The base of most marine food webs is primary producers, but not all trophic chains start with living phytoplankton. Some animals, in particular sponges, can utilize Dissolved Organic Material (DOM) as a food source . The DOM uptake in sponges can be facilitated by bacterial symbionts. While all sponges have bacterial symbionts, there is a strong distinction between High-Microbial-Assemblage (HMA) sponges and Low-Microbial-Assemblage (LMA) sponges. HMA sponges can have 10⁸ to 10¹⁰ bacteria g⁻¹ sponge, whereas LMA sponges have 10⁵ to 10⁶ bacteria g⁻¹ sponge, which is roughly equivalent to the concentration of bacteria in seawater . This difference in microbiome size also means that HMA sponges have a higher assimilation efficiency of DOM compared to LMA sponges . LMA sponges, on the other hand, compensate for the decreased size of their microbiome and their ability to utilize DOM as a food source by relying on pumping large volumes of water and filtering more particulate organic material . In HMA sponges, the high abundance of associated microbes can account for up to 35 % to 40 % of the total biomass . This high microbial biomass is crucial for the sponge's nutrition and overall health, as these microbes play essential roles in nutrient cycling and DOM processing within the sponge. Furthermore, bacterial symbionts from HMA sponges possess a wide variety of genes that confer unique biogeochemical attributes not commonly found in other animals . The ability of sponges to consume DOM can have a ripple effect throughout the food chain since HMA sponges utilize organic matter that is not available to other organisms and transfer it to higher trophic levels . In this way, they play an important ecological role as pseudoautotrophic producers and facilitate the indirect transfer of DOM via sponges to higher trophic level predators. Additionally, DOM assimilated into larger organic material by sponges can be excreted as detritus, which can be consumed by other animals in a process known as the sponge loop .

Like phytoplankton, DOM can also bind Hg, but how DOM interacts with Hg depends on several factors, including the source of DOM. Marine DOM enhances phytoplankton MeHg uptake by facilitating active transport through membrane channels , whereas terrestrial DOM inhibits the uptake of MeHg by biota due to stronger thiol binding . DOM also binds inorganic Hg (iHg), influencing its speciation and bioavailability for microbial methylation . Since sponges can consume DOM, and DOM has different Hg binding patterns from phytoplankton, it raises the question of whether DOM consumption in sponges would consequently have different Hg bioaccumulation patterns compared to phytoplankton consuming benthos.

Studies analyzing iHg and MeHg in sponges found extreme variability in the iHg and MeHg content. found that in the Mediterranean Sea sponges generally have extremely low MeHg and a low MeHg / Hg ratio of 4 %, while sponges in the intertidal area of the Celtic Sea have very high MeHg concentrations and high MeHg / Hg ratios of between 7 % and 28 %. Recent efforts in sponge identification suggest that of the 4 sponges sampled from from the Mediterranean Sea, 3 species are classified as LMA sponges; Acanthella acuta , Cymbaxinella damicornis, recently reclassified under the genus Axinella , and Haliclona fulva , while 1 sponge, Chondrilla nucula, is classified as HMA . Since sponges can form the majority of the benthic biomass in sponge reefs and are commonly predated upon, their MeHg content likely influences MeHg levels in higher trophic level animals. This raises the question of whether MeHg bioaccumulation in higher trophic levels in (HMA) sponge dominated systems differs from the bioaccumulation in non-sponge-dominated ecosystems.

The low MeHg content in sponges in the Mediterranean Sea is proposed to be caused by active in vivo MeHg demethylation by . This is because HMA sponges have been found to contain bacterial symbionts that possess the MerA and MerB genes . The MerA gene transcribes the mercuric reductase protein that converts Hg²⁺ to volatile Hg⁰, and merB transcribes for organomercurial lyase, which catalyzes the breakdown of the C-Hg bond through protonolysis and can demethylate toxic MeHg into less toxic iHg . However, this does not explain the low concentration of MeHg in LMA sponges nor the high concentration of iHg in both LMA and HMA sponges. An alternative proposal for the low MeHg concentration in HMA sponges is that it is low due to the consumption of DOM. This is suggested in , which presents a modeling study examining how the feeding strategy influences iHg and MeHg dynamics. In this model, the MeHg partitioning to DOM, which is based on the MERCY v2.0 model, results in lower MeHg concentration in DOM compared to phytoplankton. This results in DOM consuming suspension feeders, such as sponges, having less MeHg bioaccumulation than filter feeders consuming phytoplankton, zooplankton, and detritus. In that model, the feeding strategy is considered in isolation, and other biological factors that could affect bioaccumulation are not included, such as in vivo demethylation, the long lifespan of sponges, and their low metabolic rate. Additionally, the model is simulated in conditions representing the North Sea, where HMA sponges are not abundant. The model presented in this manuscript expands on the model presented by by creating a more realistic model of sponges. We focus on incorporating the life cycles and feeding behaviors of megabenthos groups to enhance the understanding of iHg and MeHg bioaccumulation in Mediterranean Sea sponges while running the simulation under hydrodynamic and climatological conditions typical of the Western Mediterranean Sea, where both LMA and HMA sponges are common.

Most ecosystems rely on phytoplankton as the base of the food web, meaning that MeHg bioaccumulation can often be estimated from trophic level and phytoplankton MeHg concentrations. Sponge-dominated systems differ in that sponges can efficiently utilize DOM as a food source and exhibit extremely low MeHg concentrations compared to other megabenthos . This is important because the main predictor of MeHg concentrations in high trophic levels is the MeHg concentration at the base of the food web and the trophic level itself . Because sponges can form the majority of the benthic biomass in sponge reefs and are commonly predated upon, the MeHg content of sponges is likely directly linked to the MeHg content of higher trophic level animals, although this relationship remains understudied. We therefore hypothesize that the low MeHg concentrations in sponges lead to reduced MeHg bioaccumulation in high trophic level animals inhabiting sponge grounds compared to animals of similar trophic position in other ecosystems.

We test the following hypotheses using a modeling approach:

The low MeHg concentrations in LMA sponges can be attributed to their consumption of DOM, while the even lower levels observed in HMA sponges are likely due to both DOM consumption and active demethylation.

