For sclerochronological analyses, three ca. 3-mm-thick sections were cut from one valve of each specimen along the axis of maximum growth ( Supplementary Figure 1 ). One of the sections was used for growth pattern analyses, the mirroring side for in situ chemical analyses, and the remaining shell slab for microstructure investigations. Before sectioning, the shells were covered along the cutting axis with an approx. 1–1.5-cm-wide and 0.3-cm-thick protective layer of metal epoxy resin (WIKO EpoFix 05). To mount the shells for cutting, acrylic glass blocks were attached to the anterior side of the valves using a plastic welder (WIKO Multi Power 03). Specimens were then sliced at approx. 200 rpm using a rotational low-speed saw (Buehler IsoMet 1000) equipped with a 0.4-mm-thick diamond-coated blade (Buehler 15LC 11-4255). After cutting, the shell slabs were ground on glass plates using F800 and F1200 SiC grit powders (suspended in deionized water) and then polished on a polishing cloth (Buehler MasterTex) using a 1 µm Al₂O₃ suspension. After each grinding and polishing step, the samples were ultrasonically rinsed with deionized water. Sections used for growth pattern analysis were then mounted to glass slides using metal epoxy (WIKO EpoFix 05).

For growth pattern analysis, one polished slab of each specimen was immersed in Mutvei’s solution (Schöne et al., 2005b) for 20 minutes at 37°C to 40°C under constant stirring. This treatment facilitated the recognition of annual growth lines. Stained sections were then photographed with a digital camera (Canon EOS 600D) mounted to a reflected-light binocular microscope (Zeiss Stemi 508). Samples were illuminated using sectoral (= one-quarter) dark-field ring-light illumination (Schott VisiLED MC1000). Growth pattern analyses were carried out on stitched images (Microsoft Image Composite Editor) using the image analysis software ImageJ (Schneider et al., 2012; Schindelin et al., 2015; Rueden et al., 2017). The number of annual increments provided information on the lifespan of the studied specimens and helped to assign each shell portion to a precise calendar year. The time of the year represented by each SEM image was determined based on the relative position between consecutive annual growth lines and existing seasonal growth models for the study sites (Schöne et al., 2021; Höche et al., 2022a). For details on the seasonal dating procedure, see also Schöne et al. (2021). These growth models provide information not only on the timing of seasonal shell growth but also on daily growth rates (in %).

Shell microstructures (see review in, e.g., Checa, 2018) consist of hierarchically organized BMUs, which can be rounded as in the homogeneous microstructures (HOMs) or needle-shaped as in crossed-lamellar, fine complex crossed-lamellar (FCCL) and crossed-acicular (CA) microstructures. Each BMU is enveloped by intercrystalline organic matrices and interwoven with intracrystalline organic matrices (proteins, lipids, sugars, etc.) (Bevelander and Nakahara, 1980; Marin et al., 2012).

In the present study, different preparation techniques were employed for qualitative and quantitative analyses ( Table 1 ) of the microstructural properties of the shells. For qualitative analyses, the third polished shell section of each specimen was briefly (10 seconds) immersed in diluted (1 M) hydrochloric acid to generate a topographical relief highlighting the BMUs. Immediately after etching, specimens were rinsed with demineralized water, air-dried, and then coated with a 5-nm-thick platinum layer using a sputter coater (Leica EM ACE200). For quantitative BMU analyses, the same shell slabs were first repolished with 1 µm Al₂O₃ suspension and subsequently ultra-polished with an alkaline polishing agent (Buehler MasterMet, 60-nm grain size) and sputter-coated with Pt.

Microstructure investigations were conducted with a Desktop SEM (Phenom Pro, third generation) equipped with a CeB₆ electron source backscatter electron detector at 10 keV. For qualitative assessment of the HCl-etched shells, mappings of large shell regions were generated automatically by stitching up to 400 individual images together. Each SEM image was taken at approx. ×3,500 magnification. Quantitative measurements of the BMU shape were performed on individual SEM images taken at a magnification of approx. ×7,500. When viewed under backscatter SEM, the intercrystalline organic envelopes appeared black, while BMUs appeared in shadings of gray reflecting variations in the amount of intracrystalline organics. This approach greatly facilitated the quantitative BMU analysis.

