From one valve of each shell, two 3 mm-thick slabs were cut out along the axis of maximum growth (Figure 1A). For this purpose, the valves were glued to acrylic glass cubes using WIKO Multi Power 3 plastic welder and covered along the cutting axis with a protective layer of WIKO 05 metal epoxy resin. After air-curing, specimens were cut with a low-speed rotational saw (Buehler IsoMet 1000) equipped with a diamond-coated wafering thin blade (0.4 mm thickness; Buehler 15LC 11-4255) operated at 200 rpm. The shell slabs were manually ground on glass plates using F800 and F1200 SiC suspensions and subsequently polished with 1 µm Al₂O₃ on a Buehler MasterTex polishing cloth. After each preparation step, specimens were ultrasonically rinsed with tap water for 6 min. One of the slabs was affixed to a glass slide for growth pattern and stable oxygen isotope analysis, whereas the other slab was attached to a one-inch SEM sample holder with carbon stickers for microstructure analysis (Figure 2B; Supplementary Figure S1).

Bivalves grow their shells periodically, leading to the deposition of growth increments and lines (Clark, 1974; Clark, 1975). As in most other bivalves, prominent growth lines are formed annually in A. islandica (Thompson et al., 1980a; Ropes et al., 1984; Karney et al., 2011). Growth line formation occurs ca. four weeks after the annual temperature maximum (Thompson et al., 1980a; Jones, 1980; Schöne et al., 2005b). The annual lines can therefore be used to determine the ontogenetic age of the animal and to place the shell record in temporal context. In order to analyze the growth patterns, shells were immersed in Mutvei’s solution (12.5% Glutaraldehyde, 0.5% acetic acid, and 5 g/L alcian blue) and kept at 38°C for 8 min under constant stirring, which stained and fixated the shell organics while gently etching the shell carbonate (Schöne et al., 2005a). After rinsing and air-drying, shells were imaged in sectoral (= one-quarter) dark-field illumination under a Leica Stemi 508 stereomicroscope equipped with a Canon EOS 600D DSLR camera. All sclerochonological measurements (i.e., distances between the annual growth lines, δ¹⁸O_shell sampling spots and SEM images) were performed with the image processing software ImageJ (Schneider et al., 2012; Rueden et al., 2017).

Since most of the studied specimens lacked clearly defined subannual growth patterns, the δ¹⁸O_shell method was used to place the shell record in temporal context. For this purpose, the paleothermometry equation by (Grossman and Ku, 1986) with the −0.27‰ PDB-SMOW scale correction (Gonfiantini et al., 1995) (Eq. 1) was solved for δ¹⁸O_shell (Eq. 2). Then, a predicted seasonal δ¹⁸O_shell (model) profile was computed for each locality based on the corresponding instrumental temperature (Figure 2B) and δ¹⁸O_water data taken from (LeGrande and Schmidt, 2006) using Eq. 2. Measured δ¹⁸O_shell data were aligned to best fit the predicted seasonal δ¹⁸O_shell(model) profiles. This way, a precise calendar date could be assigned to each sample. Finally, these data were linearly interpolated to reconstruct the daily growth rate. In case of the Baltic Sea shells, an existing seasonal shell growth model was used (Schöne et al., 2021). T = 20.6 − 4.34 × ( δ 18 O shell − ( δ 18 O water − 0.27 ) ) , (1) δ 18 O shell ( model ) = 20.6 − 4.34 × ( δ 18 O water − 0.27 ) − T 4.34 , (2)

To measure the stable oxygen isotope composition of the shells, carbonate powder samples of ca. 60 µg were taken along the axis of maximum growth using a micro drill (Rexim Minimo KM11G) firmly attached to a binocular microscope equipped with a conical SiC drill bit (300 µm diameter at tip; Gebr. Brasseler GmbH and Co. KG, model no. H52.104.003). For most specimens, sampling was accomplished on the surface of the polished shell slabs, but in case of the thin shells from deeper settings, approx. 100 µm broad swaths were milled on the surface of the valves (surface milling; Figure 1A). Carbonate powders were dissolved in He-flushed borosilicate exetainers at 72°C by adding 99.9% phosphoric acid and the released CO₂ gas was measured in continuous flow mode with a ThermoFisher MAT 253 isotope ratio mass spectrometer in coupled to a GasBench II. Stable isotope values were calibrated against an in-house Carrara marble standard (δ¹⁸O [V-PDB] = −1.91%) and are given relative to the V-PDB standard. Oxygen isotope values were not corrected for differences in acid fractionation factors between the calcite (Carrara marble) and aragonite (shell), because the source equation for Eq. 1 + Eq. 2, i.e., the paleothermometry equation by Grossman & Ku (1986) is based on uncorrected δ¹⁸O values (see Füllenbach et al., 2015, for more details). Measurement accuracy based on blindly measured reference materials of known isotopic composition (NBS-19) was better than 0.04‰.