The Bay of Villefranche is a natural bay located just west of the French-Italian border, near the French town of Villefranche-sur-Mer. The bay is known for its deep oligotrophic waters, steep underwater topography, rocky inlets, and sandy bottoms. The nearby Villefranche Canyon creates diverse habitats that support a range of benthic communities. The model was run at a 17 m deep location (43°41^′53.23^′′ N, 7°18^′57.02^′′ E) to resemble the depth at which the sponge samples, to which the model is compared, were taken. Monitoring programs within the bay have mainly focused on planktonic research in coastal waters . Therefore, similar and abundantly available long-term benthic community data from nearby areas of the Gulf of Lyon were used to evaluate our model. In Fig. , we show the model domain; where Fig. 1a) depicts the Western Mediterranean Sea, while Fig. 1b) shows the exact modeled location in the Bay of Villefranche. The maps were created in R using Esri WorldImagery.

The one-dimensional (1D) Generalized Ocean Turbulence Model (GOTM) is used as the hydrodynamic model, which calculates the turbulence of a vertical 1D water column setup by computing the solutions to the 1D version of the transport equation of momentum, salinity, and temperature . The model is nudged to observational data sets for temperature and salinity. The setups are designed using the hGOTM tool, which is driven by gridded bathymetry data for water depth (1/240°) , the ECMWF ERA5 dataset for meteorological data , and the World Ocean Atlas for salinity and temperature profiles .

The MERCY v2.0 model and the ECOSMO E2E model are coupled to the GOTM model using the Framework for Aquatic Biogeochemical Modeling (FABM) . The biogeochemical models are coded into FABM. The FABM interfaces communicate the state variables between the GOTM model and the biogeochemical models. Physics is modeled with a vertical grid resolution of 0.71 grid cells m⁻¹. The model's state variables are updated every 120 s using the forward Euler method to solve the ordinary differential equation.

Nutrient cycling is normally dependent on lateral fluxes that are not present in a 1D water column model. To compensate for this, an atmospheric deposition of 0.01 mmol NO₃ d⁻¹, 0.01 mmol NH₄ d⁻¹, 0.5 µmol PO₄ d⁻¹, and 5 µmol SiO₄ d⁻¹ (d⁻¹ denotes per day) is introduced to compensate for the burial of organic material. This approach of using atmospheric deposition as a tuning parameter follows the methodology applied in the 1D GOTM-ECOSMO-MERCY setups for the North and Baltic Seas used in . The deposition values are chosen to produce realistic wintertime nutrient concentrations and support chlorophyll levels in line with observed data. This is further evaluated in the model evaluation section.

The physiological parameters of megabenthos within our model are derived from rates observed in laboratory conditions or through field studies. Given the inherent variability present in biological data, definitive rates are not always available and have been inferred from the referenced studies. The deposit feeders, generalist feeders, benthic predators, and benthic fish all feed based on a feeding rate and half saturation. Feeding and respiration rates are presented here as daily rates. Therefore, a feeding or respiration rate of 0.1 d⁻¹ would mean that the animal consumes or respires 10 % of its body weight per day. This modeling approach is based on the macrobenthos functional group in the original ECOSMO E2E model . The feeding of filter feeders and HMA and LMA sponges is based on their filtration rate within the bottom grid cells of the model. These groups consume a fraction of all food in the bottom grid cell, calculated as their filtration rate [m³ d⁻¹] divided by the volume of the bottom grid cell [m³], giving us the uptake [d⁻¹]. There is a maximum consumption rate beyond which a higher concentration of food in the water column will not lead to a higher intake rate. It should be noted that while we estimate the feeding, respiration, and mortality rates of megabenthos based on empirical studies, the modeled animal still represents functional groups. When rates were found in wet weight, a wet weight : dry weight ratio of 1:5 was assumed, and when respiration rates were presented by oxygen consumption, a respiratory quotient of 0.85 was assumed to convert oxygen consumption to a respiration rate .

The uptake rates of the deposit feeder are taken from ex-situ studies performed on a sand-eating lugworm (Arenicula marina), whereas this also represents other deposit feeders such as deposit feeding gastropods in the model. The respiration rate based on this study is 0.01 d⁻¹, and the mortality rate is 0.0005 d⁻¹. The respiration rate is estimated based on the measured oxygen consumption and converted to carbon respiration, assuming a respiratory quotient of 0.85 . The mortality rate is based on the observed life expectancy of 5 years. The grazing rate is 0.12, based on the measured maximum carbon consumption rate .

The filter feeders in our model are modeled after mussels (Mytilus sp.). They are parameterized to have a filtration rate of 9.60 × 10⁻⁴ m³ d⁻¹, a maximum feeding rate of 0.21 d⁻¹, a respiration rate of 0.00867 d⁻¹, and a mortality rate of 0.0063 d⁻¹ . The high filtration rate means that filter feeders can grow the fastest of all megabenthos groups in our model when food is abundant, but their high respiration rate means that they are competitively disadvantageous when food is limited.

The model organism for the generalist feeder is the brown shrimp (Crangon crangon). They eat microzooplankton, mesozooplankton, detritus, and sediment organic carbon. They are estimated to have a lifespan of 3 years, or a mortality rate of 0.0083 d⁻¹, and a maximum feeding rate of 0.12 d⁻¹ . found a respiration rate of per 51 µL O₂ h⁻¹ in a 91 mg⁻¹ d.w. shrimp, or 0.55 mg O₂ g⁻¹ d.w. This can be translated into a respiration rate of 0.00752 d⁻¹ based on a respiratory quotient of 0.85 . As a generalist, they thrive in highly variable circumstances where their ability to use different food sources gives them a competitive advantage.

