heights, because CaCO 3 is much heavier than periostracum (for data summary see Supplementary Materials 1, 2). The analyses of the international standard, USGS43 demonstrate that measurements provided by the Svol-EA method are accurate. The significant reduction of sample material using the Svol-EA method improves the temporal resolution by reducing the time-averaging of each sample, which is highly beneficial for the analysis of nitrogen isotope chronologies derived from bivalve shells. Raw isotope data were corrected using a two-point correction method with L-glutamic acid reference standards, USGS40 (δ 15 N: −4.52 ) and USGS41 (δ 15 N: +47.57 ) or USGS41a (δ 15 N: +47.55 ), with m/z 28 signal intensities in the range of the analytes, and reported in per mil versus air. Four calibration batches containing triplicates of USGS40 and USGS41 (or 41a) were equally distributed across each analytical session: one calibration batch at the beginning and end, respectively, and two additional batches after each 21 to 24 samples. Quality control was accomplished with blindly measured USGS43 (Indian Human Hair; δ 15 N: +8.44 ) that were measuredtypically in triplicate-along with the samples (N = 31; accuracy = 0.08 ± 0.11 , precision = 0.09 ± 0.02 ).
As previously noted by Gillikin et al. (2017), the analytical error of the reference materials increases with decreasing signal intensity, and this relationship can provide information on the potential analytical error of the samples. For that purpose, the 1σ precision errors of the means of the reference material triplicates (35 triplicates of USGS40, 17 of USGS41, 18 of USGS41a, and 10 of USGS23) were plotted against the average mass 28 signal intensities of the respective triplicates and the data fitted with a power function (R = 0.56; p < 0.001) (Equation 1): With this non-linear equation, the estimated precision error (EPE) of a sample can be approximated from its signal intensity on mass 28 (ampl28 in mV). On average, the analytical precision (1σ) was better than 0.20 and 0.13 for shell powders and better than 0.11 and 0.05 for periostracum using the Lvol and Svol-EA methods, respectively. The precision errors ranged between 0.06 and 0.37 for shell powder, and between 0.03 and 0.26 for periostracum. In cases where an average value was computed from multiple measurements of the same sample, the variance of the aliquots needs to be taken into consideration, i.e., the respective error needs to be propagated. Actually, 41 samples were run in triplicate, 49 in duplicate and the remaining 111 only once (note, all Lvol-EA samples belong to the latter group). On average, the mean of the triplicates came with a 1σ precision error of 0.22 partly reflecting sample inhomogeneities. To provide a conservative error estimate, the estimated precision errors of all samples that were measured less than thrice were propagated with this average 1σ precision error (0.22 ) resulting were computed based on various weights of L-glutamic acid (9.52 wt%). Accordingly, shell powders contained, on average, 761 ± 305 ppm by weight (or µg/g; = 0.076 wt%) nitrogen (range: 305 to 1,846 ppm by weight = 0.031 to 0.185 wt%), whereas periostracum contained, on average, 8.79 ± 3.32 wt% N (range: 2.74 to 15.63 wt%).

Statistical Analysis
The potential trend in δ 15 N CBOM and δ 15 N P time-series was examined and quantified by linear regression with Pearson's correlation coefficient and associated p-value (df = n −2). To examine the differences between δ 15 N CBOM and δ 15 N P , a twoway Analysis of Variance (ANOVA) was conducted with two factors considered: (1) Tissues-carbonate and periostracum; (2) Shells-from NE Iceland and the Irish Sea, followed by pairwise comparisons. The assumption of normality and homogeneity of variances were diagnosed with Kolmogorov-Smirnov tests and Levene's tests, respectively. Analyses were conducted using packages stats (R Core Team, 2021) and car (Fox and Weisberg, 2019) in the R software version 3.6.3 (R Core Team, 2021). Spectral analysis (Fourier transformation, Lomb periodogram) was conducted to quantify the periodic oscillations in the timeseries of the two long δ 15 N CBOM series of shells from the Fladen Ground and the E Farn Deeps. The same analysis was performed with Atlantic Multidecadal Variation (AMV) index.