Since the contrast between shell carbonate and finely distributed intercrystalline organics in backscatter SEM is insufficient to analyze individual BMUs, they need to be revealed prior to morphometric analysis. This is typically accomplished by slight etching of the BMUs and/or removal of the intercrystalline organic matrices, e.g., by immersion of the shells in weak acids, bleach or hydrogen peroxide (Crippa et al., 2016). This treatment, however, evokes a violent reaction (Lavkulich and Wiens, 1970), alters the original BMU shape (Höche et al., 2021a) and sometimes affects different materials of the shell heterogeneously. To overcome these problems, an ultrafine chemo-mechanical polishing step was instead employed in this study: shell slabs were polished for 10 minutes on a rotational lap (Buehler MetaServ 2000) at 50 rpm using a Buehler MasterTex polishing cloth and Buehler MasterMet polishing suspension. With a grain size of 60 nm, the colloidal silica particles of this suspension were small enough to mechanically erode the finely distributed organic phases, while its alkaline pH of ca. 10.1 aided in the gentle chemical removal of the organics. The resulting shell slab surface was polished very evenly and intercrystalline organics were removed thoroughly while keeping abrasion of the carbonate phase at a minimum (Figures 3A,B).

As a previous study revealed a statistically significant relationship between the BMU size in the hinge plate of laboratory grown specimens and temperature (Höche et al., 2021a), the same shell portions were investigated here (953 images; Table 1). For comparison of BMU data in different contemporaneous portions of the shell, the outer shell layer (OSL) of the ventral margin was also studied (241 images; Table 1). Most images were taken in transects along the maximum growth axis to assess microstructural variation through time (Figure 1B; Supplementary Figure S1). Additionally, images were taken along contemporaneously formed shell portions in the OSL of the ventral margin and in the hinge plate (in specimen ICE06-6.2-A6R) to assess the spatial variability of BMU sizes and to test the reproducibility of the BMU proxy method (Figure 1B).

Images used in BMU morphometry were generated with a Phenom Pro Desktop SEM (3^rd generation) equipped with a CeB₆ electron source and backscatter electron detectors at 7700x magnification and 10 keV. The images were subsequently processed with the machine-learning based image segmentation program Ilastik (Berg et al., 2019), which classifies each individual BMU as a unique shade of gray based on a prior training of the software. In a set of exemplary input images (randomly selected among the images of this study to cover a wide range of microstructural textures), labels were painted over the pixels to classify them as either BMU or intercrystalline space. Based on the characteristics (i.e., intensity, texture, contour sharpness) of these manually classified regions, the probability of being a BMU (p_BMU) was calculated for each remaining pixel. Individual BMUs were then detected in these probability maps by applying a two-step probability threshold (Canny, 1986; Condurache and Aach, 2005). Connected pixels with p_BMU > 0.95 were considered as the cores of individual BMUs and were assigned with a unique gray value. These areas were then expanded outward to include neighboring pixels of p_BMU > 0.5. This two-step hysteresis thresholding process allows for effective separation of adjoining BMUs, which would otherwise be considered as one entity. The segmented images were inspected visually, and all invalid results (= severe over- or under-splitting of the BMUs) arising from poor image quality due to sample drift or charging effects were discarded (217 of 1194 images). Finally, the size of each BMU in each correctly segmented image was automatically measured and its elongation estimated by fitting an ellipse using the programming language python and image processing library scikit-image (van der Walt et al., 2014). The scripts and classifier used in this study are available online (Höche et al., 2021b) and can be reused in future studies to batch process SEM images acquired under the same settings for BMU morphometry.

The BMU raw data comprise one entry for every BMU of each of the 1194 images, marking the need for data reduction techniques. Since previous studies revealed significant heterogeneity among BMU size values and determined a temperature relationship predominantly for the largest BMUs (Höche et al., 2021a), the same statistical measures were used herein. Of each image, the fifteen largest BMUs were selected and means and standard deviations were calculated for their size and elongation. The shell portion, microstructure type, ontogenetic age, and relative position within the annual growth increment corresponding to each image (i.e., BMU data point) were recorded.

To assess relationships between environmental, physiological and BMU morphological parameters at each locality, Pearson correlations (r) and corresponding significance values (p; significance threshold α = 0.05) were computed. In addition, linear models between the BMU size and the environmental and physiological parameters were computed via least squares regression. Trends in BMU size through lifetime of the bivalves were estimated by locally weighted scatterplot smoothing (Cleveland, 1979) using a window size of one-third of the data to represent long-term developments and 2 years of data for the seasonal trends. All calculations were performed using python and the modules SciPy (Virtanen et al., 2020) and statsmodels (Seabold and Perktold, 2010).

