Two genetically distinct cod stocks exist in the Baltic Sea, the western Baltic stock (distributed in ICES SDs 22 – 24), and the eastern Baltic stock (distributed in ICES SDs 24 – 32) (ICES, 2022). The distribution and management areas of the two stocks thus overlap in the Arkona Basin (SD 24), with approximately 70% eastern Baltic cod stock (Hüssy et al., 2016a). Since the biology (spawning time, growth, fecundity) of the two stocks differs, we exclusively focus on genetically identified eastern Baltic cod to avoid unwanted bias. Historically, eastern Baltic cod supported a commercially important fishery. Today, the stock size has dropped dramatically, coinciding with large changes in most biological parameters (truncated length distribution, earlier maturation, decrease in body condition, growth, and recruitment), resulting in the closure of the directed fishery since 2019 (ICES, 2022). While the reasons for these changes are not addressed here (reviewed in detail by Eero et al. (2015)), the current status of the stock highlights the importance of gaining a better understanding of the cod’s lifetime habitat use to support stock recovery. Many eastern Baltic cod perform seasonal migrations from coastal feeding areas into deep spawning areas, where the salinity is high enough for the eggs to be buoyant (Nissling et al., 1994) ( Figure 1 ). Today, successful spawning of the eastern Baltic cod stock is largely limited to the Bornholm Basin due to anoxic conditions in other historic spawning sites (Aro, 1989; Bagge et al., 1994; MacKenzie et al., 2000). Eastern Baltic cod’s spawning season is longer than in any other cod stock, typically lasting from May to November (peaking July/August) (Wieland et al., 2000; Bleil et al., 2009) and lasting approximately 4 to 8 weeks for individual fish (Vallin and Nissling, 2000). Outside the spawning season, adult cod are not usually found in the deep basins (Aro, 1989; Bagge et al., 1994). Baltic cod consume prey throughout the year, including during the spawning season (Bagge, 1981), but consumption is greatest post-spawning in fall/early winter (Fordham and Trippel, 1999; Pachur and Horbowy, 2013). Juvenile cod generally occupy shallower areas compared to their larger conspecifics (Sparholt et al., 1991; Pihl and Ulmestrand, 1993; Oeberst, 2008).

The Baltic Sea is a large estuary, with consecutive deep basins with progressively larger maximum depths (53 m in the Arkona Sea, 105 m in the Bornholm Basin, and even deeper further east) separated by shallower sills ( Figure 1 ). The hydrography is characterized by deep-water inflow of saline water from the Atlantic Ocean and outflow of freshwater, resulting in a general salinity gradient from west (high) to east (low) (Meier et al., 2006; Naumann et al., 2020). Salinity also varies three-dimensionally. Riverine freshwater inputs create a low salinity (~7‰) surface layer (typically 0 –50 m depth), which sits on top of more saline bottom waters (10 to 18‰) (Matthäus and Franck, 1992; Schinke and Matthäus, 1998; Meier et al., 2006; Naumann et al., 2020). Salinity conditions are inflow dependent, but the halocline depth has remained fairly constant across multiple decades (Carstensen et al., 2014). As a result of this stable salinity stratification and decades of eutrophication, the extent of hypoxic areas has been spreading as a persistent layer in the deeper basins (Helcom, 2009; Viktorsson, 2018; Hansson et al., 2019; Naumann et al., 2020). Many shallower areas within the Baltic Sea, with restricted exchange of bottom water due to complex bottom topography, also suffer from seasonal periods of hypoxia during summer (Conley et al., 2011; Carstensen and Conley, 2019). Overall, the total volume of hypoxic areas (< 2 mg O₂ l^-1) in the Baltic Sea has been increasing through recent decades (Carstensen et al., 2014; Kõuts et al., 2021). In terms of water temperature, a seasonal thermocline occurs at 30 – 40 m depth in summer (July to September), with a relatively homogeneous mixed surface layer down to the halocline (Matthäus and Franck, 1992; Møller and Hansen, 1994; Schinke and Matthäus, 1998).