RESULTS
The modern bivalve from NE Iceland (ICE12-05-12 A-L) and the historical specimen from the Irish Sea (M071868-A1L) attained a lifespan of merely 33 years, whereas the two subfossil A. islandica specimens lived for 92 (Fladen Ground, WH241-604-BoxL-D11R) and 162 years (E Farn Deeps, WH241-677-BoxB-D1R) based on annual increment counts (Figure 2 and Table 3). δ 15 N CBOM chronologies of the shells from NE Iceland, the Fladen Ground and the E Farn Deeps showed a general and statistically significant decrease through ontogeny which were estimated with linear regression models ( Table 3). The specimens from the Fladen Ground and the E Farn Deeps shared a common weak trend (−0.006 year −1 ), while the decreasing trend occurred more than ninefold stronger in the NE Icelandic shell (−0.057 year −1 ). A weak decline in δ 15 N CBOM was also observed in the shell from the Irish Sea (−0.008 year −1 ), yet the trend is not statistically significant ( Table 3).
The two long chronologies showed markedly larger interannual variance during early ontogeny than later during life and seemed to exhibit an initial δ 15 N CBOM increase by up to approx. 1.41 (+0.071 and +0.064 year −1 , respectively) until the age of 20 or so, followed by a gradual decline (Figures 2C,D). It should be noted that the increase in δ 15 N CBOM which is coincidental with early ontogeny of A. islandica also fits into the multidecadal cycle throughout both records (See further). In the two shorter-lived specimens, such inflection points were not evident or superimposed by low-frequency components (see below) (Figures 2A,B).
Besides these potential ontogenetic trends, the two longer time-series also showed distinct multidecadal oscillations with a period length approx. 50 to 60 years and quasi-decadal variability at ca. 12-16 years as well as amplitudes ranging between 1.80 and 2.90 (Figures 2, 3 and Supplementary Material 1). A statistically sound spectral analysis of this low-frequency variability, however, would require longer time-series (95% confidence levels were barely reached, Figure 3). Furthermore, high-resolution sampling in a few annual increments of specimens from Iceland and the E Farn Deeps offered an insight into the seasonal variability. As demonstrated by Figures 2A,D, on seasonal time-scales, the δ 15 N CBOM data fluctuated almost an order of magnitude stronger (more than 2.50 ) than from year to year (0.35 ).
The largest δ 15 N CBOM difference (annual averages) through lifetime was observed in the specimen from the Fladen Ground (+2.63 to +5.92 = 3.29 ), followed by shells from NE Iceland (+4.54 to +7.29 ) and the E Farn Deeps (+5.71 to +8.58 ), both with an average range of approx. 2.90 . A significantly lower spread of only 1.01 was found in the specimen from the Irish Sea (+7.12 and +8.13 ) (Figure 2 and Table 3).
Periostracum was only preserved in the two live-collected shells from NE Iceland and the Irish Sea. Ontogenetic δ 15 N trends in periostracum were statistically insignificant yet somewhat showing a weak declining tendency, i.e., −0.007 and −0.006 year −1 , respectively ( Table 3). As in shell organics, the δ 15 N P range was larger in the NE Icelandic specimen (1.97 , +3.20 to +5.17 ) than in the shell from the Irish Sea (1.10, +6.24, and 7.34 ). In both specimens, the periostracum was depleted in 15 N relative to shell organics (NE Iceland: −2.07 , Irish Sea: −0.86 ) (Figure 4 and Table 3). Low-frequency changes of the δ 15 N P and shell δ 15 N CBOM chronologies showed some visual agreement, but the correlation was neither strong nor significant (Figure 2).

DISCUSSION
The ontogenetic δ 15 N CBOM decline was most clearly developed in the two old-grown specimens (92 and 162 years-old) (Figure 2), and the slopes of their regression lines were similar (approx. −0.60 per century; Table 3). Irrespective of locality, water depth and time of growth, shell δ 15 N CBOM values of these two long records declined through lifetime, suggesting a physiological cause and precluding low-frequency climate swings. In comparison, the ontogenetic δ 15 N CBOM trends from the two short-lived modern specimens might be influenced by the decadal fluctuations, anthropogenic and/or other environmental impacts.
In addition, the time-series revealed pronounced seasonal and multidecadal variability. The latter was best observed in the two long chronologies and comprised period lengths of ca. 50 to 60 years with considerable amplitudes of as much as almost 2.90 (Figure 2). It should be pointed out that the lifespan of these two bivalves covered two and three full cycles of multidecadal oscillations, respectively, suggesting the observed ontogenetic δ 15 N CBOM trends are unrelated to environmental changes. In short δ 15 N CBOM chronologies, fractions of such multidecadal oscillations can exaggerate or be misinterpreted as ontogenetic trends. This can explain the nine-fold stronger ontogenetic δ 15 N CBOM decrease (6.09 per century) in the shell from Iceland (Table 3). Thus, in order to disentangle multidecadal shell δ 15 N CBOM fluctuations from directed ontogenetic trends, it is crucial to study shells of sufficiently long-lived specimens.