SEM images taken for quantitative analysis of the BMU size and shape were fed into the machine learning-based image segmentation software ilastik (Berg et al., 2019). For details on how this software can assist in microstructure analysis, the reader is referred to (Höche et al., 2022a). In brief, the program classifies each pixel of the SEM images ( Supplementary Figure 2A ) as belonging to either a BMU or the interstitial space (= organic envelope, intercrystalline organic matrix) based on user-supplied examples and returns an image where the pixels of each BMU are labeled in an individual color, while the background is labeled black. This color-coded image ( Supplementary Figure 2B ) can then be used as an input for automatic particle analysis to obtain the morphological parameters of each BMU. Particle analysis was performed using the python programming language and the scikit-learn image analysis library (as well as the image analysis software ImageJ using the plugin MorphoLibJ). The trained classifier used by ilastik is available in Höche et al. (2022b).

In addition to the BMU size, width, length, elongation, perimeter, and coverage, which have already been studied in prior publications (e.g., Milano et al., 2017; Ballesta-Artero et al., 2018; Höche et al., 2021; Höche et al., 2022a), several novel shape parameters were investigated in the present study, namely, i) the perimeter-to-size ratio of the BMUs, which is a two-dimensional representation of the surface-to-volume ratio of three-dimensional particles, and ii) the BMU solidity, i.e., the size of the BMUs in relation to the size of the smallest convex polygonal envelope of the BMU (= BMU size/size of convex hull; Supplementary Figure 2C ). This value decreases when a smaller proportion of the convex hull is filled out by the BMU, i.e., when the BMU has more crevices that are not filled with CaCO₃.

For specimens from the Baltic Sea, correlation analysis was performed between the morphological parameters of the BMUs and environmental parameters. For this purpose, Spearman’s rank correlation was used because BMU morphology data were not consistently normally distributed. In Baltic Sea shells, the properties of microstructures formed under well-oxygenated conditions were compared to such formed during low-oxygen conditions during summer. For this purpose, the time covered by the shell growth records was categorized into “high-DO” and “low-DO” intervals based on the oxygenation status during biomineralization. Since we do not know the DO concentration of the water inhaled by the bivalves (see also Section 4.1), the average summer DO concentration of the study region (Mecklenburg Bight and Fehmarn Belt) at 5 m above the seafloor (where actual DO measurements were completed), i.e., 5 mL/L, was set as the boundary between “high-DO” and “low-DO”. Very likely, the formation of disturbance lines is coupled to the metabolic rate. According to Strahl et al. (2011), in A. islandica, the metabolic rate decreases markedly below DO values of 5 kPa, which likely corresponds to the threshold set for the categorization into high- and low-DO time intervals. An oxygen partial pressure of 5 kPa (measured at 1 atm and 10°C; Strahl et al., 2011) corresponds to 0.43 mL/L, 1.62 mg/L, or 44.93 µmol/L DO assuming the environmental conditions of the Mecklenburg Bight, i.e., a temperature of 7.5°C, a salinity of 19.3 (averages during the studied time interval), and water depth of 24.8 m (converted using Henry’s law and the ideal gas constant). Subsequently, all BMU data of the “low-DO” category were compared to those of the “high-DO” category using non-parametric Mann–Whitney U-tests. An analogous comparison between low- and high-salinity conditions was not performed because salinity only causes severe stress in A. islandica below values of approx. 15 (Hiebenthal et al., 2012). Furthermore, salinity in Mecklenburg Bight only fell below this threshold for 25 days during the entire studied time interval.