Based on annual increment counts, the studied shell portions of the specimens formed before the age of seventeen (Table 1). In the hinge plate, annual increment widths ranged from 0.03 to 1.81 mm and were broadest at around age four to seven (Figure 3C). According to the δ¹⁸O_shell-alignment (Supplementary Table S1), studied bivalves formed annual growth lines a few weeks after the annual temperature maximum. Among the specimens from one locality, the timing and rate of shell growth were highly synchronized (Supplementary Table S1), whereas they varied considerably between sampling sites (Figures 3C,E). The Icelandic shells mainly grew from October until late August of the following year and attained maximum growth rates during spring and summer. In contrast, shells from the Viking Bank grew between February and December, with growth rate maxima during late August. The specimen from the Norwegian Trench grew its shell from April to February of the following year, with the fastest shell formation rates during early September (Figure 3D).

The hinge plates were dominated by CA microstructure (occurring in 92.7% of the images). Annual growth lines were easily identified, because they consisted of ISP microstructure (Figures 1C, 3B). In the CA microstructures following an annual line, much smaller BMUs were formed (Figure 3B). These CA microstructures were occasionally replaced by FCCL. FCCL microstructures were also found in the most lateral regions of the hinge plate, i.e., in slower growing portions far away from the main growth axis. FCCL microstructures became more abundant with increasing age. Shells from the Baltic Sea consisted of a larger proportion of FCCL microstructures than those from the other localities (30.5% of images vs. 1.7% of images, respectively). Here, FCCL microstructures were not only found near the annual growth lines, but also in disturbance lines which frequently occurred during summer months.

The ventral margin of the shells was divided into an inner (ISL) and outer layer (OSL) separated by the myostracum (Figure 1B). The outermost portion of the OSL consisted of homogeneous microstructures, which gradually transitioned into CA and FCCL microstructures towards the myostracum. Inner shell layers consisted of FCCL microstructures. Like the hinge plate, annual growth lines of the outer shell layer of the ventral margin consisted of ISP microstructures. These ISP microstructures were most defined in the outer portion of the OSL, whereas they gradually disappeared toward the inner portion of the OSL (iOSL) close to the myostracum.

The size of the BMUs differed distinctly between study localities (Figure 4). Except for the Norwegian Trench and the Baltic Sea, this difference was statistically significant (p < 0.05). Specimens from Iceland formed the largest BMUs (9.44 ± 5.39 µm²), followed by those from the Baltic Sea (6.46 ± 2.80 µm²), while the BMUs were smallest in the deeper environments (Norwegian Trench 5.33 ± 2.51 µm² and Viking Bank 3.72 ± 1.47 µm²). BMU size data were also heteroscedastic, i.e., size variance increased in portions with larger BMUs. In both the hinge plate and ventral margin, BMUs of FCCL microstructures were significantly smaller (4.9 ± 2.3 µm²) than those of the CA microstructure (8.7 ± 5.1 µm²) and also more elongated (2.41 ± 0.41 in FCCL and 1.97 ± 0.24 in CA; Figure 4B).

Along the axis of maximum growth in the hinge plate, BMU sizes first increased and then decreased with ontogenetic age (Figure 5). In all shells, BMUs reached their maximum size during age seven to nine. At large, BMU sizes of the hinge plate correlated with the annual growth rate. The broadest annual increments, however, occurred a few years earlier than the largest BMUs, i.e., between age four to seven (Figure 3A). Furthermore, all shells, except those from the Baltic Sea, exhibited seasonal BMU size patterns (Figure 5). This was most evident in the Icelandic shells: BMUs were smallest at the beginning of each growing season, shortly after the annual growth line. Thereafter, BMU size gradually increased, reaching a maximum at around 70–90% of the annual growth increment and decreased toward the following annual line (Figure 5). This intra-annual BMU size trend was not limited to the axis of maximum growth, but was likewise observed in transects away from that axis (Figure 6). Specimens from the Viking Bank showed the same pattern as those from Iceland, yet with a smaller amplitude. The shell from the Norwegian Trench, in contrast, formed the largest BMUs about half way through the annual increments. In Baltic Sea specimens, BMU size variation was poorly synchronized among individuals and did not show a clear pattern over the course of each year. Note that 52 images of the shells from this locality had to be discarded because of problems during image segmentation (arising from highly irregular geometries of the interstitial spaces surrounding the BMUs), which obscured some portions of the shell growth records.