The geochemistry of the Baltic Sea is also characterized by a high degree of spatiotemporal heterogeneity. Many elements have a relatively uniform distribution in sea water (Walther and Limburg, 2012; Lebrato et al., 2020), the unique bathymetry and hydrography of the Baltic Sea with the associated variations in salinity and dissolved oxygen, are likely to manifest in strong geographic and temporal gradients in ambient elemental concentrations. For elements under greater environmental control, we hypothesize that these gradients will manifest in predictable patterns in otolith elemental signatures, potentially providing a tool for retrospective geolocation of individual fish.

Fish are complex organisms, and elemental processing and incorporation pathways can vary by system, species, individual physiological state and life stage (Sturrock et al., 2012; Hüssy et al., 2020b). However, many studies have shown strong relationships between water and otolith concentrations, particularly when salinity gradients are more extreme (e.g. Elsdon and Gillanders, 2003). Otolith Mn is also increasingly used to reconstruct past hypoxia-exposure in fish (Limburg et al., 2011; Mohan et al., 2012; Mohan and Walther, 2016; Altenritter et al., 2018; Altenritter and Walther, 2019). Eastern Baltic cod experience considerable variation in their physicochemical environment that are amplified by their seasonal movements between nearshore, shallow feeding grounds and deeper spawning areas ( Figure 2 ). In this study, we take advantage of these sharp environmental gradients alongside general movement patterns of tagged fish to test hypotheses of otolith element incorporation mechanisms in a natural setting, focusing on six commonly measured elements: Strontium (Sr), barium (Ba), manganese (Mn), magnesium (Mg), phosphorus (P), and zinc (Zn). We have broadly grouped these into elements under greater environmental (Sr, Ba, Mn) vs. physiological (P, Mg, Zn) control, and pose a series of hypotheses based on expected environmental gradients and otolith incorporation mechanisms [summarized below and in Sturrock et al. (2012), Heimbrand et al. (2020), and Hüssy et al. (2020b)].

Elements under greater environmental control (Sr, Ba, Mn): Sr, Ba and Mn all form divalent cations that substitute for calcium in the growing otolith crystals (Doubleday et al., 2014; Thomas et al., 2017). Multiple studies have shown that ambient water Sr : Ca and Ba : Ca ratios are strongly correlated with otolith concentrations in laboratory experiments and field studies (Bath et al., 2000; Elsdon and Gillanders, 2003; De Vries et al., 2005; Hamer et al., 2006; Hicks et al., 2010; Reis-Santos et al., 2013), but Mn only in field studies (Thorrold and Shuttleworth, 2000; Dorval et al., 2007; Mohan et al., 2012). Thus, we expect that if we can predict spatiotemporal variations in ambient Sr, Ba and Mn concentrations based on salinity and dissolved oxygen, that these variations should be reflected in the otoliths of cod inhabiting that location at that time.

Strontium: In the Baltic Sea, the lowest salinities are in the east, superimposed by a secondary salinity gradient from coast (shallow) to offshore (deep). In most systems, salinity is strongly correlated with water and otolith Sr concentrations, particularly across the salinity ranges seen in the current study area (i.e. 7 – 20 psu) (Lin et al., 2007; Hicks et al., 2010; Walther and Limburg, 2012; Albertsen et al., 2021).

Bariums: Ba is dissolved from its freshwater particulate state in low-salinity estuarine water, and precipitates as barite at higher salinities (Coffey et al., 1997; Paytan and Griffith, 2007), leading to a coastal-offshore depth gradient. We predict that this depth gradient will result in a negative relationship between water – and otolith – Ba concentrations and depth, and also to higher Ba concentrations in the shallower western than the eastern Baltic.