Possible Causes for the Ontogenetic δ 15 NC BOM Trends The most pressing question is what caused the ontogenetic δ 15 N CBOM decline. The nitrogen isotope composition of calcium carbonate-bound organics, which mainly consists of proteins (Goulletquer and Wolowicz, 1989), can be influenced by a variety of different environmental and physiological factors including but not limited to the (1) isotope signature of total dissolved nitrogen (δ 15 N TDN ), (2) composition of the diet of bivalves, (3) metabolic/turnover rate (4), reproduction, and (5) protein synthesis.
The changes in environmental δ 15 N TDN and diet composition of bivalves can be subsumed under extrinsic factors. A shift in the isotope composition of total dissolved nitrogen [i.e., the sum of dissolved inorganic N (ammonium, nitrate, nitrite) and dissolved organic N (urea, amino acids etc.)] becomes encoded in photoautotrophs (phytoplankton, benthic algae etc.) and is subsequently imparted to all consumers including bivalves. Thus, the isotopic level of all components of the food web will change proportionately to the shift in δ 15 N TDN (e.g., Ohkouchi et al., 2015). Possible reasons for δ 15 N TDN fluctuations include changes in ocean circulation, mixed layer depth, upwelling, pH value, nitrogen cycling processes, influx of freshwater and input of anthropogenic nitrogen by atmospheric deposition or through rivers (e.g., Altabet et al., 1995;Cabana and Rasmussen, 1996;Voss et al., 2005;Sherwood et al., 2011;Ellis, 2012;Wang et al., 2018;Whitney et al., 2019;Duprey et al., 2020;Hopkins et al., 2020;Doherty et al., 2021). In addition, the δ 15 N value of bivalve soft tissues can also be affected by changes in the dietary composition of the bivalve (e.g., Kang et al., 1999;Tue et al., 2012). As a suspension-feeder, A. islandica generally feed on suspended POM (Josefson et al., 1995;Cargnelli et al., 1999), representing a mixture of zooplankton, phytoplankton, bacteria, benthic algae and organic detritus (Simon et al., 2002;Verdugo et al., 2004). The nitrogen isotope value of POM depends highly on the community structure of the phytoplankton, because each photoautotroph species exhibits distinct nitrogen isotope fractionation patterns and thus comes with distinct δ 15 N values (Montoya and McCarthy, 1995;Chikaraishi et al., 2009;Ohkouchi et al., 2015). However, the above listed extrinsic aspects fail to convincingly explain the common ontogenetic δ 15 N CBOM decrease in the studied A. islandica specimens, simply because it would require that δ 15 N changes in TDN or diet had occurred at the same rate, magnitude and direction in geographically distant settings and during different time intervals of the past. Such coincidence would be highly unlikely. It appears more reasonable that the ontogenetic nitrogen isotope shifts in shells of A. islandica were triggered by physiological factors, e.g., changes in dietary preferences or food selection capabilities through lifetime.
Yet, a change in the composition of potential food items is not necessarily mirrored 1:1 in δ 15 N CBOM values, nor does an unchanged composition of available food items lead to invariant δ 15 N CBOM values. Actually, most bivalves do not indiscriminately feed on offered organic substances, but can select food items before or after ingestion (e.g., Kiørboe and Møhlenberg, 1981;Shumway et al., 1985). Based on shell 14 C/ 12 C data, Erlenkeuser (1976) concluded that ocean quahogs are "real gourmets which feed on the most recent organic matter only." The food selection capability of A. islandica has subsequently been experimentally studied under laboratory conditions. For example, according to Shumway et al. (1985), this species conducts a pre-ingestive selection of organic particles with its labial palps. A constraint that should be mentioned, however, is that the labial palps are very short in A. islandica and particle selection capabilities thus less effective than in many other species (Kiørboe and Møhlenberg, 1981). No study has yet investigated in detail if food preferences of the ocean quahog change during lifetime. Respective behavioral changes are commonly observed in facultative deposit feeders and relate to the ontogenetic growth of siphons and labial palps. While juveniles predominantly graze on the seabed, adults with larger siphons filter-feed suspended seston (Rossi et al., 2004;Tue et al., 2012). Perhaps, in A. islandica, the preference for phytoplankton and/or the capability to select food particles gradually increases during ontogeny as a consequence of larger labial palps and siphons. If that assumption holds true, A. islandica may consume an increasingly larger proportion of food items that are depleted in 15 N, i.e., more fresh phytoplankton and less organic detritus, resulting in gradually lower δ 15 N CBOM values as it grows older.