In the ventral margin, the size of BMUs of the iOSL was nearly identical to those formed contemporaneously in the hinge plate (Figure 7A). Hinge plate and iOSL also exhibited similar seasonal BMU size patterns. The only exception was observed in specimens from the Norwegian Trench, in which BMUs of the iOSL attained largest sizes at the beginning of the growing season. BMUs had a distinct size pattern from the outer shell surface to the myostracum (Figure 7B). Relative to the average size of all BMUs in a vertical transect, values were smallest in the outermost homogeneous microstructures of the oOSL (z ≈ −1; Eq. 3), but abruptly increased in size near the transition zone to the CA microstructures of the iOSL (to z of up to 3). Toward the myostracum, the BMU sizes decreased slightly (to z ≈ 0) throughout the CA portion of the iOSL until at the transition to FCCL microstructures, a sharp size peak was developed (z of up to 3). Close to the myostracum, the BMUs of the FCCL microstructures remained at a moderate size (z ≈ 0). The BMUs of CA microstructures were most variable in size, whereas FCCL BMUs had little size variation. z ( BM U area ) = BM U area − μ ( BM U area ) σ ( BM U area ) , with μ = mean   and σ = standard   deviation   (3)

The BMU morphology was only weakly correlated with the physiological and environmental parameters and did not reveal a common pattern for the studied localities. BMU sizes were weakly, but significantly positively correlated to water temperature at NE Iceland and the Viking Bank (Pearson r = 0.24 and 0.30, respectively; p ≤ 0.5; Figures 8A,E). The opposite, a significant negative correlation, was observed in specimens from the Baltic Sea (r = −0.31; p = 0.05). Linear regression models for BMU size and temperature, revealed similar slopes for Iceland and the Viking Bank (0.52 and 0.66 µm²/°C) but their intercept differed (7.32 versus −1.76°C; Figure 8A). Chlorophyll a concentration and the BMU size were weakly positively correlated at the Viking Bank and Norwegian Trench (r = 0.21 and 0.32), but only the latter was statistically significant. In the specimens from the Baltic Sea and Iceland, in contrast, BMU size was significantly (negatively) correlated to the chlorophyll a concentration (r = −0.19 and 0.31; p < 0.05; Figures 8B,E). Except for some specimens from the Baltic Sea, BMU size and shell growth (i.e., the annual increment widths and the reconstructed seasonal growth rates) was consistently positively correlated (Figures 8C–E). Note, that BMU sizes in specimens from low-seasonality environments (Norwegian Trench, Viking Bank) were stronger correlated to the annual increment width than those of the environments with high seasonality (Iceland, Baltic Sea).

At Iceland and the Viking Bank, BMU size covaried with the water temperature (Figures 8A,E). As in other bivalve microstructures (Olson et al., 2012; Gilbert et al., 2017; Milano et al., 2017b), larger BMUs were formed in warmer waters, i.e., seasonal BMU size maxima coincided with the seasonal temperature maxima (Figures 2B, 5D). Moreover, seasonal BMU size amplitudes were larger at NE Iceland than at the Viking Bank, likely owing to the greater seasonal temperature range (Figure 2B). At both localities, the temperature sensitivity of BMU size was similar (+0.52 and +0.66 µm²/°C, respectively), and agreed well to findings of previous laboratory experiments using the same bivalve species (+0.50 µm² /°C; Höche et al., 2021a). All this argues for a causal link between BMU size and temperature, which functions analogously to abiogenic systems, where rising temperature also leads to increased grain size of the precipitated carbonate (Mejri et al., 2014). In the Icelandic specimens, differences in BMU size between individuals of the same age could be explained by temperature differences. For example, specimen ICE06-6.2-A11R formed, during age three to five (corresponding to 2001-2003), BMUs on average 8.8 µm² larger than specimen ICE06-6.2-A6R at the same age (corresponding to 1997-1999, Figures 5A,C), because temperatures were on average 1.1°C warmer (5.34 and 4.19°C, respectively; Supplementary Material). However, based on the previously determined relationship, this temperature difference would only account for ca. 0.6 µm² of the observed 8.8 µm². The intercepts of linear models between BMU size and temperature also differed strongly between specimens and sites (Figure 8A). Furthermore, though statistically significant (p < 0.05), temperature was only weakly correlated to BMU size (r = 0.28–0.29; Figure 8E) suggesting that other environmental and/or biological factors controlled the BMU sizes. In conclusion, BMU size in naturally grown specimens of A. islandica can only serve as a proxy for relative temperature changes.