Manganese: Mn is an active redox participant that cycles between dissolved and particulate phases as a function of pH and dissolved oxygen, with lower pH and oxygen favouring the dissolved forms (Slomp et al., 1997; Trouwborst et al., 2006). Therefore, dissolved Mn concentrations are presumed to be elevated in the deep hypoxic areas in the eastern Baltic Sea, and in shallow coastal areas during summer throughout the Baltic Sea.

Assuming that spatiotemporal variations in salinity and dissolved oxygen result in predictable differences in water and otolith Sr, Ba and Mn in this system (an essential pre-requisite for using these markers for movement and/or environmental reconstructions) we make a series of predictions. Note that – while temperature can influence salinity, dissolved oxygen and fish physiology – effects on otolith element concentrations will be indirect, and thus while we account for temperature in subsequent analyses, we do not explicitly use it to frame these predictions.

Elements under greater physiological control (P, Mg, Zn): These three elements are all either essential constituents of the otolith’s organic matrix, or co-factors in metabolic processes, including those in the endolymphatic epithelium (Wojtas et al., 2012; Sturrock et al., 2013, 2014; Maret, 2017; Thomas and Swearer, 2019). Their ions appear to be primarily incorporated into the otolith randomly trapped in the crystal lattice (Miller et al., 2006; Izzo et al., 2016; Thomas et al., 2017). As such, these elements are assumed to be under strong physiological control, thus – while their ambient concentrations exhibit large horizontal, vertical and seasonal variations gradients (Sylva, 1976; Cox, 1989; Conley et al., 2002; Lebrato et al., 2020; Naumann et al., 2020) – we predict that cod otolith concentrations will be uncorrelated to these and instead driven by endogenous factors such as feeding rate, growth, size, sex, age and/or reproduction. Previous studies have suggested positive relationships between growth rate and otolith Zn (Halden et al., 2000; Limburg and Elfman, 2010), and with otolith P and Mg (Limburg et al., 2018; Heimbrand et al., 2020; Hüssy et al., 2021a). Concentrations in otolith Mg and P were furthermore linked to seasonal temperature fluctuations, with lowest concentration during the coldest months in late winter/spring and highest concentrations in late summer (Hüssy et al., 2021a). Sturrock et al. (2015) also speculated that the drop in plasma and otolith Zn in mature female plaice during the spawning season was caused by rerouting of Zn to the ovaries during vitellogenesis (Sturrock et al., 2014). Thus, for these elements, we suggest the following hypotheses:

The general seasonal patterns in cod movements and expected temperature, salinity and dissolved oxygen concentrations are conceptualized in Figure 2 , alongside the predicted relative element concentrations in the water (low/high) as a function of depth, month and area. Here, we combine otolith and Data Storage Tag (DST) records from two cod tagging programs to estimate the influence of environmental and physiological different drivers on otolith elemental concentrations in situ and to test whether our hypotheses (H1 – 10) are supported or refuted by these data.

The otolith samples used in this study originate from two tag-recapture programs carried out in the Baltic Sea: CODYSSEY (Cod spatial dynamics and vertical movements in European waters and implications for fishery management, 2002 – 2006, https://cordis.europa.eu/project/id/Q5RS-2002-00813) and TABACOD (“Tagging Baltic Cod”, 2016 – 2020, https://tabacod.dtu.dk/) ( Figure 1 ). Details on the tagging protocols of CODYSSEY and TABACOD are summarised in the Supplementary Material (Section 1.1 and Supplementary Table 1 ). Further details may be found in Nielsen et al. (2013) and for TABACOD in Hüssy et al. (2020a, 2021a). Given very low samples sizes (n = 5) for western stock cod, only cod belonging to the eastern stock were used in this study. For stock identification, see Hemmer-Hansen et al. (2019) and Schade et al. (2019). All procedures were carried out under the necessary animal test permissions: German DST tagging: AZ 7221.3.1-007/18; Danish T-bar and DST tagging: 016-15-0201-00929, Swedish T-bar and DST tagging: Dnr 5.8.18-14823/2018.