While the hypotheses on changing food preferences and particle selection capabilities can explain the declining values, they fail to elucidate the observed increase in δ 15 N CBOM values of some shells prior to the age of 20 (Figures 2C,D). Changes in the allocation of energy to different compartments (somatic growth, reproduction, maintenance etc.) during ontogeny and associated changes in turnover rate offer a more convincing explanation of the observed isotope patterns. Such changes are described by combined Dynamic Energy Budget and Dynamic Isotope Budget models (Emmery et al., 2011). For example, during early youth, most energy is allocated to somatic growth, while energy flows predominantly into the production of eggs and sperms once maturity is reached (see below), provided that enough food is available. If energy supply is limited ( = less food available), somatic maintenance is prioritized over body growth and reproduction (Emmery et al., 2011). Energy availability determines the metabolic rate in the respective compartment and affects isotopic fractionation patterns. In small, young individuals, the structural volume of the body is lower and tissue turnover rates are higher than in larger, older specimens resulting in lower bulk δ 15 N values in tissues. With the increase in body size, the overall turnover rate decreases and-due to stronger isotope fractionation-tissues get gradually enriched in 15 N (Emmery et al., 2011). While this model could potentially explain the observed rise of shell δ 15 N CBOM values until the age of 20, it cannot be completely ruled out that such increasing trend might also be coincidental with the multidecadal δ 15 N CBOM variations due to extrinsic factors, which also observed in other parts of the record, e.g., 50-92 years in shell from the Fladen Ground (See sections "Seasonal to Multidecadal Variability" and "Variation of δ 15 N CBOM in Modern Shells" for more details).
Furthermore, the gradual change of the δ 15 N CBOM slope at around age 20 could be due to a gradual re-routing of energy from somatic growth to reproduction and/or a change in food preferences and particle selection capabilities (see above). Age 20 is nearly in the middle of the range at which A. islandica reaches sexual maturity (Thorarinsdóttir and Steingrímsson, 2000) and somatic growth strongly diminishes associated with major metabolic reorganizations (Abele et al., 2008). Since the gametes, especially eggs, mainly consist of proteins (Baptista et al., 2014) and the heavier nitrogen isotope typically becomes enriched in the proteins, the soft tissues and shell organics would become depleted in 15 N during gametogenesis. Yet, this hypothesis still does not fully explain why the δ 15 N CBOM values continue to decline thereafter, unless an increasingly larger amount of energy is allocated to reproduction or the capability to select particles becomes increasingly more sophisticated as the bivalve grows older.
A more likely explanation for the general ontogenetic decrease of δ 15 N CBOM values after the onset of sexual maturity (or even for the entire chronology if the initial rise in δ 15 N CBOM values was an artifact of superimposed multidecadal variability) is that different proteins were synthesized during different stages of life, which aided in the biomineralization process and became included in the shell organics. In fact, the composition of amino acids, the building blocks of proteins, varies over lifetime. In A. islandica, glutamic acid (Glu), aspartic acid (Asp) and Alanine (Ala) facilitating crystal growth are favored during fast growth in early ontogeny, whereas the relative proportion of Phenylalanine (Phe), Tyrosine (Tyr) and Lysine (Lys) enhancing shell strength increases in shell organics later during ontogeny (Goodfriend and Weidman, 2001). The different amino acids differ strongly in their δ 15 N value, because they undergo specific nitrogen metabolic processes (McClelland and Montoya, 2002;Chikaraishi et al., 2007Chikaraishi et al., , 2009. During trophic transfer, the so-called "trophic amino acids" (Popp et al., 2007) (Glu, Asp and Ala) experience strong nitrogen isotopic fractionation (TDF = 8 in case of Glu). These amino acids are synthesized de novo by consumers and record the isotope signature of the diet. In contrast, the socalled "source amino acids" (Popp et al., 2007) (Phe, Tyr and Lys) can only be fabricated by microorganisms (and plants) and are absorbed by consumers with minimal isotopic fractionation (TDF = 0.4 in case of Phe). Their original isotope signature imparted by the primary producers is thus largely preserved at higher trophic positions. In brief, Glu, Asp and Ala are enriched in 15 N, whereas Phe, Tyr and Lys are depleted in this isotope. Therefore, the prevalence of trophic amino acids in shell portions laid down during early ontogeny would explain the higher bulk δ 15 N CBOM values, whereas the relative increase of weakly fractionating source amino acids later during life resulted in an ontogenetic decrease of bulk δ 15 N CBOM values. It would be worth testing this interpretation in future studies through the use of protein amino acid-specific (compound-specific) nitrogen isotope analyses by examining amino acid composition as well as the δ 15 N values of each amino acid through the lifetime of the A. islandica shells.