In contrast to NE Iceland and the Viking Bank, the BMU size of the shell from the Norwegian Trench was negatively correlated to the water temperature, whereas BMU data of shells from the Baltic Sea showed no significant link with temperature (Figure 8A). In case of the Norwegian Trench (243 m depth), this may partly be explained by the lack of high-resolution temperature data, i.e., only monthly average values were available. For comparison, at NE Iceland, the BMU size correlated stronger to the daily temperature data than to the monthly averages (r of 0.29 versus 0.21; r ² of 0.09 versus 0.04). However, at the Viking Bank (100 m depth), BMU size was significantly linked with temperature even though monthly average water data were used. Hence, at the Norwegian Trench and the Baltic Sea, temperature effects on the shell microstructure were minor, and other, unknown factors played a larger role.

In the specimen from the Norwegian Trench, observed negative correlations between BMU size and water temperature were likely spurious, because seasonal minima or maxima of BMU size did not coincide with those of temperature (Figures 2B, 5D). A causal link also seems unlikely because all results published to date indicate the formation of larger BMUs at higher temperatures, rather than the opposite (Milano et al., 2017b; Gilbert et al., 2017). Perhaps, seasonal temperature variability at the Norwegian Trench (∼1.2°C; Figure 2B) was insufficient to produce a seasonal BMU size pattern. However, in specimens from the Viking Bank, seasonal fluctuation of BMU sizes was clearly related to the seasonal temperature oscillation of approx. ±2.2°C. Taken together, this could imply that temperature changes smaller than 1.2–2.2°C may not visibly affect the shell microstructure of A. islandica.

Instead of temperature, dietary conditions could have controlled BMU size, because at the Norwegian Trench, BMU size was weakly positively correlated to the chlorophyll a concentration (r = 0.32, r ² = 0.04; Figure 8). For a variety of reasons, however, food availability was challenging to assess at the study sites. Firstly, as a deposit feeder, A. islandica can rely on various food sources (Winter, 1970; Cargnelli, 1999; Morton, 2011). Thus, a range of different indicators, such as nutritional value of various primary producers in the water, particulate organic matter content, and hydrogeography, should ideally be used to assess food availability. These data, however, are not widely available, so that chlorophyll a concentration was used in this study. Secondly, chlorophyll a concentration does not always represent a good estimation of algal biomass (Kruskopf and Flynn, 2006). And lastly, in situ chlorophyll a data were only available for the Baltic Sea, whereas less precise, remotely sensed satellite data were used for all other localities. For example, satellite data are likely less suitable to characterize the dietary conditions at the Viking Bank (100 m depth) and Norwegian Trench (243 m depth), which are both located well below the photic zone. Chlorophyll a data for NE Iceland were also less reliable, because satellite data are challenging in these optically highly complex coastal surface waters (IOCCG, 2000; Chen et al., 2013; Gholizadeh et al., 2016). Accordingly, quality and resolution of the chlorophyll a data were not sufficient to assess seasonal and interannual variations in food availability. Thus, positive correlations between BMU size and chlorophyll a at the Norwegian Trench and Viking Bank should be interpreted with care. Furthermore, at other localities, e.g., NE Iceland, the BMU sizes were negatively correlated to the chlorophyll a concentration (Figure 8B). A link between food supply and BMU size thus seems highly unlikely, as has also been confirmed by recent laboratory experiments (Milano et al., 2017a; Ballesta-Artero et al., 2018). At most, the deeper water of the Norwegian Trench could represent a special case in which the BMU size is more strongly linked to food availability than in surface waters. This scenario is supported by the fact that seasonal BMU size maxima occurred during autumn at this locality (Figure 5D), the time at which the seasonal thermocline destabilizes and food items are transported to the bottom water (Pohlmann, 1996; Meyer et al., 2011; Rovelli et al., 2016).