ECOSMO E2E has two forms of pelagic aquatic organic carbon, detritus, and labile DOM (lDOM) . Since DOM is a key driver in giving sponges their competitive advantage over other suspension feeders, we also added semi-labile DOM (sDOM). Modeled sDOM does not have a sinking speed; both HMA and LMA sponges can consume it with the same efficiency as lDOM, and bacteria degrade it at the same rate as detritus. When organic material is formed in the model, it is formed as 38 % lDOM, 2 % sDOM, and 60 % detritus. There is an additional rate of sDOM formation from detritus of 0.001 d⁻¹. These formation ratios and rates are chosen to have a realistic sDOM value of 0–50 mgC m⁻³, based on , rather than on established rates of sDOM formation. This is described in more detail in .

In addition to the sDOM added to the model, we also added refractory DOM (rDOM). This DOM is estimated to represent 97 % of the global DOM pool . Estimates of rDOM in the Bay of Villefranche are not available, but measurements from a bay on Medes Island estimate the concentration of total DOM at 2.56 gC m⁻³ . Since our model has 0–50 mgC m⁻³ sDOM we estimate that an additional 2.5 gC rDOM gC m⁻³ could be present in our modeled domain. Given the uncertainty surrounding the extent to which sponges utilize rDOM when lDOM, sDOM, and detritus are available, and the substantial uncertainty of rDOM's binding to Hg compared to sDOM and lDOM, we run the model simulations both with and without 2.5 gC rDOM m⁻³. When rDOM is added, it is implemented as a fixed background concentration that is not consumed by bacteria or catabolized into nutrients. This rDOM reacts within the model in two ways. First, HMA sponges can take up rDOM. Since rDOM is represented as constant background concentrations, it is unaffected by sponge uptake. To address this, and considering its low energy value as a food source for sponges, its uptake is limited to 10 % efficiency compared to fresh DOM. Therefore, for HMA sponges, 2.5 gC rDOM m⁻³ corresponds to an equivalent of 0.25 gC m⁻³ lDOM, sDOM, or detritus in terms of its food availability. The difference in efficiency between the consumption of rDOM and other food sources is not based on a known value, but it is a necessary estimation, since if HMA sponges could feed on rDOM as efficiently as on other food fractions, they would overgrow everything, which is not what is observed in the Mediterranean Sea benthic food web. Because of this, it is used as a tuning parameter to increase the sponge biomass in a somewhat natural way, so we can evaluate the importance of the HMA sponges under different possible circumstances, and can evaluate the influence of sponges on fish MeHg bioaccumulation under a scenario with a high sponge biomass. rDOM is assumed to be mostly organic carbon with no other nutrients. The original ECOSMO E2E model assumes a Redfield Ratio in all consumers, but we deviate from this when HMA sponges consume rDOM to account for the low C : N and C : P ratio in rDOM. When HMA sponges consumes rDOM, the biomass originating from rDOM is tracked, and when they release carbon due to respiration or mortality, the consumed rDOM is not released as labile nutrients for phytoplankton. The same is true when predators and top predators consume HMA sponges directly or indirectly. It is traced how much of the organic carbon originates from rDOM, and when this rDOM-originating organic carbon is released, it is not transferred to dissolved nutrients that are usable by phytoplankton. It is possible that in real life sponges consume rDOM that contains micronutrients, and these nutrients later may be bioavailable, but this is understudied and out of the scope of this model. To summarize, the rDOM approach offers an approximation that allows HMA sponges and their predators to achieve greater biomass without violating the mass conservation principles of the rest of the model, and rDOM does not react directly with any biota, except HMA sponges in the model.

The second way in which rDOM interacts in our model is through partitioning to iHg and MeHg. In the MERCY v2.0 model, the binding partitioning between lDOM and iHg and MeHg is assumed to be based on partitioning coefficient for organic material (the Kd). We use the same partitioning coefficient for lDOM, sDOM, and rDOM. A partitioning coefficient of log(Kd)=6.4 for iHg and a coefficient of log(Kd)=5.9 for MeHg is used. This value is the same as is used in and is based on .

Our rDOM implementation addresses its absence in the previous version of the ECOSMO E2E model since it is essential for understanding sponge dynamics. The constant concentration reflects rDOM's refractory nature, and the 10 % efficiency assumes lower accessibility compared to fresh organic matter. This enables exploration of an under-studied mechanism in Hg cycling. We acknowledge that the rDOM implementation represents a simplified model designed to test whether this mechanism could explain observed patterns or to determine that this is numerically unlikely. The constant concentration and 10 % efficiency assumptions, while necessary for model stability, require empirical validation. This means that this approach can assist our hypothesis-generating model but is not able to generate definitive answers about the importance of rDOM for either sponges or Hg cycling.

LMA and HMA sponges were modeled separately. It must be stated that sponges are an incredibly diverse group, and comprehensive studies analyzing rates in sponges are rare. Because of this, we take the published rates of sponges and evaluate them as well as possible. LMA sponges have a maximum consumption rate of 0.014 d⁻¹, a filtration rate of 4.8 × 10⁻⁴ m⁻³ d⁻¹, and a combined respiration and mortality rate of 0.00257 d⁻¹, based on the LMA sponge Halichondria panicea . LMA sponges feed on phytoplankton, detritus, lDOM, and sDOM.

To distinguish the LMA and HMA sponges, we make two adjustments. First, the microbiome of HMA sponges gives them the ability to consume rDOM. However, this comes at a cost. HMA sponges are denser and filter water 52 %–94 % slower, due to their investment in maintaining bacterial symbionts . However, their uptake efficiency is higher because of the uptake of additional DOM through their symbionts. Because of this, we give HMA sponges a filtration rate of 3.60 × 10⁻⁴ m⁻³ d⁻¹, which is 25 % lower than the LMA sponges. In our model, HMA sponges feed on detritus, lDOM, sDOM, and rDOM and phytoplankton. When sponges are compared to observations, we assume a carbon to dry weight ratio of 1:5, based on . It is worth mentioning that while the 1:5 ratio was measured in Atlantic Deep Sea sponges rather than Mediterranean Sea sponges, it originates from the demosponges Geodia atlantica, which should resemble the demosponges sampled in .

