ABSTRACT
Although geographic patterns of species’ sensitivity to global environmental changes are defined by interacting multiple stressors, little is known about the biological mechanisms shaping regional differences in organismal vulnerability. Here, we examine large-scale spatial variations in biomineralisation under heterogeneous environmental gradients of temperature, salinity and food availability across a 30° latitudinal range (3,334 km), to test whether plasticity in calcareous shell production and composition, from juveniles to large adults, mediates geographic patterns of resilience to climate change in critical foundation species, the mussels Mytilus edulis and M. trossulus. We find mussels produced thinner shells with a higher organic content in polar than temperature regions, indicating decreasing shell calcification towards high latitudes. Salinity was the major driver of regional differences in mussel shell deposition, and in shell mineral and organic composition. In low-salinity environments, the production of calcite and organic shell layers was increased, providing higher resistance against dissolution in more corrosive waters. Conversely, under higher-salinity regimes, increased aragonite deposition suggests enhanced mechanical protection from predators. Interacting strong effects of decreasing salinity and increasing food availability on the compositional shell plasticity in polar and subpolar mussels during growth predict the deposition of a thicker external organic layer (periostracum) under forecasted future environmental conditions. This marked response potential of Mytilus species suggests a capacity for increased protection of high-latitude mussel populations from ocean acidification. Our work illustrates that mechanisms driving plastic responses to the spatial structure of multiple stressors can define geographic patterns of unforeseen species resilience to global environmental change.
INTRODUCTION
Unprecedented global environmental changes are driving scientists towards increased understanding of the mechanisms underlying geographic variation in species’ responses to future environmental conditions (1, 2). However, our ability to forecast emergent ecological consequences of climate change on marine populations, communities and ecosystems remains limited (3). Ecosystem-wide projections are severely constrained by heterogeneous patterns of ocean warming and acidification (4), multiple interacting stressors (5), and species-specific effects (6), as well as predictive models which often exclude important biological mechanisms when projecting changes to species and ecosystems in response to climate change (2). A better mechanistic understanding of the biological processes and environmental sources mediating species’ responses to disturbances is critical for building the theoretical baseline necessary to forecast the combined effects of multiple emerging stressors (2, 3).
Advances in macroecology suggest that permanent environmental mosaics, defined by spatial overlaps of non-monotonic environmental gradients (7), as well as regional adaption or acclimatization (8–10), dictate geographic variations in species performance and sensitivity to environmental change in marine ecosystems. Key to these works is that responses vary among populations and individual taxa (6, 8), which often play disproportionately strong roles in structuring benthic communities (11). Thus, species-specific biological mechanisms driving organismal variability may shape differential regional responses of foundation species to co-occurring multiple drivers. This can establish spatial patterns of unexpected susceptibility of marine communities to future conditions.
Climate change is considered a major threat to marine ecosystems worldwide, with ocean warming and acidification profoundly affecting species life history and ecology (6, 10), as well as community structure and ecosystem dynamics (11, 12). Species producing calcium carbonate (CaCO3) shells and skeletons are possibly experiencing the strongest impacts of rapid environmental changes (6). Knowledge on their sensitivity is derived largely from experimentally induced responses in model organisms (1, 6), while complex variations under multiple stressors have rarely been investigated in natural environments (7, 11–13). Therefore, inferences made from experimental studies can be misleading and not fully applicable to marine ecosystems (9). Indeed, species-specific mechanistic responses to habitat alterations (14) on top of mixed outcomes of environmental interactions (additive, synergistic or antagonistic) make future ecosystem predictions extremely challenging. This leaves open the question: do differences in biological mechanisms, shaping regional calcifiers’ responses to interacting environmental stressors, define geographic patterns of unforeseen species sensitivity or resilience to global environmental change? A body of research has focused on responses of marine calcifiers to altered water chemistry (1, 6), but studies have rarely considered changes in biogeochemical cycles that strongly mediate biological responses to environmental alterations (4). Among those, a marked intensification of the global water cycle in response to warming (+4% for +0.5°C) has been documented over recent decades through changes in ocean salinity (15). Salinity is a major ecological factor dictating survival of aquatic organisms and ecosystem functioning. Multidecadal studies have revealed a global salinity pattern following the “rich-get-richer” mechanism, where salty ocean regions (compared to the global mean) are getting saltier (mid-latitudes), whereas low salinity regions are getting fresher (tropical convergence zones and polar regions) (15). In a future 2-3°C warmer world (16), a substantial 16-24% intensification of the global water cycle is predicted to occur making salinity gradients much sharper (15). However, emergent ecological effects of changing salinity on calcifying species and marine communities are largely unknown.