Seasonal to Multidecadal Variability
Aside from ontogenetic trends, δ 15 N CBOM data varied on seasonal, interannual and multidecadal time-scales (Figure 2), which certainly merits further attention in subsequent studies. A detailed analysis of the possible causes for such variations, especially the longer-term variations require a larger number of long δ 15 N CBOM chronologies from modern specimens along with high-resolution environmental, ecological and physiological data. Some considerations for the observed δ 15 N CBOM patterns are laid out in the following.
It cannot be ruled out that the seasonal to multidecadal δ 15 N CBOM variability was linked to physiological changes discussed earlier, but a response to extrinsic factors, δ 15 N Diet and δ 15 N TDN , appears more reasonable. A likely candidate for the observed seasonal δ 15 N CBOM fluctuations includes changes in the isotopic composition of the food of the bivalves, i.e., δ 15 N Diet . Due to Rayleigh fractionation kinetics during primary production, phytoplankton formed at the beginning of the bloom thus have a lower δ 15 N value than such formed near the end of the bloom when the nitrate pool is exhausted, because photoautotrophs preferably assimilate 15 N-depleted nitrate (Altabet and Francois, 1994). This isotope range could be reflected in the diet of the bivalves. Additional variability in δ 15 N Diet can arise from seasonally varying phytoplankton community structures induced by the physical conditions of seawater, e.g., nutrient levels, light penetration and stratification (Barton et al., 2014). Since the different species are isotopically distinct (Needoba et al., 2003; see also overview in Chikaraishi et al., 2009), a change in phytoplankton species composition throughout the year can cause variations of δ 15 N Diet , and eventually δ 15 N CBOM . Furthermore, during different seasons, phytoplankton can assimilate different nitrogen species (Bronk and Glibert, 1993), which can also lead to a considerable variance of the δ 15 N baseline.
The most prominent pattern in the isotope chronologies aside from ontogenetic trends was a distinct multidecadal δ 15 N CBOM variability. These low-frequency changes could mirror fluctuations of δ 15 N TDN values induced by changes in water mass source. For example, a southward shift of the Polar Front would result in a larger proportion of Arctic waters reaching the study site in NE Iceland with the East Greenland Current. In contrast, a northward shift of the Polar Front would increase the influence of Atlantic waters mediated by the Irminger Current (Marali and Schöne, 2015). Arctic waters carry a higher δ 15 N TDN signature (Hansen et al., 2012;ca. 6-8 : Lehmann et al., 2019) than Atlantic waters (4 , Marconi et al., 2019) and this environmental isotopic baseline shift would be reflected in the entire food web. However, this explanation would not apply to the specimens from the remaining three sites which were only fed by Atlantic waters (Figure 1).