The parameters and interactions with respect to bioaccumulation are based on the 1D model presented in . Importantly, our model models both the biomagnification and bioconcentration of both iHg and MeHg at every trophic level. In it is shown that this model can accurately model MeHg bioaccumulation at the base of the food web without tuning. The bioconcentration in phytoplankton depends on the size-dependent bioaccumulation and the release rate for both iHg and MeHg. The bioconcentration in all megabenthos is based on the macrobenthos functional group from and they have an uptake rate of 1.68 × 10⁻⁵ m³ mgC⁻¹ d⁻¹ and 2.20 × 10⁻⁵ m³ mgC⁻¹ d⁻¹ for iHg and MeHg, respectively. All megabenthos have a release rate of 0.04 d⁻¹ for iHg while there is no additional release rate for MeHg. Additionally, both MeHg and iHg are released with respiration and mortality. For biomagnification, an iHg assimilation efficiency (AE) of 0.31 is used and a MeHg AE of 0.95, when biota or detritus is consumed. When sediment is consumed, the AE of iHg is 0.07 and 0.95 for MeHg, based on . When DOM is consumed, the AE is 0.95 for both iHg and MeHg. This is based on the work of in phytoplankton, which shows that DOM can play an essential role in facilitating iHg transfer into cells. The assumption is that the DOM molecules are absorbed as a whole, and therefore the iHg associated with this would have a high assimilation efficiency for both iHg and MeHg.

We suggest that there are 2 pathways leading to HMA sponges having a lower MeHg content than LMA sponges; the consumption of DOM and in vivo demethylation. To assess the relative importance of both pathways we first model the bioaccumulation in HMA sponges under base conditions, where there is no rDOM for the sponges to feed on and the sponges do not demethylate. Afterwards, we simulate sponge bioaccumulation in a scenario where 2.5 gC rDOM m⁻³ is present. This allows us to quantify the effect of rDOM consumption on the bioaccumulation of MeHg in HMA sponges. We then perform a sensitivity analysis by varying demethylation rates in HMA sponges to identify the demethylation required to explain the observed MeHg bioaccumulation in LMA and HMA sponges.

Finally, we assess MeHg bioaccumulation in benthic fish by comparing the base case to the scenarios with 2.5 gC rDOM m⁻³. This allows us to estimate the potential impact of HMA sponges on MeHg concentrations in fish. We focus on the 2.5 gC rDOM m⁻³ scenario because it represents a high HMA sponge biomass resembling a sponge reef. The goal is to evaluate whether this effect is plausible, rather than to quantify its real-world magnitude, as the 1D water column model used here is not suitable for capturing spatial complexity.

The evaluation of MeHg bioaccumulation in LMA and HMA sponges involves assessing the Mean Percentage Bias (MPB), the equation for which is shown in Table in the Appendix. This metric was chosen because typically, in assessing models, the bias is used, which is the average difference between the model and observations; however, we could not use this due to the uneven amount of data between the model output and observations. Therefore, we first take the mean of the model output and then estimate the percentile bias between the modeled and observed means. This is further supported by performing a Kolmogorov–Smirnov (KS) and Wilcoxon signed-rank (W) tests. Statistical comparisons are used exploratorily to compare model scenarios. The KS test is conducted using the ks.test() function, and the Wilcoxon signed-rank test is performed with the wilcox.test() function in R (version 4.1.2, “Bird Hippie”). The evaluation considers the D-statistic, the Kolmogorov–Smirnov p-value (KS-p), and the Wilcoxon p-value (W-p). All tests are based on comparisons between the average daily model outputs from the last decade of the simulation and the observations reported by .

The D-statistic measures the maximum deviation between the cumulative distributions of the model and observations, thus measuring the maximum differences in the distribution. A D value below 0.1 suggests very similar distributions, values between 0.1 and 0.5 indicate minor to moderate differences, and a D value greater than 0.5 shows a large difference. The KS-p value calculates whether a significant difference between the modeled and observed distributions exists, while the W-p value estimates the significance of the difference between their medians. P-values below 0.05 for either KS-p or W-p are considered significant.

There is a large difference in sample sizes, with 3652 model data points compared to only 4 (HMA) and 6 (LMA) observations. This imbalance reduces the power of the KS and Wilcoxon tests, making them especially susceptible to false-negative outcomes. This means that high p-values cannot confirm model fit, whereas low p-values (<0.05) are more reliable indicators of a mismatch between the model and observations.

This limitation is particularly relevant for the Kolmogorov–Smirnov test, as it evaluates the distribution. With only 4 and 6 data points, the observational distribution cannot be robustly defined, but significantly low (<0.05) p-values still strongly indicate a mismatch between the modeled distribution and the limited observations.

Therefore, these metrics are mainly used to compare which setup performs better relative to other setups and to flag poor-performing configurations. However, they cannot be used to fully determine whether the model simulations are in agreement with observations due to the limited number of observations.

As mentioned in the methods section, the nutrient deposition is selected to create a realistic phytoplankton community to drive our megabenthos model. In our model, the wintertime nutrient concentrations are 0.12 µmol L⁻¹ NH4+, 5.0 µmol L⁻¹ NO 3-, 0.3 µmol L⁻¹ PO43-, and 1.6 µmol L⁻¹ Si(OH)₄. These values are on the higher end of observations in the Western Mediterranean Sea but align with elevated nutrient concentrations found in the Gulf of Lion, where surface nitrate and phosphate concentrations reach up to 3.58 ± 1.16 and 0.2–0.3 µmol L⁻¹, respectively . While the modeled silicate concentration is on the lower end of the observed range (1.6–5.8 µmol l⁻¹), based on the World Ocean Atlas 2018, silicate is not typically a limiting factor for phytoplankton growth in the Western Mediterranean, making this value acceptable for our modeling purposes . Based on this, we conclude that the modeled nutrient concentrations are in line with observations and can be used to simulate phytoplankton in our model.

We evaluate the modeled chlorophyll a concentrations against NASA MODIS-A satellite chlorophyll time series, for the same geographic location (https://oceancolor.gsfc.nasa.gov/, last access: 20 May 2026). The observational chlorophyll time series was sourced from the COPEPOD project . The NASA combined-satellite chlorophyll data presents monthly mean values, allowing for a comprehensive comparison. To evaluate whether the model captures the general chlorophyll a dynamics of the region, we focused on monthly means. Specifically, we averaged modeled chlorophyll values for each calendar month over the last decade of the simulation (e.g., the mean of all Januarys, all Februarys, etc.), and compared these to the corresponding monthly means from the NASA dataset.