Atlantic mussels, Mytilus edulis and M. trossulus, are important bed-forming foundation species throughout the eulittoral ecosystems of the northern hemisphere (up to 90% of epibenthic biomass), and represent valuable resources for aquaculture (192,000 t produced in 2015 worth 325 million USD) (17). Growing awareness of the consequences of climate change on biodiversity and industry that Mytilus species support have stimulated a number of studies to estimate the response potential of these habitat-forming calcifiers to changing ocean conditions (18–20).
Calcareous shells perform a range of vital functions including structural support and protection against predators. Because shell integrity determines survival, shell traits are subject to strong selection pressure with functional success or failure a fundamental evolutionary driver. Mytilus shell consists of three layers (Fig. 1A-B): (1) the outer organic periostracum, (2) the calcified prismatic and (3) nacreous layers. The periostracum provides the substrate and a protected environment for shell secretion, and is made of sclerotized proteins, protecting shells from corrosive, acidic waters as well as predatory and endolithic borers (21). The prismatic and nacreous layers are composed of different mineral forms of CaCO3, calcite and aragonite respectively, dispersed in an organic matrix (22). These calcareous layers are characterized by different microstructures and more (e.g. aragonite) or less (e.g. calcite and organics) soluble components the combination of which determines chemical and mechanical shell properties (23). Differences in energetic costs of making shell components (13) combined with future shifts in environmental gradients (4) may influence variations in shell production, composition and structure, shaping regional patterns of shell strength and resistance to acidification.
Mytilus growth, biomineralisation and fitness are linked to multiple drivers, including water temperature, salinity and food supply [chlorophyll-a (Chl-a) concentration] (24, 25). In the North Atlantic and Arctic Oceans, these key environmental factors vary heterogeneously with latitude (Fig. 1C-D), encompassing a range of conditions predicted under different future climate change scenarios (16). Here we hypothesize that biological mechanisms driving spatial variations in shell production, mineral and organic composition: i) shape regional differences in the responses of Mytilus species to interacting environmental drivers, and ii) define geographic patterns of unanticipated mussel vulnerability in the face of global environmental changes.
Despite projected environmental alterations (4, 15), salinity gradients have been overlooked in large-scale models predicting emergent effects of climate changes on marine organisms. This knowledge is essential to predict whether environmental changes affect shell variability (i.e. thickness, mineral and organic content) and its properties, especially in calcifying foundation species such as M. edulis and M. trossulus. These factors are crucial for understanding species susceptibility to other rapidly emergent stressors, such as warming and acidification (3).
In this study, we examine the relationships between the plasticity in Mytilus shell production and composition (from juveniles to large adults) and interactive environmental gradients of temperature, salinity and Chl-a concentration in 17 populations spanning a latitudinal range of 30° (3,334 km) across the Atlantic-European and Arctic coastline (Fig. 1C-D). In particular, we test for a latitudinal effect on Mytilus shell calcification (variation in shell thickness and organic content) that we hypothesize will show a general decrease from temperate to polar regions. We also identified environmental sources and magnitude of regional variations in shell deposition, to test whether salinity affects shell production and mineral composition during growth, driving changes of mechanical and chemical shell properties. Finally, we modelled spatial trends in the production of individual shell layers with environmental gradients, to test whether biological mechanisms, driving variations in shell structure and properties, shape regional responses of Mytilus to interacting stressors (especially salinity) and define geographic patterns of sensitivity to future changes.