For comparison with the microstructure data, instrumental temperature, salinity, DO, and chlorophyll a (Chl a) data were acquired for Þistilfjörður, NE Iceland, and Mecklenburg Bight, western Baltic Sea ( Table 1 ). In the case of Þistilfjörður, temperature data were obtained from Grímsey station (Hanna et al., 2006), and salinity data were taken from satellite measurements (Gaillard et al., 2016). High-resolution DO time series were not available for the site at NE Iceland, but given that the habitat belongs to the well-agitated surface water body and the sandy substrates prevail, permanent well-oxygenated conditions can be assumed. According to the World Ocean Atlas (https://www.ncei.noaa.gov; last access Nov. 21, 2022), surface waters in that region contained ca. 0.35 mol O₂ per g, corresponding to 11.2 mg/L or 4.9 mL/L O₂ at the water depth in which the bivalves lived, i.e., 6.6 m, and the local water temperature (see below). For Mecklenburg Bight, temperature, salinity, and DO concentration data were taken from in situ measurements in a water depth of 20 m at Station 12 (Zettler et al., 2017; online database, reference Leibnitz-Institut für Ostseeforschung Warnemünde, 2021, at https://odin2.io-warnemuende.de/, last access Nov. 20, 2022), located ca. 172 m NW of the sampling site. Note that these DO data may overestimate the actual oxygen availability at the seafloor where the bivalves dwelled (see also discussion in Schöne et al., 2022). Within the timespan covered by the shell records (1986–2006; Table 1 ), the instrumental DO dataset ( Figure 2 ) comprises 179 points of data (on average, ca. nine per year), with 20 of them falling into summer months (i.e., one per year). All environmental datasets are provided in Höche et al. (2022b).

The water at the study site at Mecklenburg Bight was, on average, >3°C warmer than at NE Iceland (7.4°C ± 3.5°C vs. 4.3°C ± 2.3°C; 1983–2012) but significantly lower in salinity (19.4 ± 2.7 vs. 34.4 ± 0.2) and contained highly variable amounts of dissolved oxygen (2.0–8.9 mL/L vs. ca. 4.2–5.1 mL/L) (Höche et al., 2022b). Food availability (Chl a) was nearly identical at both study sites (Mecklenburg Bight, 1.8 ± 0.7 mg/m³; NE Iceland, 1.8 ± 0.8 mg/m³) but more extreme in the Baltic Sea (0–10.3 vs. 0.6–4.7 mg/m³). Seasonal temperature and salinity amplitudes at Mecklenburg Bight (−0.4°C to 16.9°C; 14.9–28.9) clearly exceeded those at NE Iceland (−1.1°C to 10.3°C; 34.4–34.7). During summer, the study site at Mecklenburg Bight experienced temperatures as high as 12.2°C ± 2.1°C and salinities, Chl levels, and DO concentrations of 23.6 ± 2.7, 1.76 ± 0.02 mg/m³, and 3.0 ± 1.2 mL/L, respectively.

In the western Baltic Sea, shells of A. islandica grew up to 0.6% of the annual growth increment within a single day (Höche et al., 2022a). In general, daily growth rates increased with higher oxygen supply (r = 0.55, p < 0.05; Figure 4A ). The fastest shell production (0.3%–0.6% of annual growth per day) was observed at DO concentrations above ca. 6 mL/L. Below this threshold, however, no more than 0.25% of the annual growth increment was produced per day. Shell growth was also negatively correlated to water temperature (r = −0.54, p < 0.05; Figure 4B ). The fastest growth rates occurred between 2.5°C and 5°C. In contrast, in Iceland, growth rates were weakly positively linked to the water temperature (r = 0.22, p < 0.05), and daily shell growth varied between 0.1% and 1% of the annual increment ( Figure 4B ).

Irrespective of locality, the outer shell layer (OSL) of the ventral margin of the studied specimens ( Table 1 ) was subdivided into microstructurally distinct sub-layers ( Figure 3 ), i.e., an outer portion (oOSL) dominated by HOM ( Figures 5A, D, G, J ) and an inner portion (iOSL) in which CA microstructure prevailed ( Figures 5B, K ). From the oOSL/iOSL transition (i.e., the boundary between HOM and CA microstructure) toward the myostracum, individual BMUs became gradually smaller and more elongated. In the innermost part of the iOSL, i.e., the slowest growing portion near the myostracum, FCCL microstructure was observed ( Figures 5C, L ).