Days at liberty (DAL) were defined as the number of days between release and recapture. The samples used in this study were restricted to fish with DAL > 30 to ensure sufficient otolith accretion to have multiple chemical records post-release. From CODYSSEY n = 146 recaptured cod in the length range 435 – 960 mm (released in 2003 – 2006) were available and from TABACOD n = 32 recaptures in the length range 298 – 503 mm (released in 2016 – 2018). Release and recapture locations are shown in Figure 1 . For details on length distribution and DAL between tagging and recapture, see Supplementary Figure 2 . After recapture, fish total length (TL, mm), weight (g), sex (female, male) and maturity (ICES 6-stage standard) were recorded. Somatic growth was calculated as change in total length per day G _fish = (TL _recapture – TL _release) DAL ^-1. Sagittal otoliths were removed, cleaned, and stored dark and dry.

In both experiments, the DSTs used measured temperature and pressure (converted into depth) ( Figure 3 ). The DSTs used in the CODYSSEY project also measured conductivity. Star-Oddi micro-TD and milli-TD tags were used for TABACOD, and Star-Oddi DST-CTD for CODYSSEY. While depth was estimated with high resolution, given differences in the geolocations available among individuals and studies, the geographic resolution used for cod movement reconstructions in this study was relatively coarse, with fish assigned to ICES subdivision on a daily scale (SD 24 or 25, Figure 1 ). Further details about the DST data processing are provided in the Supplemental Material and Supplementary Table 1 .

For all fish, mean daily depth was used to classify spawning and feeding season for each individual: Spawning of Baltic cod occurs in the deep waters of the Bornholm Basin (SD 25), while the feeding grounds are found in shallower areas (Aro, 1989; Bagge et al., 1994; MacKenzie et al., 2000). Spawning migrations therefore leave a distinct pattern in the DST records: a sudden change in distribution towards depths below 50 m. In the Baltic Sea, these conditions only exist in the Bornholm Basin. Thus, individual spawning periods were classified as any periods when an individual remained at depths >50 m for more than 7 days during May – September, the known spawning season of the stock (Wieland et al., 2000; Bleil et al., 2009). See Nielsen et al. (2013) for further details on spawning time identification.

Otoliths were marked chemically with strontium chloride in the CODYSSEY program and with tetracycline in the TABACOD program. Methods to identify the chemical mark therefore differ by tagging program. Mark identification of the CODYSSEY samples is based on acquisition of chemical profiles and described under 2.4. Chemical analyses. In the TABACOD samples, mark identification was based on visual identification and described in the following.

Otolith preparation procedures were described in detail in Hüssy et al. (2020a, 2021a) In brief, otoliths were embedded in Epoxy resin (Struers ^®) and sectioned through the core using an Accutom-100 multicut sectioning machine to obtain a 10 mm wide block with the nucleus exposed at the sectioned surface. The TABACOD otolith sections, which were marked with tetracycline, were viewed under UV light using a Leica DMLB microscope (with a BP 355 – 425 excitation filter, magnification of 1.36 μm·pixel^-1, 3648 pixel x 2736 pixel frame) and images digitized using a Leica DCF290 camera ( Figure 4 ). The otolith growth during the tagging period (G _oto) was measured as the distance from the tetracycline mark (or the strontium spike, see below) to the otolith edge along the dorsal axis, while the lifetime growth was inferred via the otolith radius (OR) from the core to the dorsal edge. All measurements were made using Image J ver. 1.54d (Schindelin et al., 2015).