RESULTS
Generalized linear (mixed) models, GL(M)Ms, were used to explain shell thickness and composition, from juveniles to large adults, with respect to latitude and environmental drivers, and to compare between the individual shell layers.
Latitudinal patterns of shell deposition
GLMMs indicated a general decrease of Mytilus whole-shell thickness with increasing latitude from warm-temperate to polar regions (Fig. 2A). We detected a significant negative relationship between the prismatic layer thickness and latitude (Fig. 2A), while no variation in nacreous thickness, periostracum thickness and relative proportion of prismatic layer thickness (calcite%) was found (table S1). Shell length was positively correlated with thickness in all layers indicating thickening during growth (table S1).
The weight proportion (wt%) of organic content in the prismatic layer was modelled with a GLM as a function of collection site and shell thickness. Prismatic layers were characterized by a significantly higher organic content (lower proportion of CaCO3) in mussel shells from polar than temperate regions, indicating decreased shell calcification at higher latitudes (Fig. 2B). Polar shells [sites 15, 16; mean (SD) = 1.8 wt% (0.31)] were characterized by an average of 29% more organic content compared to temperate mussels [sites 1, 11; mean (SD) = 1.4 wt% (0.16)]. The organics wt% was negatively correlated with prismatic thickness (Fig. 2C), indicating a lower proportion of CaCO3 and thinner, so less calcified, shells at polar latitudes.
Environmental influence on shell production and composition
Individual GLMMs were fitted to explain spatial variations in the whole-shell thickness, periostracum thickness and calcite% with environmental gradients during shell growth. We identified significant trends in shell thickness with environmental gradients depending on the shell measurement considered (Fig. 3A-E, fig. S1, Table 1). Whole-shell thickness was positively related to temperature, salinity and shell length, but there was no influence of Chl-a (cR2 = 0.93; Fig. 3A). Salinity had an effect on shell thickness that was 3.4 and 2.1 times larger than temperature and length, respectively (Fig. 3A, Table 1). We detected a negative relationship between calcite% and salinity (95% CI = -12.03 to -2.38, cR2 = 0.56) (Fig. 3B, table S2), with none of the other drivers having a significant effect.
Prismatic and nacreous layers thickness were analysed within the same GLMM. After model selection, fixed continuous covariates of the optimal model, equation (1), were the standardised temperature, salinity, Chl-a concentration and shell length in addition to shell layer (categorical, two levels: prismatic and nacreous) and their interactions. The random component was the collection site used as a random intercept. The model was of the form: where Thicknessijk is the kth thickness observation from layer j (j = prismatic, nacreous) and site i (i = 1,…, 17). Siteij is the random intercept for layer j, which is assumed to be normally distributed with mean 0 and variance
Sea surface temperature, salinity and shell length all successfully predicted (cR2 = 0.93) variations in the thickness of prismatic and nacreous layers, while no influence of Chl-a was detected (Table 1). The mean effect size of salinity on the response was twice as large as the effect of shell length, while it was 2.9 and 4.7 times larger than the effect of temperature on the prismatic and nacreous layers, respectively (equation (2), Fig. 3C-D). This indicates salinity had a stronger contribution to predicting shell structure than the effects of temperature, Chl-a and shell length combined (Fig. 4).
Interactions between shell layer and both salinity and shell length (equation (2)) indicate deposition of proportionally thicker prismatic layers under low salinities and proportionally thicker nacreous layers under higher salinities across the entire range of shell lengths (Fig. 4).