In contrast, the hinge plate (portion formed in the outer extrapallial space (i.e., the same from which the oOSL of the ventral margin was precipitated) was microstructurally more uniform than the ventral margin and showed less gradual microstructural transitions ( Figure 6 ). CA microstructure prevailed, with occasional intercalations of FCCL microstructure ( Figures 6A, D, G, J ). In shells from both study sites, the annual growth lines in the hinge and ventral margin were typically composed of irregular simple/spherulitic prismatic (ISP) microstructure (e.g., Figures 6B, C, E, F, I, K ), but significant differences were detected regarding their relative proportions. In the following, these differences will be described in more detail.

The morphology of BMUs was quantitatively assessed in a total of 1,875 SEM images taken in six shells ( Table 1 ). In the following, the relationships between different BMU shape parameters will be described. Then, the BMU shape data of the two localities will be contrasted with each other, and finally, for specimens from the Baltic Sea, data of shell regions formed under low and high DO will be compared with each other.

Disturbance lines were exclusively found in specimens from the Baltic Sea, specifically in shell portions formed under severe oxygen depletion during summer ( Figures 2 , 3 ). None of the other environmental variables for which instrumental data were available caused comparable shell growth patterns at any other study locality. For example, the study site at NE Iceland is episodically strongly influenced by nutrient-poor polar waters (Marali and Schöne, 2015). Despite that, the bivalves did not form disturbance lines ( Supplementary Figure 5 ) but just grew slower (Marali and Schöne, 2015). Likewise, no disturbance lines were formed in the present study at temperatures close to 0°C or salinities as low as 14.6 ( Figure 10 ), conditions which are well-known to cause physiological stress and slow growth in ocean quahogs (Witbaard et al., 1998; Hiebenthal et al., 2012). In shells from the western Baltic Sea, disturbance lines formed in summer despite water temperatures (−0.5°C to 15.9°C; Figure 10 ) were near the species-specific optimum (6°C to 16°C; Merril et al., 1969; Witbaard et al., 1998; Cargnelli, 1999) and salinities (16.7–28.9 between June and August; Figure 10 ) well above values that exceed the species-specific tolerance (approx. 10–12: von Oertzen and Schulz, 1973; Zettler et al., 2007; Darr et al., 2014). Given that the Baltic Sea is eutrophicated, the bivalves should also not have experienced any lack of food supply that could have led to disturbed shell growth in summer. While high algal densities were reported to negatively affect the feeding behavior of A. islandica (Winter, 1970), this effect is probably negligible in comparison to the effects of an actual lack of food supply. It is also unlikely that the disturbance lines resulted from predatory attacks. First, growth distortions related to predation are typically associated with distinct notches and damages on the outer shell surface (e.g., Kennish and Loveland, 1980; Gaspar et al., 1994; Ramsay et al., 2001), which were not observed. Second, predatory attacks should not be limited to low-DO times but occur year-round. Third, predation, for instance, by predatory fish (Arntz and Weber, 1970; Rumohr and Krost, 1991), is also common in other localities outside the Baltic Sea as well. The only reasonable remaining environmental factor that could have triggered the formation of the regularly developed disturbance lines in Baltic Sea shells is severe oxygen deficiency. This conclusion is supported by the concurrent enrichment of manganese in the shells as well as an inverse relationship between the prominence of disturbance lines and DO concentration (and the height of Mn/Ca_shell peaks).