Bivalves from the Baltic Sea, however, likewise did not show a significant correlation between BMU size and temperature (Figure 8A) despite warm waters and highly variable temperatures (Figure 2B). However, as demonstrated in laboratory experiments, under normal marine conditions, BMUs of A. islandica shells are larger in warmer waters (Höche et al., 2021a). Perhaps, the BMU size of specimens from the Baltic Sea is still weakly linked to temperature, but effects were overruled by those of other environmental stressors unique to this locality and absent under controlled conditions. These stressors could be low and variable salinity (Supplementary Material), episodic oxygen depletion and eutrophication (Gray et al., 2002; Karlson et al., 2002), high pCO₂ (acidification) or pollution (Kremling et al., 1997). pCO₂ was shown to leave the shell microstructure of A. islandica largely unaltered (Hiebenthal et al., 2013; Stemmer et al., 2013). Effects of the remaining environmental aspects on the microstructure are poorly studied. At dump sites polluted by trace metals, mortality of A. islandica increases (Kapp, 1980), and juveniles do not settle (Liehr et al., 2005). Polluted sediments thus appear to be unfavorable for this species (Leipe et al., 2005) and may negatively affect its biomineralization processes resulting in smaller BMUs. While A. islandica can survive low-oxygen conditions to some extent (Theede et al., 1969; Philipp et al., 2012), metabolic rates and ventilation rates decline substantially below dissolved oxygen (DO) concentrations of 5 mg/L (Taylor and Brand, 1975a; 1975b). During hypoxia (i.e., DO < 2 mg/L), respiration becomes anaerobic, so that body functions almost come to a complete halt (Oeschger, 1990; Strahl et al., 2011). The extrapallial fluid becomes increasingly acidified (Oeschger, 1990; Strahl et al., 2011), so that biomineralization stops. Since at the Baltic Sea, DO concentrations occasionally fell below 5 mg/L, but rarely below 2 mg/L (Supplementary Data Sheet 1), smaller BMUs may have been formed due to restricted body functions. Low and variable salinity likewise decreases shell growth rate (Hiebenthal et al., 2012) and shortens longevity (Begum et al., 2010). Hence, in the Baltic Sea, smaller BMUs may have been formed due to the low and variable salinity (approx. 19.2), when compared to that of all other studied localities (approx. 34.4; Supplementary Data Sheet 1). Even if the ultimate cause of the reduced BMU size at the Baltic Sea cannot be identified here, the smaller BMUs are likely the result of physiological stress.

To summarize, at Iceland and the Viking Bank, the BMU size was affected by changes in water temperature. The temperature sensitivity of BMU size was comparable among these localities (0.5–0.62 µm² /°C), suggesting that BMU-based temperature reconstructions can be completed in fully marine, unpolluted settings. The correlation between BMU size and temperature, however, was weak (Pearson r = 0.21–0.26 with monthly averages; 0.29 with daily temperature data) and did not occur at the other studied sites. In the deeper waters of the Norwegian Trench, temperature variability (approx. 1.2°C) was likely too low to evoke significant microstructural changes, whereas at the Baltic Sea, temperature effects were probably overruled by environmental stressors such as the low and variable salinity and/or reduced dissolved oxygen content. BMU size ranges of individuals from the different sites did not reflect the temperature regimes in which they lived. This suggests, that BMU size is not solely controlled by temperature.

At all localities except the Baltic Sea, a weak positive link existed between BMU size and annual growth rate (Figures 8C,D). BMU size ranges of specimens from the respective localities reflected shell growth rate ranges (Figures 3E, 4B). Moreover, in the hinge plate, BMU size and shell growth rate varied non-linearly through lifetime (Figures 3C, 5A,B). At first sight, this argues for a coupling between the shell growth rate and the BMU size. However, the following observations speak against this hypothesis. Firstly, the broadest annual increments were formed two to 4 years before the largest BMUs were produced (Figure 9A). Secondly and more importantly, BMU sizes in the hinge plate were nearly the same as in the ventral margin (Figure 7A), even though the latter grew about five times as fast (Höche et al., 2021b). Irrespective of whether a direct or an indirect link exists, BMU size was evidently influenced by the physiology of the bivalve.

Perhaps, BMU size and shell growth are regulated by the same underlying biological processes, so that both parameters develop similar patterns over lifetime (Figures 3C, 5A,B) and are coupled to the same environmental variables (Figure 8E). This hypothesis seems reasonable because both BMU size and shell growth are related to biomineralization performance. Bivalve biomineralization is controlled at many levels by the cellular activity of the organism, be it in production and secretion of the organic matrices, in the uptake and transport of carbonate precursor ions (i.e., Ca²⁺ and HCO₃ ⁻), or in the establishment of the chemical boundary conditions necessary for mineralization (Wilbur and Saleuddin, 1983; Marin et al., 2012). Organic components, calcium ions and carbonate ions can be acquired via food uptake or filtration (Wilbur and Saleuddin, 1983; Marin et al., 2012), so that the supply of shell material is bound to the feeding and respiration activity of the bivalve as well as its assimilation efficiency (Winter, 1978). Furthermore, HCO₃ ⁻ ions can be hydrated from metabolized CO₂ (Marin et al., 2012), so that HCO₃ ⁻ supply is coupled to the metabolic rate. Metabolic energy also supports transmembrane Ca²⁺-ATPase-pumps, which are required to transport Ca²⁺ ions (Carré et al., 2006) and to extrude them into the extrapallial fluid in order to reach supersaturation (Marin et al., 2012; Stemmer et al., 2019). Similar ATP-ase pumps are also involved in the removal of H⁺ from the calcification site for pH regulation (Marin et al., 2012; Stemmer et al., 2019). Since the bivalve metabolism supports these processes, improved metabolic rate ultimately results in accelerated shell growth and, perhaps, larger BMUs. This hypothesis seems plausible because numerous other proxy systems of biogenic carbonates are also influenced by metabolism. For example, it has been suggested that metabolism controls the Sr/Ca ratios and carbon stable isotope composition of aragonitic bivalve shells (Purton et al., 1999; Gillikin et al., 2005, 2007), including A. islandica (Schöne et al., 2011; Wanamaker and Gillikin, 2019).