Periostracum plasticity
Models of periostracum thickness revealed significant exponential relationships with Chl-a and shell length (cR2 = 0.81) (Table 1). Length had a mean effect that was 3 times larger than Chl-a (Fig. 3E), showing a rapid thickening of the periostracum during shell growth. The interactions between shell length and both salinity and temperature indicate that the effects of these variables on periostracum were interdependent. At low salinities, the higher values of shell length had a greater positive effect on periostracum thickness, while the reverse was true for higher temperatures having a marginal effect only on thickening rates (Fig. 5A-B). This suggests that increasing shell size was a more important factor for periostracum growth in fresher waters than in relatively saltier conditions.
Among-site shell variation
GLMMs showed no difference in collection site-level effects (conditional modes) on each thickness measurement (Fig. 5C). This indicated no residual effect of species identity or hybridization on the thickness of individual shell layers at different sites after accounting for the effects of environmental factors and shell length.
DISCUSSION
Our results demonstrate that plasticity in shell production in Mytilus species and the spatial structure of environmental conditions drive geographic variations in shell responses shaping regional differences in the resilience of these foundation species to global environmental change. An understanding of the biological mechanisms driving regional species’ responses to multiple interacting stressors is crucial for improving predictive accuracy and informing more realistic projections of species and ecosystem resilience to climate change (2). Heterogeneous population-level responses from different climates suggest that environmental stressors, especially salinity, drive regional variations in Mytilus shell production, mineral (prismatic and nacreous) and organic (periostracum) composition during growth, which is reflected in the relative proportion of each shell layer. Variations in shell production and composition determine geographic differences in chemical and mechanical protection of shells, shaping the vulnerability of these habitat-forming species to future conditions.
Decreasing shell calcification (increasing organic content and thinner shells) towards high latitudes (Fig. 2) supports documented patterns of skeletal production (13, 26). Two explanatory paradigms exist for decreased skeletal size at higher latitudes: increased calcification costs (13) and reduced predation pressure (27). Given the higher production cost of organics than CaCO3 deposition (13) and problematic protein production at polar temperatures (28), we might expect a reduced proportion of organic matrix. Moreover, decreasing predation pressure (27) should result in thinner shells of the same composition irrespective of geographic area. However, the wt% of organic matrix was higher at Arctic latitudes. This could suggest either (or a combination of) a marked increase in the cost of calcification in polar regions (13), altering significantly the relative costs of CaCO3 and organics production, or a decreased saturation state (increased dissolution) of CaCO3 due to low temperatures and, more importantly, salinity (low [Ca2+] availability) (25). In either case, these underlying effects would result in decreased shell calcification at high latitudes. This increased proportion of organic matrix could protect the calcified shell components from dissolution and have an adaptive beneficial effect in more corrosive conditions.
Our results illustrate that different drivers significantly affect both shell thickness and composition in Mytilus (Fig. 3). For over 60 years, temperature and shell size have been considered key drivers of CaCO3 shell mineralogy across latitudes, dictating the formation of predominantly aragonitic structures in temperate regions and increased calcite precipitation in cold climates (29–31). Although our study partly supports previous findings, we demonstrate that salinity has the strongest influence on shell production and composition in Mytilus, which is contrary to the general assumption of temperature and shell size being the primary drivers of shell compositional plasticity.
The interaction between shell layer, salinity and shell size indicates heterogeneous, age-related compositional changes in Mytilus shells across different salinities (Fig. 4A). Shifts in shell properties from juveniles to large adults are strongly modulated by salinity, which leads to the formation of exclusively prismatic-dominated shells in brackish waters and nacreous-dominated structures under marine conditions (Fig. 4B). These patterns, which we show were independent of species or hybrid status (Fig. 5C), indicate that mussel shell plasticity during growth (the Length × Layer interaction, equation (1)) has an indirect effect on layer thickness by allowing salinity-induced compositional changes and, therefore, the production of the most appropriate shell structure for specific environmental conditions.