The rationale for using Mn/Ca ratios of bivalve shells as a surrogate for DO is as follows. Under reducing conditions (e.g., within the sediment), manganese becomes dissolved in the water column (Hem, 1963; Balzer, 1982; Rue et al., 1997) and is biologically available as Mn²⁺. Sedimentary porewater diffuses out of anoxic sediments (Ben-Yaakov, 1973; Kremling et al., 1997), mixes with water above the substrate, and may be inhaled by A. islandica (and other filter-feeding benthos). Dissolved manganese and other redox-sensitive elements can then reach the body fluids of the bivalves and eventually be incorporated into the shell. In the case of Mn²⁺, this occurs swiftly after uptake from the water (Langlet et al., 2007). Mn²⁺ concentrations are highest near the sediment–water interface and gradually decline with increasing distance from the seafloor as well as rising DO concentration in the free water column (Kremling and Petersen, 1978). The correlation between shell Mn/Ca values and DO measured at some distance from the sediment surface (e.g., approx. 5 m above the substrate in the case of Mecklenburg Bight) indicates that the bioavailability of Mn²⁺ near the seafloor is proportional to DO concentrations in the free water column. This relationship was previously evoked to suggest that Mn/Ca_shell serves as a semiquantitative DO proxy in Astarte spp. and A. islandica (Schöne et al., 2021; Schöne et al., 2022).

Due to the lack of direct measurements of DO concentration at the very sites at which the bivalves lived and inhaled water (i.e., the fluffy layer), or even better, the DO content of the coelomic fluid, we can only provide the average DO value that prevailed 5 m above the sediment surface when the disturbance lines were formed (below approx. 5 mL/L ≡ 516 µmol/L at 2.43 atm and 7.5°C). The water inhaled by the bivalves at the seafloor, however, most certainly contained much less oxygen than indicated by the instrumental DO data because of sedimentary porewater efflux. Presumably, DO concentration was less than 45 µmol/L (≡ 5 kPa, 0.43 mL/L ≡ 1.62 mg/L), i.e., the threshold at which the metabolic rate depression of A. islandica starts (Strahl et al., 2011). Controlled rearing experiments in future studies would be very useful to constrain and quantify the relationship between the prominence of disturbance lines and actual DO levels close to the seafloor where the bivalves live.

Disturbance lines were associated with a strong reduction in shell growth rate ( Figure 4 ) and a concomitant formation of FCCL microstructure ( Figures 4 , 5 ). FCCL BMUs became increasingly smaller, more elongated, and less convoluted (= lower surface roughness, higher solidity) as DO levels dropped. The correlation between BMU size and DO fits well with the observation that A. islandica strongly reduces the metabolic rate under low-oxygen conditions (Theede et al., 1969; Taylor and Brand, 1975; Strahl et al., 2011). As a consequence, much less energy would be available to transport Ca²⁺ and HCO₃ ^– toward the site of biomineralization and H⁺ ions away from it as well as to enzymatically convert metabolic CO₂ into bicarbonate ions. This would inevitably lead to a reduction of the BMU size because an active ion transport is crucial to form abundant CaCO₃ (Wilbur and Saleuddin, 1983; Carré et al., 2006). Furthermore, based on the observation that low-DO-induced stress changes the expression of certain genes (Philipp et al., 2012), it is speculated here that low-DO-induced physiological stress also influences the synthesis of proteins responsible for the formation of a specific BMU habit. A change in the type and composition of shell-forming proteins may have been responsible for the production of increasingly smooth biominerals as DO supply dropped. If that hypothesis holds true, there may be scope to detect oxygen deficiency in the past solely by detailed BMU morphometry or even to quantify the DO depletion based on BMU habit. Future work should thus further explore the diagnostic potential of the morphological properties of the BMUs. As shown here, the relationship between DO and BMU size or elongation was statistically significant, but the explained variability was too small and the error too large to be used to reconstruct DO concentration from morphometric data of the BMUs ( Figure 9 ). Perhaps, the surface roughness, size, shape, or other parameters reveal a closer relationship with DO concentration if studied in three dimensions (e.g., Howes et al., 2021; Mehra et al., 2022).

As demonstrated by many controlled laboratory experiments (Oeschger, 1990), under the persistent absence of oxygen (anoxia) or the presence of H₂S (euxinia), A. islandica closes its valves for extended periods of time, ceases biomineralization, and switches to an anaerobic metabolism once all entrapped oxygen is consumed (Crenshaw and Neff, 1969). Acidic metabolites generated by this type of adenosine triphosphate production (i.e., anaerobic glycolysis) are neutralized by the dissolution of inner shell surfaces (Crenshaw and Neff, 1969; Crenshaw, 1972; Oeschger, 1990). This process can readily explain the thinness (ca. 1 mm-tick; Supplementary Figure 1 ) of ocean quahog shells from the western Baltic Sea as formerly hypothesized by Weigelt (cited as a personal communication in Oeschger, 1990).