While this approach smooths over interannual variability, it offers a way to assess whether the model captures the typical seasonal cycle and magnitude of phytoplankton. Given our focus on bioaccumulation processes, this comparison helps ensure that the modeled phytoplankton concentrations are within a realistic range and appropriate for driving biogeochemical dynamics in the Bay of Villefranche.

The results of the chlorophyll comparison between the model and the observations are shown in Fig. a, and the corresponding Taylor diagram is shown in Fig. b. The equations from which the metrics are derived are shown in Table . The modeled and observed chlorophyll concentrations exhibit a Pearson correlation of 0.59. The root mean square error (RMSE), is 0.127 mg m⁻³. The maximum observed chlorophyll is 0.54 mg m⁻³ and modeled is 0.55 mg m⁻³, while the observed mean is 0.31 mg m⁻³ and the modeled mean is 0.28 mg m⁻³, resulting in a mean bias of -0.028 mg m⁻³. Overall, the comparison reveals a low bias and relatively high correlation, indicating that the modeled phytoplankton dynamics are consistent with expectations for the Bay of Villefranche and provide a suitable basis for the biogeochemical processes explored in this study.

The benthic communities in this area have high specialization and display a patchy distribution, complicating the evaluation of our 1D model. According to , biomass values in the shallow Bay of Lyon vary widely, between 0 and 40 g ash-free dry weight m⁻², showing notable regional differences. Despite this, our modeled benthic biomass at 17 m ranges between 3.5 and 5 gC m⁻², as shown in Table and visualized in Fig. . This means that the biomass in our model is within the observed range. In the Gulf of Lyon, distinct benthic communities dominated by filter feeders (e.g., Spisula subtruncata, Venus ovata), deposit feeders (Scoloplos armiger), and generalist predators (Nephtys hombergii) have been identified . Although our 1D model cannot resolve these complex spatial differences, our model does support stable populations of all megabenthic functional groups, making it a reasonable representation of the general benthic community structure in the area.

Bioaccumulation starts with the dissolved concentration of the pollutants. Therefore, it is essential to assess that the MERCY v2.0 model generates concentrations of Hg and MeHg that are in line with observations. Our model estimates Hg and MeHg concentrations in the Bay of Villefranche of 0.97 ± 0.087 and 0.018 ± 0.005 pM respectively. Observations in shallow coastal stations in the Gulf of Lion by found mean Hg concentrations of 1.52 ± 1.00 pM and MeHg concentrations of 0.026 ± 0.024 pM; however, some MeHg concentrations were below the detection limit, introducing uncertainty. This means that our modeled Hg and MeHg concentrations are well within the observed range and within 1 standard deviation, with a MPB of -36 % for Hg and -31 % for MeHg compared to the observed mean. Based on these results, we conclude that our model reproduces observed Hg and MeHg concentrations in the water, making it suitable for use in our bioaccumulation modeling framework.

The biomass and bioaccumulation biota are shown in Table . We first discuss the base case, the simulation without rDOM. Then we discuss the differences in the simulation with 2.5 gC rDOM m⁻³ and how this compares to observations. When data were reported in dry weight, we assumed a 1:3 dry-weight-to-carbon ratio for plankton and a 1:2 ratio for megabenthos and fish to compare our model output in mgC with the bioaccumulation observed in dry-weight measurements. When fish were reported in wet weight, an additional 1:5 dry-weight-to-wet-weight ratio was assumed . Overall, the model reproduces the observed ranges and trophic-level trends reported in the literature. MeHg bioaccumulation in phytoplankton is observed at 2.4 ± 0.8 ng Hg gC⁻¹ by , compared to 1.8 ± 4.2 ng Hg gC⁻¹ in the model. Zooplankton observations range from 9–21 ng Hg gC⁻¹, with copepods reaching up to 31 ng Hg gC⁻¹ while the model predicts 11.8 ± 4.5 ng Hg gC⁻¹ for microzooplankton and 12.6 ± 3.6 ng Hg gC⁻¹ for mesozooplankton, so the model is within the observed range.

Filter feeders (oysters and mussels) show observed bioaccumulation of 18–290 ng Hg gC⁻¹ along the French coast, and 48–148 ng Hg gC⁻¹, with a mean of 100 ng Hg gC⁻¹, in the Bay of Cannes, which are their closest observations to the Bay of Villefranche . The model predicts 35 ng Hg gC⁻¹, indicating that the model likely underestimates MeHg bioaccumulation in filter feeders, but the modeled mean still falls within the observed range. Benthic predators have roughly double the MeHg of filter feeders, consistent with biomagnification. Observed MeHg in benthic fish (422 ng Hg gC⁻¹) aligns with mid-trophic Mediterranean species such as common sole (400 ng Hg gC⁻¹) and white seabream (310 ng Hg gC⁻¹). In the 2.5 gC rDOM m⁻³ scenario, low-trophic biota show slightly higher bioaccumulation, likely due to reduced growth dilution. In this scenario, filter feeders are largely outcompeted by sponges. The benthic predator shifts its diet towards HMA sponges, reducing its MeHg to 32 ng Hg gC⁻¹ (still approximately three times higher than its diet), and benthic fish MeHg decreases to 205 ng Hg gC⁻¹, consistent with lower-end observations for common sole and closer to species with lower MeHg, such as blackspot seabream or common pandora. Due to the 1D model structure and limited observations, detailed quantitative validation is not possible, but both setups, with and without rDOM, simulate bioaccumulation consistent with observations.