Under current scenarios, plasticity in shell production could confer Mytilus species an advantage when facing different water chemistries and predation levels. In fact, at high-latitudes and in the Baltic region, where durophagous (shell-breaking) predators are rare or absent and the water is more corrosive (13, 27), mussels are characterized by thinner, prismatic-dominated shells, providing a generally higher protection from dissolution. Conversely, at mid-latitudes, where durophagous predators are more abundant and the CaCO3 solubility of the water is lower (13), mussels display thicker, nacreous-dominated shells with higher mechanical resistance.
Despite rapid global changes in the water cycle and salinity gradients (15), Mytilus species shows a strong capacity to respond to heterogeneous environments. This plasticity in shell production could help to mitigate the emergent negative effects of changing water chemistry. In fact, the interacting effects of salinity and shell length, as well as a minor influence of temperature, on the periostracum (Fig. 5A-B), which represents a strong chemical barrier to dissolution in molluscs (21, 32, 33), suggest enhanced periostracal thickness under decreasing salinities could mediate impacts of ocean acidification.
Although populations in high-latitude ecosystems will experience globally the most rapid acidification (4), the concurrent decrease in salinity predicts thicker prismatic layers and periostraca will be produced which increase protection from higher solubility conditions. Conversely, in temperate areas, increasing salinity would determine deposition of thicker shells and a relatively thicker nacreous layer and thinner periostracum, favouring mechanical shell resistance. However, predicted changes in periostracal thickening rate under different salinities depend on shell size and would be more evident in larger individuals (length > 48 mm) (Fig. 5A).
In Greenland, where the rate of melting of the ice sheet has doubled in the last decade (34), low salinities during summer (< 20 psu) and high productivity (food supply) in coastal areas and fjords (35) predict formation of thicker periostraca and increased relative thickness of organic-enriched (high wt%) calcitic layers. These changes could make Arctic Mytilus populations more resilient to future acidification. Differently, in the Baltic Sea, the forecasted decrease in salinity (maximum 45% reduction) (36), combined with a considerable physiological stress, would be particularly critical for mussels inhabiting already unfavourable conditions for calcification (salinity from 22 psu to 3 psu, low water [Ca2+], and CaCO3 saturation state) (25). Moreover, the reduced shell size of Baltic Mytilus does not predict formation of thicker periostraca, which would further increase vulnerability to dissolution. Impacts of changing salinity on this habitat-forming species, which contributes up to 90% of the Baltic benthic biomass, could strongly affect the ecosystem, most likely resulting in substantial range restrictions towards higher salinity areas.
Although our results strongly support the hypothesis that biological mechanisms for variations in shell production can shape regional responses in Mytilus, changes of other biological drivers, such as predation pressure and primary production, could also have profound influences (13, 37). In fact, as temperature rises, durophagous predators could expand their ranges towards polar regions (38), suggesting an increased vulnerability of thin-shelled individuals. However, predicted northward phytoplankton expansions and an overall increase in primary production at high latitudes (37), could favour periostracal growth potential in Mytilus and, thus, increased resistance to dissolution for all the size classes in polar and subpolar regions.
Mytilus shells have a thick periostracum and a marked compositional plasticity compared to other calcifiers that often compete with it for space (e.g. barnacles and spirorbid polychaetes). This layer provides a strong defence against shell dissolution allowing mytilids to survive in oligohaline waters (∼5 psu) and extremely acidified conditions (e.g. hydrothermal vents) (33). These factors may shift the ecological balance and community structure in favour of species with stronger resistance to corrosive conditions, such as mussels, when ocean waters become fresher and more acidic in future decades.