The increase of BMU size and shell growth rate during youth can be explained by changes in metabolic rate, as the latter increases with body mass following a power law (respiration = a × mass^3/4; Figure 9B; Kleiber, 1932; da Silva et al., 2006; Ballesteros et al., 2018). With increasing ontogenetic age, a larger amount of energy becomes available that could support the formation of larger BMUs and faster shell growth. However, BMU size and shell growth rate decrease after the age of ca. four to nine (Figures 3C, 5A,B), whereas metabolic rates of A. islandica do not seem to decline before the age of 100 (Begum et al., 2009; Basova et al., 2012). Moreover, while it was observed that Mytilus edulis down-regulate mitochondrial respiration, ATP production and pH regulation after ca. age five to six (Sukhotin and Pörtner, 2001; Sukhotin et al., 2020), this is likely not the case in A. islandica (Begum et al., 2009). Decreases in BMU size and shell growth rate after age four and seven are thus not simply caused by a decrease in metabolic energy. Instead, according to dynamic energy budget models, during lifetime, A. islandica increasingly shifts its metabolic energy expenditure from shell production toward reproduction (Figure 9B; Abele et al., 2008; Begum et al., 2010). Accordingly, ontogenetic patterns in BMU size and shell growth rate can reflect the amount of metabolic energy available for shell production, which reaches peak values during early stages of life (Begum et al., 2010; Ballesta-Artero et al., 2018).

BMU size trends during the first ca. 20 years of life can thus correspond to the changing amount of energy available for shell growth. However, even if a direct link does not exist, an understanding of the early physiological developments of A. islandica, upon which the aforementioned metabolic budgets were based (Begum et al., 2010), can help to predict and detrend the ‘ontogenetic bias’ of the BMU sizes. Just after settlement of the larvae, metabolism and shell growth occur only at slow rate due to the small body size of the individuals (Figure 9). However, during this early stage of life, shell growth also accelerates most rapidly, because it receives the largest proportion of the increasing metabolic power in order to improve individual survival chances (Kooijman and Kooijman, 2010; Ballesta-Artero et al., 2019). Accordingly, the shell of young A. islandica grows faster than its soft body (Murawski et al., 1982). Shell growth rate reaches a maximum already at around age four (Figure 9). BMU sizes increase concurrently with the metabolic rate and shell growth rate, but formation of the largest BMUs is delayed by two to 3 years. Perhaps, larger BMUs can only form after the first few years of life, when lateral gshell growth rates exceed a certain threshold. Once the shells are large enough to escape predation, shell growth becomes less important and metabolic priority shifts toward soft body growth and maturation (Figure 9B; Kooijman and Kooijman, 2010). Accordingly, the maximum somatic production activity is attained after the formation of the largest BMUs. After maturation, shell growth reaches the negative asymptotic stage (Philipp and Abele, 2010) because the majority of available metabolic energy is now directed toward reproduction (Begum et al., 2010; Kooijman and Kooijman, 2010; Ballesta-Artero et al., 2019). However, the onset of sexual maturity varies greatly between localities and among specimens (Thompson et al., 1980b; Cargnelli, 1999; Thorarinsdottir and Steingrimsson, 2000), whereas BMU size maxima were always attained around age seven to nine (Figure 5). Perhaps, not enough shells of each population were studied in order to portray the full temporal variation of the BMU size maxima. Still, ontogenetic BMU size patterns can be assessed relatively precisely by corresponding shell growth rates. Alternatively, the timing of ontogenetic BMU size changes can be assessed by habitat-specific models of metabolic budgets or by estimates of longevity, age of maturation and shell size to escape the predation window.