Otolith biomineralization and protein concentrations can result in variations in opacity and differences in element incorporation rates (e.g. Sturrock et al., 2015; Hüssy et al., 2020b). Thus, while not a central hypothesis of the study, we still wanted to account for potential influences of otolith opacity on element concentrations. To measure opacity, images of the otolith sections were taken under reflected light (32-bit black and white images). Otolith optical characteristics were examined as a profile of grey values corresponding to the otolith material deposited during time at liberty. Grey values ranged between 0 (black = translucent) and 255 (pure white = opaque), with high values corresponding to high dispersion of light due to high otolith protein content (Watabe et al., 1982; Seyama et al., 1991). These grey values are hereon referred to as otolith opacity.

Trace element analyses were carried out by laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) at the Geological Survey of Denmark and Greenland, using a NWR213 frequency-quintupled Nd : YAG solid state laser system from Elemental Scientific Lasers that was coupled to an ELEMENT 2 double-focusing, single-collector magnetic sector field ICP-MS from Thermo-Fisher Scientific. The otoliths were analysed along a transect from the core to the dorsal edge of the otolith following the axis of maximum growth. We used a beam diameter of 40 μm, a laser fluence of 9.5 J·cm^–2, a repetition rate of 10 Hz, and a travelling speed of 5 μm·s^-1. Concentrations are reported in element:calcium ratios in ppm, using calcium (⁴³Ca) as an internal standard to account for any variable sample introduction parameters affecting the ablation yield. Values > 4x standard deviations from the mean were treated as outliers and discarded (representing<1% of the data for all elements except for Zn, for which 10% of the data were excluded). Further details on operating conditions, standards used, data acquisition parameters, analytical protocols and data processing techniques are described in Serre et al. (2018) and Hüssy et al. (2021a), and summarised in Supplementary Table 2 . This study focuses on magnesium (²⁵Mg), phosphorus (³¹P), manganese (⁵⁵Mn), zinc (⁶⁶Zn), strontium (⁸⁸Sr), and barium (¹³⁷Ba).

From these line transects, only measurements corresponding to the period between the chemical mark and the otolith edge were selected. In the CODYSSEY samples marked with strontium chloride, a sharp increase in otolith Sr concentration to values > 10 times the reported maximum concentrations of wild Baltic cod (Heidemann et al., 2012; Heimbrand et al., 2020), followed by a rapid subsequent decline, was identified as the chemical mark to use for selecting the appropriate transect data.

To align otolith chemical measurements with calendar dates and associated DST data, we assigned calendar dates using the proportional distance between the otolith measurement and the chemical mark (release date). Eastern Baltic cod do not form clearly defined seasonal otolith growth zones, preventing a direct validation of seasonal otolith accretion (Hüssy, 2010; Hüssy et al., 2016b). The assumption of linear otolith growth was therefore tested – for details see Supplementary materials . Otolith accretion (G _oto) from the chemical mark to the otolith edge was linearly correlated with DAL, without any seasonal patterns in model residuals ( Supplementary Figure 1 ). Thus – while we recognise that individual deviations from this could result in some decoupling of ‘otolith time’ with ‘calendar time’ and some noise in relationships between sub-annual measurements – otolith accretion during the tagging period was largely constant within fish throughout the year. Each individual measurement was then assigned to a date of formation using Equation 1 and based on the measurements distance to the chemical mark, the proportional relationship between G _oto and DAL, using the date of release as the start date:

While the temporal resolution of DST data is high (sub-daily), elemental data typically span several days or weeks, depending on beam size and otolith growth rate. In this study, with a beam travelling speed of 5 μm·s^-1, each chemical measurement represented an advancement of approximately 4 days, but the 40 μm beam size incorporated roughly 10 days worth of otolith material in total. Matching otolith chemical measurements with the corresponding DST-derived data required the DST data to be averaged over the same period as each chemical measurement. Here, we averaged DST measurements from the date of release to the date of formation corresponding to the mid-point of the first chemical measurement, and between consecutive days of formation of the following chemical measurements throughout the rest of the transect data, resulting in average values for temperature T _ij and depth D _ij for each individual i and each measurement j. This approach invariably results in considerable smoothing of the depth and temperature data, potentially masking variation on short temporal scales. Data from the first 7 – 10 days post-tagging, corresponding to the first chemistry measurement, were discarded to avoid effects of the tagging procedure on the behaviour of the fish (van der Kooij et al., 2007).