More likely, the multidecadal δ 15 N CBOM variations reflect shifts in the dietary composition of the bivalves. In fact, changes in phytoplankton species composition are not only known from seasonal, but also interannual and longer time-scales (e.g., Benedetti et al., 2019). However, the underlying controls of long-term changes of the phytoplankton community structure remain less clear. Potential causes include temperature-induced changes in zooplankton assemblages preying on phytoplankton and variations in ocean circulation (Barton et al., 2014), winddriven changes of the mixed-layer depth affecting the timing of the phytoplankton bloom (Henson et al., 2009) and resource competition dynamics among primary producers (Benedetti et al., 2019). Doherty et al. (2021) recently suggested that fluctuations in nitrate utilization controlled by nitrate supply and changes in large-scale circulation patterns could have caused δ 15 N patterns observed in a six-hundred-year coralline red algae record from the Labrador shelf. Following their arguments, a larger proportion of nutrient-poor polar waters delivered by the Labrador current resulted in a complete utilization of nitrate and δ 15 N values in algae as high as δ 15 N of nitrate. In contrast, an increased advective supply of relatively nutrient-rich Atlantic waters caused the opposite. These hydrographic changes were coupled with the Atlantic Multidecadal Variation (AMV) and Atlantic Meridional Overturning Circulation (AMOC): Polar waters dominated during negative AMV phases (corresponding to negative SST anomalies in the northern North Atlantic) and weak AMOC, whereas Atlantic waters prevailed when the AMV was in its positive state and the AMOC strong (note, AMV and AMOC are positively coupled, Zhang et al., 2019). The AMV index describes longer-term variations of sea surface temperature (SST) in the North Atlantic Ocean (Schlesinger and Ramankutty, 1994) that appear to be forced by volcanic activity (Birkel et al., 2016;Mann et al., 2021) and vary within a frequency band corresponding to periods of 50 to 70 years (Schlesinger and Ramankutty, 1994;Delworth and Mann, 2000;Keenlyside et al., 2008;Mann et al., 2021).
Interestingly, the period length of the multidecadal δ 15 N CBOM cycles of A. islandica (Figure 3) falls in the same range as that of the AMV (Figure 5). A direct comparison with the AMV index, which could shed more light on a potential link between the two records, however, is only meaningful with the NE Iceland δ 15 N CBOM chronology. The lifespan of all other specimens does not overlap sufficiently long or not at all with the AMV, which is computed from instrumental SST data and thus only covers the time interval since 1856 CE. Although the short NE Iceland δ 15 N CBOM chronology showed a strong (visual) inverse covariation with the AMV, not just in the low-frequency band (multidecadal component), but also on quasi-decadal timescales ( Figure 5). To verify this coupling, longer shell nitrogen isotope time-series of modern A. islandica are certainly needed.
In anticipation of a verification of this finding in future studies, some thoughts are provided in the following that could explain an inverse relationship between δ 15 N CBOM data and the AMV. These ideas were inspired by the works of Edwards et al. (2001); Barton et al. (2003) and Doherty et al. (2021). During negative AMV phases and weak AMOC, lower (ocean and air) temperatures and stronger northerly winds resulted in intensified vertical mixing and delayed water column stratification with negative effects on primary production (Sverdrup, 1953: criticaldepth concept). In response to a sluggish meridional circulation, the supply of relatively nutrient-rich Atlantic waters onto the NE Atlantic shelf areas was decreased. Thus, nitrate became depleted and completely utilized by photoautotrophs leading to δ 15 N Diet and δ 15 N CBOM values as high as δ 15 N of nitrate. In contrast, during positive AMV phases and stronger AMOC, warmer water promoted water column stratification and a shallow mixed-layer depth early in the year. The increased AMOCmediated advection of nutrient-rich Atlantic water caused a surplus of nitrate in surface waters that was only incompletely consumed leading to lower δ 15 N Diet and δ 15 N CBOM values. The quasi-decadal fluctuations in the NE Iceland δ 15 N CBOM chronology, which mirror such in the AMV index, can likely be explained by similar mechanisms described above. Different temperature regimes during positive and negative AMV phases were presumably also associated with a change in phytoplankton species composition which further affected the isotopic signals of diet and CBOM. As outlined further above, this is because each phytoplankton species carries a distinct δ 15 N signature (Needoba et al., 2003) related to fractionation differences during nutrient uptake (Altabet and Francois, 1994).

Variation of δ 15 N CBOM in Modern Shells
In contrast to the studied subfossil specimens, ontogenetic δ 15 N CBOM trends were more challenging to identify in the two live-collected modern specimens. Firstly, their lifespan barely exceeded 50 % of the period length of the multi-decadal variability. Existing trends may thus be masked or biased by the multidecadal fluctuations (viz. the strong δ 15 N CBOM slope of the NE Iceland specimen; See section "Seasonal to Multidecadal Variability"). In addition, the two specimens lived at times during which anthropogenic environmental perturbations have already started to impact marine ecosystems and altered nitrogen dynamics in the ocean as evidenced by changes in δ 15 N TDN (Häder et al., 2015). For instance, ocean acidification facilitated nitrogen fixation pathways (Hopkins et al., 2020) and resulted in 15 N-depleted ecosystems (Sigman and Casciotti, 2001;Gruber, 2008). It is therefore hard to tell if a negative δ 15 N CBOM trend reflects a physiological or human-induced pH change or both. In addition, the influx of fertilizers to the coastal environments which the studied bivalves inhabited could have affected the δ 15 N TDN signatures. The application of fertilizers which are typically 15 N-depleted in nitrate and/or ammonia could have had significant impacts on coastal δ 15 N TDN and δ 15 N POM (Heaton, 1986), and were consequently transferred through the food chains. Presumably, such an effect played a greater role in the Irish Sea than in more pristine waters of NE Iceland. The changes in δ 15 N CBOM values from the anthropogenic impacts could have been compensated by the physiology-related δ 15 N shift resulting in the nearly flat curve observed in the specimen from the Irish Sea. Hence, it is again stressed out that longer δ 15 N CBOM chronologies, especially from the modern specimens, were needed to mathematically disentangle physiological and other factors.