Differences in BMU size range among individuals of the studied localities can also be explained by regional differences in shell growth and/or metabolic rate. At NE Iceland for example, BMUs were large and variable in size (Figure 4A) and shells grew fast (Figure 3C), due to the relatively warm and variable temperatures (Figure 2B). Accordingly, the total metabolic throughput estimated for A. islandica of this locality is also quite large (approx. 43,700 kJ; Begum et al., 2010), because warmer temperatures increase metabolic performance (Gillooly et al., 2001). On the other hand, BMUs of specimens from the Baltic Sea were small despite substantial temperature variability (Figures 2B, 4B), which can be explained by the negative effects of low and variable salinity and/or DO content on the metabolism of A. islandica (Taylor and Brand, 1975b; Oeschger, 1990; Strahl, 2011; Basova et al., 2012; Hiebenthal et al., 2012), leading to slow shell growth and formation of smaller BMUs. Lower salinity and oxygen levels have also been found to impair metabolism and shell formation processes of other bivalve species (Sobral and Widdows, 1997; Sillanpää et al., 2020), supporting this hypothesis. Accordingly, estimated lifetime metabolic budgets of Baltic Sea populations are much lower than those of fully marine ones (∼300 kJ at Kiel Bay, approx. 70 km NE’ the site of this study; Begum et al., 2010). Mass-specific metabolic rates, filtration rates, and oxygen consumption rates, in contrast, are all enhanced under physiological stress (Taylor and Brand, 1975b; Basova et al., 2012). However, most of the excess metabolic energy is used for coping with stress and reproduction to ensure population survival (Begum et al., 2010), so that the amount of energy used for shell formation does not increase and BMUs remain small. Lastly, populations of low longevity show almost no metabolic reaction to warming (Begum et al., 2010; Basova et al., 2012) which could explain why Baltic Sea specimens lack a correlation between BMU size and temperature (Figure 8A).

In summary, BMU size of A. islandica is strongly coupled to its physiology, which complicates temperature reconstructions based on the shell microstructure. Changes in BMU size during the first ca. 20 years of life corresponded to variations in shell growth rate and/or shifts in metabolic activity. This early stage of the shell growth record should thus be excluded in BMU-based proxy reconstructions in order to minimize ontogenetic bias. Suboptimal growth conditions, i.e., low salinity and/or low DO conditions, also prevented the formation of large BMUs, likely because physiological stress exerted a negative control on metabolism and shell growth. Hence, BMU size only serves as a temperature proxy in fully oxygenated, fully marine settings.

The different microstructure types formed BMUs of distinct size and shape. Under the same environmental conditions and at the same ontogenetic age, BMUs of homogeneous microstructure are comparatively small, FCCL BMUs are moderately sized with little size variation but highly elongated, and CA BMUs are larger and more variable in size and of moderate elongation (Figures 4B, 7B). Within one individual, BMUs of the same microstructure formed at the same time varied only slightly in size, regardless of shell portion (hinge and ventral margin) or position within the shell layer (Figures 6, 7), except for two increments of specimen ICE06-6.2-A6R. If BMU size is indeed coupled to the metabolism, the size heterogeneity within these increments might be caused by variations of metabolic rate in different portions of the mantle. A similar effect exists, for example, in Mytilus edulis, where at lateral margins, trace element incorporation into the shell is more strongly controlled by metabolism than along the ventral margin (Klein et al., 1996). However, only the seventh increment of the hinge plate of ICE06-6.2-A6R was substantially curved (Figure 6A), whereas increment eight was not. Still, BMU sizes of different transects through increment eight correlated poorly, so that variations in shell curvature and associated changes in metabolism cannot explain the BMU size heterogeneity.

Baltic Sea specimens formed FCCL microstructures much more frequently than specimens of the other localities, at which CA was predominantly deposited. Since FCCL BMUs are smaller than those of CA microstructures (Figure 4B), BMU size data of Baltic Sea specimens are generally lower than that of the remaining studied localities. Perhaps, correction factors can be applied to facilitate comparisons of BMU size data of different microstructures. Interestingly, FCCL microstructures were mostly formed after disturbance lines and annual lines, i.e., during times of slow shell growth and reduced biomineralization rate. This aligns with the hypothesis that physiological stress reduces metabolism and leads to the formation of small BMUs. Environmental stressors likely evoked the formation of an entirely different microstructure (FCCL instead of CA), as was previously postulated in other studies (Prezant et al., 1988; Tan Tiu and Prezant, 1989; Dunca et al., 2009).

To summarize, each microstructure type (homogeneous, crossed-acicular and fine complex crossed-lamellar) exhibits a specific BMU size range. In order to use BMU size as a proxy for temperature or other environmental or physiological variables, respective measurements should be completed in microstructurally identical shell portions (e.g., CA in iOSL and hinge plate). Otherwise, corrections may be needed. When selecting shell portions for BMU-based proxy reconstructions, higher temporal resolution can be achieved in the ventral margin than in the hinge plate. However, due to microstructural gradation of the ventral margin of A. islandica, preparation quality varied more strongly and image targeting was more challenging to accomplish than in the hinge plate.