To examine how much variation in otolith elemental concentrations were explained by environmental and/or physiological variables we fitted linear mixed-effects (LME) models similar to (Sturrock et al., 2015) using the “lme4” package (Bates et al., 2015) in R, ver. R-4.1.3 (R Core Team, 2020). Models were fitted using individual fish and release year as random variable to allow variable intercepts, accounting for inter-individual differences and differences between tagging years (Crawley, 2007; Zuur et al., 2009).

The initial global model [model (2)] included all fixed effects, including area (each individual assignment to western (SD 24) or eastern (SD 25) Baltic ( Figure 1 ), season (spawning/feeding season based on individual depth profiles – see methods), month (month of formation 1 – 12, treated as categorical variable), T (temperature), D (10-m depth stratum, 0 – 10 to 80 – 90, treated as categorical variable), sex (males/females), TL (total length), G _fish (somatic growth between release and recapture), and opacity (0 – 255), with subscripts i representing individual fish and j individual chemistry measurements. Note that G _fish and G _oto (otolith extension distance between release and recapture) were highly correlated (r² = 0.52, df = 1.75, p< 0.001) so only G _fish was included in the global model. For the CODYSSEY samples, where direct measurements of salinity were available, model (2) also included salinity S_ij (salinity), whereas salinity was not available and hence not included in the TABACOD samples.

Element concentrations were log+1 transformed to meet assumptions of homogeneity and normally distributed model residuals. Collinearity among the fixed effects may result in variance inflation (VIF). We tested the degree of VIF using “vif” test of the “car” package (Fox and Weisberg, 2019). Models with VIF > 5 are considered to suffer from collinearity among predictor variables (Zuur et al., 2010). The highest VIF values were observed for month (4.42), season (2.69) and depth (2.23) as spawning usually occurs within specific months in the deepest parts of the Bornholm Basin. However, all VIF values were< 5 and therefore all fixed effects were included in the initial global model [model (2)]. Prior to model testing and selection, the predictor and response variables were centred and scaled using the “scale” function in R in order to standardise the model (Schielzeth, 2010). Finally, the most parsimonious model was identified by ranking all possible models by the Akaike information criterion (AIC) using the “dredge” function of the MuMIn package (Burnham and Anderson, 2002). The final model was selected as the model with an AIC difference > 2 compared to all other models (Zuur et al., 2009). Model statistics, including partial p values and conditional and marginal correlations, were extracted using the “lmerTest” package (Kuznetsova et al., 2017).

For samples where salinity was directly measured by the DST attached to the fish (CODYSSEY samples), global model (2) was fitted including salinity as a possible predictor, however it was never selected for the final model. Even when model (2) was fitted excluding potentially covarying variables area and depth (see hypotheses) there was still no relationship between salinity and otolith Sr and Ba as we had expected (H1; Table 1 , Supplementary Figure 4C ). As such, the results presented herein are based on the combined data from the two programmes (hence without salinity as a predictor).