Figure shows the bioaccumulation of MeHg in HMA sponges. Without demethylation, the average bioaccumulation of MeHg is higher than the observations in HMA sponges. Having a demethylation rate of 1 % d⁻¹ reduces the bioaccumulation of MeHg to 9 ng Hg gC⁻¹, which better aligns it with the average observed MeHg concentration of 13 ng Hg gC⁻¹ in HMA sponges. Furthermore, increasing the demethylation rate to 2.5 % d⁻¹ aligns the MeHg content of the HMA sponges with the lowest observations in the HMA sponges. The statistical comparison of the base model and the scenario with 1 % and 2.5 % demethylation per day is shown in Table . In the base model, there is a positive MPB for the MeHg content of HMA sponges (+113 %) and the KS-p of <0.001 and W-p of <0.001 show a significant variation between the observed and modeled data, both in their distribution and median values. In the scenario where HMA sponges have 1 % demethylation, the HMA sponges have a smaller negative bias (-30 %) while the D statistic of 0.48 shows a minor difference in the distribution of the modeled and observed bioaccumulation, while the KS-p of 0.309 shows that this difference is not statistically significant. Additionally, the W-p of 0.76 shows that there is also no statistically significant difference between the observed and modeled median MeHg bioaccumulation. In the 2.5 % demethylation per day scenario, the KS-p values (0.023) show no significant difference, while W-p (0.003 ) value does at the 95 % confidence internval. However, both values are considerably lower compared to the 1 % d⁻¹ scenario. These reduced p-values, combined with the high D value of 0.75 and an increased negative MPB of 65 % show that the 2.5 % demethylation per day underestimates the MeHg content of HMA sponges and has lower agreement with the observations than the scenario with 1 % demethylation per day. Under our parameterization assumptions, the model suggests that demethylation rates of 1 % d⁻¹ provide reasonable agreement with observations, though this depends critically on uncertain uptake parameters.

We investigate the hypothesis that HMA sponges can have low MeHg due to their reliance on DOM as food. To investigate this, we ran the model with 2.5 gC rDOM m⁻³, which could be consumed by HMA sponges. For HMA sponges, there is no significant discrepancy between the observed and modeled distributions, as signified by a KS-p value of 0.77. Additionally, the D-statistic of 0.33 indicates a moderate similarity between the modeled and observed distributions. The MPB of -2 % along with a W-p value of 0.87 indicates a small bias in the mean, and no significant difference between the modeled and observed median. To summarize, in the scenario with 2.5 gC rDOM m⁻³ there is no significant difference in either the median or the distribution of MeHg bioaccumulation in HMA sponges. Since the consumption of rDOM alone accounts for the low MeHg levels in HMA sponges, the model was not rerun with a demethylation rate and 2.5 gC m⁻³, as this was unnecessary to explain the observed low MeHg concentrations in HMA sponges.

In our model, consumption of rDOM can influence MeHg bioaccumulation in three ways. First, in our current parameterization, rDOM binds dissolved MeHg, with the same strength as lDOM and sDOM. If we include rDOM, the MeHg concentration per DOM can be reduced due to biomass dilution. This will result in lower concentration of MeHg in suspension feeders consuming DOM, and consequently in predators consuming these suspension feeder. Secondly, the consumption of rDOM by HMA sponges allows the consumption of carbon that is otherwise not bioavailable. This increases the biomass of HMA sponges, and consequently, it increases the biomass of the benthic predator and benthic fish. The shift in biomass causes a shift in the diet of the benthic predator; in the base model, the majority of their diet would consist of deposit feeders and LMA sponges, whereas in the 2.5 gC rDOM m⁻³ setup, their diet would mostly consist of HMA sponges followed by deposit feeders. Since HMA sponges have an extremely low MeHg concentration, this results in low MeHg in benthic predators and, consequently, in benthic fish. The concentration of MeHg in benthic fish is 422 ng Hg g⁻¹ in the base model and only 205 ng Hg g⁻¹ in the setup with 2.5 gC rDOM m⁻³. If we run the model with 2.5 gC rDOM m⁻³, but without HMA sponges, the bioaccumulation of MeHg in fish is 437 ng Hg mgC⁻¹, so comparable to the base case. This means that the HMA sponges in this setup reduce the MeHg content of the fish by 53 %. This effect is not present in the base model, nor in the setup with 1 % demethylation per day, which can be explained by the low biomass of HMA sponges in the base setup compared to the 2.5 gC rDOM m⁻³ setup. This shows that the 53 % reduction is an upper bound of the reduction that can be expected in benthic fish in ecosystems that are dominated by HMA sponges at the base of the food web compared to ecosystems without sponges.

Sponges are observed to have an extremely high iHg content. observed a mean iHg content of 1523 and 696 ng Hg mgC⁻¹ in HMA and LMA sponges, respectively, assuming a carbon to dry weight ratio of 1:5. Even assuming an AE of 0.95 iHg when DOM is consumed, our modeled iHg bioaccumulation is still an order of magnitude lower than observations. Although fully understanding this is out of the scope of the model, we reproduced the observed concentrations only when the turnover and release rate of iHg in LMA and HMA sponges was completely removed. Under these circumstances, we obtained a bioaccumulation of 1448 and 592 ng Hg mgC⁻¹, or a MPB of 5 % and 15 % in HMA and LMA sponges.

Since we can replicate the low MeHg content of HMA sponges in the 2.5 gC rDOM scenario, we conclude that in vivo MeHg demethylation is not necessary to explain the observed iHg and MeHg content of HMA sponges. However, the bioconcentration rates of iHg and MeHg in HMA sponges in the model are highly uncertain. If the actual uptake rates are higher, a higher demethylation rate would be required to reproduce the observed values. To estimate an upper bound for the demethylation rate, we assume uptake rates sufficient to reproduce the observed high iHg concentrations in LMA and HMA sponges.

This is achieved by simulating a tenfold increase in LMA sponge uptake rates and a twentyfold increase for HMA sponges in the base case scenario. The resulting uptake rates are 1.68 × 10⁻⁴ and 3.35 × 10⁻⁴ m³ mgC d⁻¹ for iHg in LMA and HMA sponges, respectively, and 2.20 × 10⁻⁴ and 4.41 × 10⁻⁴ m³ mgC d⁻¹ for MeHg. Modeled iHg and MeHg concentrations under these conditions are shown in Table . Under these circumstances, a demethylation rate of 13 % d⁻¹ in HMA sponges gives the best agreement with observations. The demethylation required to reproduce the low MeHg concentrations, therefore, represents a loosely constrained upper bound of the maximum expected demethylation rate.