As hypothesised, plasticity in shell production and the spatial structure of environmental conditions drive regional differences in Mytilus shell deposition and composition, shaping spatial patterns of chemical and mechanical shell properties. Overall, mussel shell calcification decreased towards high latitudes, with salinity being the major driver of geographical variations in shell production, mineral and organic composition. The marked compositional plasticity in calcareous shell components (prismatics and nacreous layers) suggest an higher resistance against dissolution for mussels in polar, low-salinity environments, and an enhanced mechanical shell protection from predators in temperate, higher-salinity regions. The strong response potential of Mytilus shell periostracum to heterogeneous environments predicts an increased resilience to ocean acidification in polar and sub-polar mussels, and a higher sensitivity of Baltic populations under future environmental conditions.
In conclusion, our findings demonstrate that biological mechanisms, driving spatial variability of mussel responses to interacting environmental factors, shape the complex geographic pattern of shell deposition and properties, dictating regional differences in Mytilus species sensitivity to future environmental change. As the magnitude of anthropogenic impacts continue to increase, further studies are need to better understand the key biological processes mediating species’ response to habitat alterations, especially for those having both high climate sensitivity and disproportionately strong ecological impacts in shaping marine communities. This knowledge underpins our ability to predict accurately and reduce the damaging effect of climate change on future biodiversity under any range of scenarios (2). Our study has important implications because it clarifies the links between (i) the mechanisms of biological variation, (ii) the predicted shift in spatial co-occurrence of multiple environmental drivers, and (iii) regional differences in the plastic responses and sensitivity of calcifying, foundation species to changing habitats. This understanding is of critical importance for making realistic projections of emergent ecological effects of global environmental changes, such as altered salinity regimes, and to improve our predictive accuracy for impacts on marine communities and ecosystems, and the services they provide.
MATERIALS AND METHODS
Mytilus collection
We sampled individuals from a total of 17 Mytilus (Mytilus edulis and M. trossulus) populations along the North Atlantic, Arctic and Baltic Sea coastlines from four distinctive climatic regions (warm-temperate, cold-temperate, subpolar and polar) covering a latitudinal range of 30° (a distance of 3,334 km), from Western European (Brest, North-West France, 48°N) to Northern Greenlandic (Qaanaaq, North-West Greenland 78°N) coastlines (Fig. 1C). During December 2014 - September 2015, mussels of various size classes for each site (shell length 26-81 mm) were sampled from the eulittoral zone on rocky shores for a total of 424 individuals (table S3). For each specimen, shell length was measured with digital calipers (0.01 mm precision) and used as a within-population proxy for age.
We analysed Mytilus populations of which the genetic structure was known, with particular focus on species identity and hybrid status (M. edulis × M. trossulus). Mytilus shells used were either from specimens already evaluated in genetic investigations or mussels obtained from sites routinely used in regional monitoring programs that provided information on species identity (table S3). Areas where the Mediterranean mussel, Mytilus galloprovincialis, was present were avoided. We did, however, sample few sites with very low levels of M. edulis × M. galloprovincialis hybridization.
Mussel shell preparation
We set left shell valves in polyester resin (Kleer-Set FF, MetPrep, Coventry, U.K.) blocks. Embedded specimens were sliced longitudinally along their axis of maximum growth (Fig. 1A) using a diamond saw and then progressively polished with silicon carbide paper (grit size: P800-P2500) and diamond paste (grading: 9-1 µm). Photographs of polished sections (Fig. 1B) were acquired with a stereo-microscope (Leica M165 C equipped with a DFC295 HD camera, Leica, Wetzlar, Germany) and shell thickness (µm) was measured using the Fiji software (v1.51u). Since larger individuals had undergone evident environmental abrasion or dissolution which removed the periostracum and prismatic layer closer to the umbo, we estimated the thickness of the whole-shell, prismatic and nacreous layers at the midpoint along the shell cross-section. The proportion of calcite was estimated as
Periostracum thickness was measured at the posterior edge where it attaches to the external side of the prismatic layer, to estimate the fully formed organic layer that was unaffected by decay or abrasion (21).