Despite the apparent absence of a salinity effect, otolith Sr concentrations were significantly higher and Ba and Mn concentrations lower in the western than in the eastern Baltic Sea, as predicted (H2; Figure 5 , Table 1 ). Contrary to expectation, otolith Sr concentrations decreased with depth, while otolith Ba concentrations, as expected, were highest in shallow water (H3). Overall, however, Sr and Ba were positively correlated (ANOVA, df = 2006, r² = 0.15, p< 0.001) unlike observations throughout much of the literature. Otolith Mn concentrations were highest in shallow water during the late summer – winter (August – February) feeding season. (H4) All three elements varied seasonally with the highest Sr concentrations in August to September and lowest concentrations in November to January, coinciding with general spawning- and feeding seasons respectively. Seasonal signals were even more pronounced in Mn, but out of phase with the Sr, with a peak in fall/winter and a sharp drop in spring. Ba concentrations followed the same seasonal pattern as Mn, but somewhat less pronounced. Contrary to hypothesis H5 we detected size-related patterns in otolith Sr, Ba and Mn concentrations, with all elements negatively related to fish size at recapture. Mean fish growth was not included in the final model. There was a negative relationship between otolith opacity and both Ba and Mn, and a positive relationship with Sr. The only element that differed between the sexes in this group of elements was Ba, with higher otolith concentrations in males than females ( Table 1 ). For all three elements reported in this section, the fixed effects explained a considerably lower proportion of the variability in otolith concentrations than the random effects of tagging year and fish (Sr: marginal r² = 19.5%, conditional r² = 58.5%; Ba: marginal r² = 13.2%, conditional r² = 73.4%; Mn: marginal r² = 10.0%, conditional r² = 58.9%).

P, Mg and Zn all exhibited similar relationships with environmental and biological variables ( Table 1 , Figure 6 ) in the final Equation 2. All elements showed a distinct seasonal pattern with lowest concentration in late spring (April – May) and highest concentrations in winter (November – February), as predicted (H6), except that the peak concentrations were predicted to occur in fall/early winter when consumption (and presumably also growth) is highest. P and Mg were also positively correlated with temperature (H7), while there was no temperature effect on Zn concentrations. We also predicted higher element concentrations during the feeding season compared to the spawning season (H9), but this was only evident for P, while the opposite was observed for Zn (highest Zn during spawning), and Mg showed no difference between seasons. Also contrary to our predictions was that all three elements were not related to mean fish somatic growth across the tagging period (H8). They were also negatively related to otolith opacity. Also not consistent with our predictions were the observations of increased Mg with depth, and higher concentrations in P and Mg in the eastern than the western Baltic Sea (H10). Similar to the previous section, the fixed effects explained a considerably lower proportion of the observed variability than the random effects of tagging year and fish (P: marginal r² = 23.2%, conditional r² = 79.4%; Mg: marginal r² = 6.3%, conditional r^{2 =} 72.0%; Zn: marginal r² = 5.9%, conditional r² = 60.7%).

Sr, Ba and Mn all showed expected differences among geographic areas (H2) and months of the year (H4) and expected lack of relationships with fish age, somatic growth rate, and ambient temperature (H5). However, some elements also showed unexpected relationships with depth (H3), salinity (H1), sex and fish length (H5). Each of the elements will therefore be discussed separately.

P, Mg and Zn are important co-factors in many physiological processes. Spatiotemporal differences occur in the environmental concentrations of these elements, because they are taken up by primary producers during annual phytoplankton blooms (Elsdon and Gillanders, 2005; Walther and Limburg, 2012). Here, otolith P, Mg and Zn concentrations showed remarkably similar seasonal trends, with the lowest concentrations in April and May and – as predicted – the highest concentrations in fall and winter (H6). Also as predicted, Mg and P were positively related to temperature (H7) but were unexpectedly related to geographic area and/or depth (H10). Also unexpectedly, none of these elements were influenced by fish size or overall growth rate (H8), or their effects were already accounted for by factors such as month and season.