An additional method for the model to reproduce the high iHg content of sponges is to reduce the release rate. As shown in Table , the model also reproduces the high iHg content of both LMA and HMA if we assume that iHg is strongly retained in sponges and is only released at a rate equal to the respiration rate of sponges. This model is referred to as the low release rate model (LR). In this parameterization, there is no additional release rate of iHg in sponges, and iHg is only released with the metabolic rate. This parameterization can reproduce both the iHg and MeHg bioaccumulation in both LMA and HMA sponges with an MPB of 30 % or lower.

As discussed in the Methods section, the uptake of rDOM and its efficiency are poorly understood, and we consequently used this parameter for model tuning. In the 2.5 gC rDOM m⁻³ scenario, we assumed an rDOM uptake efficiency of 0.1. To analyze how rDOM consumption and sponges influence MeHg bioaccumulation in benthic fish, we run the model with rDOM uptake efficiencies of 1, 1/5, 1/10, 1/15, 1/30, 1/75 and 1/100. The results are shown in Fig. . Figure a shows that increasing the efficiency of uptake of rDOM leads to an increase in the biomass fraction of sponges, while Fig. b shows that HMA sponges comprise a larger fraction of benthic biomass. At 10 gC m⁻², further sponge growth becomes limited in our model due to the assumed substrate limitation. As a result, the fraction of benthic biomass composed of HMA sponges increases rapidly until, at an uptake efficiency of approximately 0.25, sponges represent about 70 %–80 % of the benthic biomass. Further increases in uptake efficiency of up to 1 do not lead to additional increases in the HMA sponge fraction. Figure b also shows that as the fraction of benthic biomass composed of HMA sponges increases, the bioaccumulation of MeHg in benthic fish decreases. A visual clustering is observed once sponges exceed approximately 50 % of benthic biomass, suggesting a potential saturation effect by which sponges dominate predator diets. However, this pattern is likely influenced by model structure and feeding parameterisations and may not reflect a general ecological mechanism, as natural systems typically include multiple specialized predators rather than a single generalist consumer.

The difference in bioaccumulation of iHg and MeHg highlights Hg as a unique pollutant. Certain biological pathways, such as the direct absorption of MeHg via membrane channels in phytoplankton, enable highly efficient MeHg bioaccumulation. As a result, MeHg concentrations in biota can exceed iHg concentrations, even in primary producers, despite the fact that dissolved iHg is generally more abundant than MeHg. In DOM, which consists primarily of dead organic particles, these preferential uptake pathways are absent, and DOM binds iHg and MeHg only through scavenging. The composition of DOM strongly influences its binding capacity: thiol-rich fractions of terrestrial DOM efficiently bind to MeHg , while marine DOM fractions show a higher affinity for iHg , which underpins the partitioning behavior used in the MERCY v2.0 model .

The lack of efficient MeHg uptake mechanisms, combined with low dissolved MeHg concentrations, results in low MeHg bound to DOM in our model, while iHg is higher due to the efficient scavenging of iHg by DOM and the higher dissolved concentration of iHg compared to MeHg. Consequently, when HMA sponges feed on DOM, they accumulate very low MeHg concentrations, whereas the efficient partitioning of iHg to DOM results in elevated iHg levels, even though iHg is not typically transferred via diet very efficiently.

As mentioned in the model development segment, an rDOM concentration of 2.5 mgC m⁻³ is observed. However, as the name suggests, this rDOM is refractory and generally unresponsive. This raises the question as to what degree it should be incorporated in the model. There are 2 key questions;

To what degree do sponges feed on rDOM compared to lDOM and sDOM

While demethylation and DOM consumption can both play a role in causing the low MeHg content of sponges, especially HMA sponges, they certainly have an extremely low MeHg content. In addition, HMA sponges can have large biomass and be the main part of low-trophic-level biomass. Our model shows that this can dramatically reduce the MeHg content of higher trophic level benthic fish; in our model, this is a reduction of 53 %. This percentage is, of course, highly dependent on the MeHg concentration of sponges, the MeHg concentration of other megabenthos, and the degree to which sponges are consumed. While the role of sponges in the ecosystem is understudied, it is clear that sponges are both directly and indirectly consumed by important commercial species of fish. The 53 % reduction in MeHg bioaccumulation of fish represents a potential quantification of this effect under extremely high HMA sponge biomass scenarios, while assuming HMA sponges are a critical link in the food chain. These conditions represent an idealized scenario for this effect, which means that it represents an upper bound estimate that requires further study under natural conditions to be quantified.

The model mainly focuses on explaining the low MeHg content of the sponges and the observed difference between LMA and HMA sponges. However, investigating the unusually high iHg content of HMA sponges is an essential part of the Hg sponge biochemistry and should be taken into account. We demonstrate that we can explain both the iHg and MeHg concentrations found in HMA sponges by assuming an iHg and MeHg uptake rate of 3.35 × 10⁻⁴ m³ mgC d⁻¹ for iHg and 4.41 × 10⁻⁴ m³ mgC d⁻¹ for MeHg. The issue with this parameterization is that we can offer no explanation for the low MeHg content of LMA sponges. An additional way we can reproduce the observed iHg content of sponges is sponge retain Hg stronger than other macrobenthos. We can reproduce the high iHg content of both HMA and LMA sponges if we assume that the release rate of iHg is equal to the carbon respiration rate of the sponge, which is much lower than what has been observed for other animals such as small crustaceans and bivalves . While this is speculative at the moment, a potential explanation can be found in the unique structure of sponges. A study on the glass sponge Euplectella aspergillum found that they have elevated Hg levels in the spicules. C. nucula is a demosponge, not a glass sponge, but it still has glass spicules. But a similar buildup of iHg in structural components of the sponge might be happening in other components of the sponge. Demosponges rely on a protein called spongin and sulfated polysaccharides (SPs) to bind the sponge together. A possible explanation for the reduced release rate of iHg in sponges could be that iHg remains bound to the SPs in sponges.