Organic content analyses
We performed thermogravimetric analyses (TGA) to estimate the weight proportion (wt%) of organic matrix within the prismatic layer. Mytilus edulis specimens were selected from four populations (sites 1, 11, 15, 16) to explore differences in shell organic content under temperate and polar regimes. We removed the periostracum by sanding, and prismatic layer tiles (8 × 5 mm, N = 20 × 4 sites) were isolated along the posteroventral shell margin. Tiles were cleaned, air-dried and then finely ground. We tested ten milligrams of this powdered shell with a thermogravimetric analyser (TGA Q500, TA Instruments, New Castle, DE, U.S.A.). Samples were subjected to constant heating from ∼25°C to 700°C at a linear rate of 10°C min-1 under a dynamic nitrogen atmosphere and weight changes were recorded (Supplementary Material and Methods). We estimated the wt% of organic matter within the shell microstructure as the proportion of weight loss during the thermal treatment between 150°C and 550°C (fig. S2).
Environmental characterization
We selected three key environmental drivers based on their known influence on mussel growth, their level of collinearity across the geographic scale investigated and the forecasted major ocean alterations under climate change (16). For each site, measurements of sea surface temperature, salinity and Chl-a concentration, the latter being used as a proxy for food supply (24), were generated using the Copernicus Marine Environment Monitoring Service (CMEMS) (http://marine.copernicus.eu/). These climate datasets are composed of high-resolution physical and biogeochemical assimilated (integration of observational and predicted information) daily data (N = 2,191 per parameter) (see data file S1). To provide a first order approximation of the water conditions prevailing during the near-maximum rates of shell deposition (30), we expressed parameters as mean May-October values averaged over the 6-year period 2009-2014 and used these as input variables (Fig. 1D, table S4).
Direct environmental monitoring for each site was not feasible due to the number of populations analysed, their geographic range (> 3,300 km) and the temporal resolution (6 years) required to estimate the average growth conditions during the lifespan of sampled specimens. For this large-scale study, remote-sensing and assimilated data presented potential advantages compared to traditional measurements due to their high spatio-temporal resolution, advanced calibration and validation (i.e. high correlation with discrete field measurements).
Statistical analysis
GLMMs were applied to account for the hierarchical structure of the dataset consisting of multiple specimens (N = 24-26 replicates) from each collection site and to generalize our results to Mytilus populations beyond the study sample.
We carried out data exploration following the protocol of Zuur et al. (39). Initial inspection revealed no outliers. Pairwise scatterplots and variance inflation factors (VIFs) were calculated to check for collinearity between input variables. VIF values < 2 indicated an acceptable degree of correlation among covariates to be included within the same model. We applied residual regression to uncouple the unique from the shared contribution of temperature and Chl-a concentration to the response (40). This allowed us to account for the existing causal link between these two parameters and to avoid inferential problems from modelling non-independent covariates without losing explanatory power (40). To directly compare model estimates from predictors on different measurement scales, estimate biologically meaningful intercepts and interpret main effects when interactions are present, we standardised all the input variables (environmental parameters and shell length). For standardisation, we subtracted the sample mean from the variable values and divided them by the sample standard deviation . Preliminary inspection of residual patterns showed heteroscedasticity in most models. The use of different continuous probability distributions (i.e. gamma) and link functions did not stabilize the variance, therefore a ln-transformation of the response was required, except for calcite% and wt% measurements. Response variables did not require further transformations.
We used separate GLMMs to explore patterns of shell thickness of individual layers with latitude and shell length (size). The proportion (wt%) of organic matrix (N = 80) was modelled as a function of site (categorical, four levels) and prismatic thickness (continuous) to test for differences between polar and temperate regions and association with shell thickness. The response variable was coded as a value from 0 to 1; therefore, we used a GLM with a beta distribution and a logistic link function. Pair-wise contrasts with a Bonferroni correction were then used to test for differences in wt% among sites within and between climatic regions.