Can δ 15 N Values of Periostracum Serve as an Environmental Proxy?
In the great majority of hitherto studied bivalves, the δ 15 N values of the soft tissues (specifically the muscle), periostracum and shell organic matrix were significantly correlated (O'Donnell et al., 2003;Carmichael et al., 2008;Delong and Thorp, 2009;Watanabe et al., 2009;Kovacs et al., 2010;Graniero et al., 2016Graniero et al., , 2021Darrow et al., 2017). This finding has opened new perspectives for studies evaluating the anthropogenic impact on ecosystems (Black et al., 2017;Fritts et al., 2017), hydrographical reconstructions (Whitney et al., 2019) and can be exploited in trophic (paleo)ecology, i.e., studies where soft tissues are not available or the agents used to preserve soft tissue altered the δ 15 N signals (e.g., Versteegh et al., 2011).
Although the number of periostracum samples was limited in the present study, mostly due to preservation issues, data agreed at large with previous findings, specifically that by Whitney et al. (2019) on the same species: Time-series of δ 15 N P and FIGURE 5 | Monthly AMV index (note y-axis inverted) based on Kaplan SST V2 (calculated by NOAA PSL1, http://www.psl.noaa.gov/data/timeseries/AMO/; last access: 7 July 2021) including a 25-month smoothed AMV curve (black) and stable nitrogen isotope chronology of shell carbonate-bound organic matrix (CBOM) from specimen ICE12-05-12 A-L (NE Iceland). Note, (visual) inverse agreement between the two series. For interpretation see main text. δ 15 N CBOM showed some agreement with each other, at least in the low-frequency domain, but with an offset (Figure 4 and Table 3). CBOM was enriched in 15 N relative to periostracum, on average, by 0.86 (Irish Sea) to 2.07 (NE Iceland). We follow the interpretation by Whitney et al. (2019) and conclude that both materials, periostracum and shell organics capture similar signals, but are isotopically offset due to differing amino acid composition. However, what is difficult to understand is why the nitrogen isotopic offset between the periostracum and shell organics differed between the two specimens by 1.21 (Figure 4 and Table 3). One possible explanation is that the historical shell was unknowingly treated with chemical agents or had been kept immersed in fluids that modified the isotope composition of the periostracum and/or the shell organics. As demonstrated in existing studies, storage in solutions can sometimes alter the amino acid composition and thus the isotope signatures (see reviews in Versteegh et al., 2011;Gillikin et al., 2017;Whiteman et al., 2019). For example, storage in formalin has been shown to increase δ 15 N values of fish tissues by 0.5 irrespective of duration of immersion (Edwards et al., 2002. Two months of immersion in formalin, followed by 2 months in ethanol shifted the δ 15 N values of fish muscle by 1.41 (Bosley and Wainright, 1999). In contrast, bivalve shell organic matrix became depleted in 15 N by up to 5.9 after storage in ethanol for 73 years (Versteegh et al., 2011). An alternative explanation would be the varying diets of A. islandica from the two different periods and locations. As Whitney et al. (2019) pointed out that the biomineral consisted of a higher proportion of trophic amino acids (e.g., Lys and Phe) compared with those in periostracum, it is possible to observe a higher δ 15 N offset between shell and periostracum if the most recent NE Icelandic A. islandica belonged to a higher trophic position, e.g., incorporating POM with a larger proportion of detritus deriving from animals of higher trophic levels.