The shells from the Baltic Sea were not only very thin but also overall enriched in organic matrix compared to specimens inhabiting permanently normal marine, well-oxygenated settings ( Figures 4 , 5 ) (see also Li et al., 2015). Presumably, this is attributed to their lower BMU coverage values ( Figure 7A ) and to the fact that smaller BMUs come with a proportionally larger surface area ( Figure 7C ), which is occupied by an organic matrix ( Figure 7 ). The surplus of organics demonstrably increases the elastic properties of shells (cf. Li et al., 2015), which results in higher toughness and strength (Chen et al., 2008; Meyers et al., 2008). Given that the shells were very thin, this may have served as an effective strategy to withstand predatory attacks. Toward the outer shell surface, the organic content increased as indicated by broader inter-crystalline organic sheets ( Figure 5J ) and the gradual broadening of the organic-rich annual growth bands ( Figure 5J ). This organic enrichment at the shell periphery may have further contributed to the toughness of the shell and could also have provided better protection against shell dissolution. However, porewaters of anoxic sediments are demonstrably buffered, and the pH value rarely drops below 7 (Ben-Yaakov, 1973; Bonaglia et al., 2019). An alternative explanation for higher organic content in shell portions closer to the periostracum is improved resistance against microbial attack. While increased amounts of organics can attract microorganisms resulting in even stronger bioerosion of the shell, in dys/anoxic sediments—which exist at Mecklenburg Bight—the rate of microbial decomposition of organic compounds is strongly reduced relative to a setting where the upper few centimeters of the sediment are well-oxygenated (cf. Hedges and Keil, 1995; Hartnett et al., 1998; Jessen et al., 2017) like at the study site at NE Iceland so that an organic-rich outer rim may still have fulfilled its task as an efficient physical barrier against predation.

The larger proportion of organic matrix relative to CaCO₃ in ocean quahogs from the Baltic Sea can be explained by generally smaller BMUs resulting from overall slower shell growth rates and lower metabolic rates (Strahl et al., 2011). A. islandica from localities outside the Baltic Sea not only grew considerably faster ( Supplementary Figure 6 ; Supplementary Table 1 ) but also produced larger BMUs, specifically in warmer, food-rich water (Höche et al., 2021; Höche et al., 2022a). In the Baltic Sea, however, the positive correlation between BMU size and temperature or food supply seems to have been offset by the negative effects of persistently reduced salinity (and episodically recurring severe oxygen depletion) on shell growth rate. For the stenohaline A. islandica, lowered salinity (in particular below 15) exerts physiological stress that demonstrably leads to slower shell growth (Hiebenthal et al., 2012). While salinity at the study site (approx. 15 to 25) remained within the physiological tolerance of that species (Oeschger and Storey, 1993; Cargnelli, 1999), the osmotic stress may have triggered the rerouting of energy to maintain cellular homeostasis (e.g., osmoregulation; Hiebenthal et al., 2012) instead of supporting shell formation. Alternatively or in addition to that, low salinity demonstrably limits calcification due to increasing energetic requirements (Sanders et al., 2018) for the uptake and transport of calcium and bicarbonate ions across mantle epithelia toward the site of biomineralization and proton removal from the biomineralization front, as well as the catalytic conversion of metabolic CO₂ into HCO₃ ^– by means of carbonic anhydrase (Marin et al., 2012; Sanders et al., 2018). Furthermore, with declining salinity, HCO₃ ^– and, more importantly, the Ca²⁺ concentration decrease, which can impede calcification. For Mytilus spp., a lower salinity threshold of 11 has been reported for calcification (Sanders et al., 2021). Assuming that the production of organic matrices proceeded at nearly similar rates, the reduced production of the mineral phase led to smaller BMUs and a higher proportion of organics vs. CaCO₃ in A. islandica.