Different approaches were used to investigate the relationships between shell thickness and environmental gradients. Whole-shell thickness, periostracum thickness and calcite% were modelled separately (N = 424 each). Prismatic and nacreous layer thickness were analysed within the same GLMM, allowing the simultaneous prediction of common and divergent environmental effects on both layers and to reduce the probability of type I error. To model shell thickness (N = 424 × 2 layers) as a function of the environmental predictors we used a GLMM with a normal distribution (equation (1)). In the initial model, fixed continuous covariates were the standardised temperature, salinity and Chl-a in addition to shell layer (categorical, two levels) and their two-way interactions. Shell length (continuous) was included to control for possible effects of within population age on layer thickness. To incorporate the dependency among observations for a specific layer from the same collection site, we used site as a random intercept.
Models were optimized by first selecting the random structure and then the optimal fixed component. The principal tools for model comparison were the corrected Akaike Information Criterion (AICc) and bootstrapped likelihood ratio tests. Random terms were selected on prior knowledge of the dependency structure of the dataset. Visual inspection of residual patterns indicated violation of homogeneity in most cases. This required the use of variance structures (generalized least squares) allowing the residual spread to vary with respect to shell layer. The fixed component was optimized by rejecting only non-significant interaction terms that minimized the AICc value. For all model comparisons, variation of AICc between the optimal (lowest AICc value) and competing models were greater than 8, and fixed-effect estimates were nearly identical, indicating that competing models were very unlikely to be superior (41). The proportion of variance explained by the models was quantified with conditional or pseudo determination coefficients (cR2 or pseudoR2). We used variograms to assess the absence of spatial autocorrelation. Final models were validated by inspection of standardised residual patterns to verify GLMM assumptions of normality, homogeneity and independence. We used optimal models to estimate the mean effect sizes (same measurement scale) of environmental drivers on the response. Ninety-five per cent confidence intervals (95% CI) for the regression parameters were generated using bias-corrected parametric bootstrap methods (10,000 iterations). 95% CIs were used for statistical inference due to estimation of approximated significance values (P-value) in mixed-modelling. If the confidence intervals did not overlap zero, then the effect was considered significant. All data exploration and statistical analyses were performed in R (v3.4.1) (for packages see table S5).
Funding
The work was funded by the European Union Seventh Framework Programme, Marie Curie ITN CAlcium in a Changing Environment (CACHE), under grant agreement n° 605051. JT acknowledges additional financial support from the Danish Council for Independent Research, Individual Post-doctoral Grant n° 7027-0060B.
Author contributions
L.T., L.S.P. and E.M.H. conceived the original project and designed the study; L.T. performed laboratory work and thermogravimetric analysis, generated environmental datasets, performed modelling work and analysed output data; L.T., T.S., J.T. and M.K.S. collected materials; L.T., L.S.P. and E.M.H. wrote the first draft of the manuscript, and all co-authors contributed substantially to revisions.
SUPPLEMENTARY DATA
Environmental datasets
List of the datasets used for the calculation of mean annual values of environmental descriptors. Water temperature, salinity and chlorophyll-a concentrations were expressed as mean values (May - October) averaged over the 6-year period 2009 - 2014. This study has been conducted using the Copernicus Marine Service Products: COPERNICUS - Marine Environment Monitoring System (http://marine.copernicus.eu/).
DATASET #1
DATASET #2
DATASET #3
DATASET #3
DATASET #4
DATASET #5
DATASET #6
Acknowledgments
We thank Iain Johnston (Scottish Oceans Institute, St. Andrews, UK), Sarah Dashfield (Plymouth Marine Laboratory, Plymouth, UK), Dr. Peter Thor (Norwegian Polar Institute, Tromsø, Norway), Alexander Ventura (University of Gothenburg, Kristineberg, Sweden), Dr. Henk van der Veer and Rob Dekker (Royal Netherlands Institute for Sea Research, Texel, Netherlands) for help with specimens collection, and the Statistics Clinic (University of Cambridge, UK) for statistical advice.