Whereas it might be easier to measure a nearly pure organic material than a material which consists of approx. 95 to 99 wt% CaCO 3 , we still feel that the shell organic matrix serves as the more useful and robust ecological archive. Firstly, sampling can be done more accurately in the shell carbonate, whereas dry periostracum is very brittle and flakes off from the shell in large pieces. It is therefore more difficult to sample periostracum without time-averaging. Secondly, periostracum is often not well preserved in old-grown specimens, even if live-collected, and missing in portions of the shell that formed during early ontogeny. In dry samples from museum collections, the periostracum is rarely well preserved on the entire shell surface, particularly if the samples have been touched frequently or were shuffled around. Thirdly, with increasing geological age, the likelihood to recover shells with intact periostracum decreases. The organic matrix of the shell is then the only remaining material that can be measured to reconstruct paleoecological conditions. Fourthly, contamination is a major concern. The periostracum protects the animal during lifetime against dissolution and microbial attack and as such is exposed to seawater. It is thus not surprising that epibionts and a rich microbiome settle on or even in the periostracum. Sometimes, sediment grains, diatoms, coccolithophorids etc. can be found trapped within the periostracum. It appears impossible to remove such contaminants effectively, and the use of oxidants and acids bears the danger to modify the nitrogen isotope signal of the periostracum unwittingly. In contrast, the shell carbonate, specifically the "inner core" of the shell, i.e., the iOSL is well protected against contamination, and should, in our view, receive prioritized use.

SUMMARY AND CONCLUSION
The δ 15 N values of the organic matrix in shells of the ocean quahog, Arctica islandica vary on seasonal to multidecadal time-scales and show a gradual decrease through lifetime by ca. 0.6 per century. The latter is most likely related to physiological changes of the bivalve. Should subsequent studied substantiate the findings herein and long δ 15 N CBOM chronologies of A. islandica from other sites and time intervals also show the same decline through lifetime, these ontogenetic trends need to be mathematically removed before these data are used for ecological and environmental reconstructions. Presumably, the ontogenetic decline in δ 15 N CBOM values results from a change in the type of proteins synthesized at different stages of life (Goodfriend and Weidman, 2001), i.e., a gradual shift from proteins rich in trophic ( = strongly fractionating) amino acids (Glu, Asp, Ala) during youth toward proteins rich in source amino acids (Phe, Tyr, Lys) during adulthood. Alternative explanations include ontogenetic changes in food preferences and particle selection capabilities (i.e., a larger proportion of phytoplankton than organic detritus containing remains of organisms from higher trophic positions) or changes in the allocation of energy (to somatic growth during youth, and to reproduction and maintenance later during life) and associated changes in turnover rate. Aside from this ontogenetic trend, distinct multidecadal δ 15 N CBOM variations (ca. 50 to 60 years) exist, modulated by quasi-decadal variability. These low-frequency fluctuations were likely governed by variations in nutrient supply mediated by the AMV/AMOC combined with changes in nitrate utilization by photoautotrophs and associated Rayleigh fractionation processes. To reliably identify and distinguish multidecadal variability and ontogenetic shifts, δ 15 N CBOM time-series need to be sufficiently long. Furthermore, the presence or absence of ontogenetic trends requires specimens from different time intervals of the past and ideally also different localities and water depths. Otherwise, there is a danger to misinterpret directed environmental trends as ontogenetic shifts, or vice versa.
The stable nitrogen isotope composition of periostracum can offer an alternative means to assess ecological changes in the more recent past as long as chemical and biological contamination can be precluded. However, sampling of dry, brittle periostracum is challenging and prone to produce timeaveraged chronologies. In the current study, the nitrogen isotope offset between contemporaneously formed shell organics and periostracum differed considerably (1.51 ) between two livecollected specimens, of which one lived in the mid-19th century and was obtained from a museum collection and the other lived between the 1980s and 2012 CE. The difference may be explained by an unreported treatment of the historical shell with agents that modified the isotope signals of the periostracum or alternatively, be explained by the varying trophic levels of the two bivalves from different times and locality.
To address open questions raised herein and test hypotheses offered to explain the reasons for ontogenetic and multidecadal δ 15 N CBOM variations, subsequent studies should: (1) employ compound-specific isotope techniques, and (2) compare highresolution nitrogen isotope data from several old-grown, livecollected specimens with a plethora of environmental, ecological and biological data sets.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author/s.

AUTHOR CONTRIBUTIONS
BS: conceptualization, funding acquisition, resources, sampling, data analysis, interpretation, visualization, and writing. QH: data analysis, interpretation, visualization, and writing. Both authors contributed to the article and approved the submitted version.