As otolith opacity relates primarily to the relative concentration of otolith matrix compounds in the calcium carbonate lattice, and is driven by daily, seasonal, and ontogenetic patterns in otolith biomineralization, we discuss it separately here. Otolith matrix protein concentrations exhibit a negative relationship with fish size (Baba et al., 1991; Sasagawa and Mugiya, 1996; Hüssy et al., 2004) and a positive relationship with temperature (Hüssy et al., 2004; Fablet et al., 2011), due to reduced matrix synthesis, increased calcium carbonate accretion, or an interaction between the two. Otolith opacity increases with increasing concentrations of organic matrix compounds (Hüssy et al., 2004; Pilling et al., 2007). Thus, one would expect the concentration of matrix-bound elements to be positively related to otolith opacity. However, P, Mg and Zn were all negatively correlated with opacity. Elements that primarily substitute for Ca in the crystal lattice, on the other hand, would be expected to be unrelated to otolith opacity. Yet Sr and Mn were positively related, and Ba negatively related to otolith opacity. These contradictory results highlight our still limited understanding of the organic matrix chemistry and how these compounds interact with endolymph ions during biomineralization (Thomas et al., 2019).

The key lessons learned from this study were (1) the value of performing controlled experimental studies to disentangle variables that are intrinsically collinear in nature, and (2) the value of measuring element concentrations in freshwater endmembers entering a brackish or marine system. While fully marine environments typically lack large physicochemical gradients that can impede our ability to reconstruct individual migrations using otolith chemistry (Sturrock et al., 2012), the Baltic Sea contains perhaps too many gradients. The heterogeneous brackish hydrography in the Baltic Sea coupled with the cod’s complex migration patterns covaried temporally, horizontally, and vertically, making it impossible to tease apart physiological signals from environmental signals relating to changes in water temperature and chemistry as the fish moved among habitats. Thus, contrary to our initial expectation, tagged Baltic cod were not the ideal case study for testing the validity of hypotheses on otolith biomineralization and element uptake. To improve our understanding of the dynamics in this environment, we recommend the combined sampling of water, fish plasma and otoliths (e.g. Sturrock et al., 2015) in order to establish a comprehensive library of information for calibrating measured otolith concentrations. Improved understanding of biomineralization mechanisms have been highlighted as essential research need in this field for many decades (Campana, 1999; Hüssy et al., 2020b). Our results reinforce this perception, particularly for understanding patterns in P, Mg, Mn, and Zn that are co-factors for proteins involved in fish metabolism and otolith growth and opacity.

Another caveat of this study was the potential for otolith time (element concentrations) to be decoupled from calendar time (environmental measurements) given assumptions of constant growth over the tagging period. Other studies have used fine-scale otolith δ¹⁸O measurements to add a temperature-driven otolith timeline (Sturrock et al., 2015), but this was impossible here given considerable salinity variations and lack of a detailed three-dimensional water δ¹⁸O isoscape. Nor could we use the periodicity of growth bands to infer intra-annual growth rates as they are famously indistinct in Baltic cod otoliths (Hüssy, 2010; Hüssy et al., 2016b). Future work should focus on expanding the research using otolith Mg and P (Heimbrand et al., 2020; Hüssy et al., 2021a; Reis-Santos et al., 2022) to link their seasonal concentration patterns to calendar month.

A persistent pattern observed in this study was that the fixed effects explained a relatively low fraction of the variation in otolith element concentrations (6 – 23%), with the fish and year level random effects explaining – on average – more than 5.5 times as much (40 – 60%). This observation is far from unusual. In the most comprehensive study of its kind, using fish reared under identical mesocosm conditions and including water and blood plasma element concentrations and El/Ca ratios as possible explanatory variables, individual variation still dominated the signal (fixed vs. random effects explaining 16 – 50% vs. 16 – 59% of the variation otolith elemental concentrations; Sturrock et al., 2015). Together, these studies highlight the need to understand the factors driving individual differences in order to effectively partition variance in otolith elemental chemistry before it can be reliably used for environmental and movement reconstructions in a heterogeneous environment.

The focus of future studies should include targeted laboratory and field studies to better understand elemental uptake and transport from water, food, and existing body pools to the otolith (Hüssy et al., 2020b). While challenging to do, it is important to carry out these studies across life stages and growth and maturation cycles in order to quantify the impact of physiology, and to always include sex as a covariate, in case sexual dimorphism or factors associated with vitellogenesis prove to be important.