We propose this for three reasons.

SPs in C. nucula have a very high sulfate to sugar ratio of 1:5 .

Since our model can reproduce MeHg concentrations without demethylation in both HMA and LMA sponges, it shows that demethylation may not be necessary to describe the observed patterns. This does not mean that demethylation cannot play a role, but since LMA sponges do not contain large amounts of sulfate-reducing bacteria that can demethylate MeHg, a compelling argument can be made based on our model that the low MeHg concentration in LMA sponges is caused by the consumption of DOM and detritus, which make up a substantial part of their diet. The even lower MeHg content of HMA sponges can then be explained by an increased dependence on DOM by HMA sponges, in vivo MeHg demethylation, or a combination of both.

The expected demethylation rate is linked to the uptake rate of MeHg, which is uncertain. Because of this, we see two potential pathways that lead to the observed bioaccumulation of iHg and MeHg in LMA and HMA sponges.

The scenario with no in vivo MeHg demethylation, as is explored in the 2.5 gC rDOM scenario, and low in vivo MeHg demethylation, as is explored in the 1 % demethylation scenario, can both reproduce the observed MeHg concentration in LMA and HMA sponges, but lack a full explanation for the elevated iHg concentration in sponges. Because of this, we conclude that both no or low in vivo MeHg demethylation in HMA sponges is possible, and we supplement this paper with a proposed explanation on why sponges could have high iHg based on their chemical composition.

Consequently, we consider it most likely that the high iHg content of sponges results from a reduced release rate of iHg, possible due to iHg accumulation on SPs or spongin. The low MeHg levels in LMA sponges are attributed to the consumption of DOM, while the reduced MeHg levels in HMA sponges arise from DOM consumption, possibly accompanied by an in vivo demethylation rate of 1 %.

A major simplification in our model lies in the implementation of rDOM. In our setup, rDOM serves as a proxy for organic material that can be consumed by HMA sponges but is not dissimilated by non-sponge-associated microbes on timescales relevant to our model. This approach is particularly suited to a 1D configuration, where a constant flow of rDOM can be assumed throughout the system. The current implementation was chosen to ensure consistency in nutrient availability for phytoplankton growth, to preserve mass balance across the model, and to enable HMA sponges to use rDOM for growth; as such, it serves as a useful proof of concept to explore whether rDOM could play a meaningful role in ecosystem functioning. However, this implementation is limited in scope. In a 3D model or in stratified water columns where vertical mixing is reduced, rDOM transport may significantly influence the distribution and cycling of Hg. Furthermore, potential effects of rDOM, such as its role in light attenuation, are not incorporated in the model. To fully assess the role of rDOM in both HMA sponge ecology and Hg biogeochemistry, cycling, and bioaccumulation, a more realistic representation will be necessary.

Beyond the limitations of the rDOM implementation, several major limitations affect the model. The most important is the low data availability. Our conclusions rely on a single study of MeHg and iHg bioaccumulation in Mediterranean sponges, with small sample sizes (n = 4 for HMA and n = 6 for LMA sponges). While the model–data comparison allows identification of clear mismatches between the model and observations, the limited number of observations prevents reliable validation of simulated Hg bioaccumulation patterns. Therefore, the agreement between model simulations and observations should be interpreted with care.

In addition, key processes remain poorly mechanistically understood. The mechanisms and rates of (r)DOM consumption by sponges, the role of sponges in ecosystem-level Hg bioaccumulation, and the functional differences between LMA and HMA sponges are not fully understood. These knowledge gaps limit process-based parameterization and increase structural uncertainty in the model.

Because of this, targeted empirical studies on Hg bioaccumulation in sponges, including controlled experiments focused on the bioaccumulation pathways that were flagged as potentially significant in this model, such as the role of the uptake of DOM by sponges on Hg bioaccumulation and the consumption and trophic transfer of Hg of sponges by predators, could substantially improve model development by enabling more robust validation.

We focus on qualitative model insights rather than extensive sensitivity analyses. While all parameters carry uncertainty, our primary objective is to demonstrate that the proposed mechanisms are quantitatively plausible alternatives to the demethylation hypothesis proposed by . Comprehensive sensitivity analyses would confirm parameter uncertainty without providing additional mechanistic insights. Rather, we demonstrate that our base parameterization can reproduce the observed bioaccumulation of MeHg in both LMA and HMA sponges linked to the consumption of DOM and rDOM, which shows that DOM consumption by LMA sponges and rDOM consumption by HMA sponges could reproduce the observed patterns.

Based on the literature, we assume that sponges form an integral part of the food chain, leading to our hypothesis that sponges might reduce MeHg bioaccumulation in animals above them in the food web, including commercially important fish species. The predicted 53 % reduction in benthic fish MeHg concentrations represents a hypothesis-generating finding that demonstrates the potential magnitude of sponge-mediated effects. This should be interpreted as a loosely constrained upper bound estimate, given our generous parameterization of HMA sponge biomass and role in the food chain. The actual magnitude would depend on local sponge abundance, community structure, and food web dynamics. Both our mechanistic understanding of food web dynamics in the Bay of Villefranche and the role of sponges in them, and the dynamics between sponges and iHg should be validated with empirical studies before this estimation can be meaningfully constrained using sensitivity studies.

Sponges are unique in their ability to bioaccumulate Hg, as they both have extremely high levels of iHg while maintaining very low levels of MeHg. This characteristic makes them a prime candidate for bioremediation studies. If the model is accurate for sponge grounds, they could be advantageous for Hg remediation. On the one hand, in Hg-polluted regions, sponges could bioaccumulate the less toxic iHg, which could then be extracted for bioremediation, with a diminished risk of MeHg since the majority of the Hg remains in its unmethylated form. Furthermore, the low MeHg content in HMA sponges means that organisms feeding on them would accumulate less MeHg, ultimately resulting in lower MeHg levels in higher trophic levels, as demonstrated in our model. These findings suggest that actively managing sponge grounds could both mitigate MeHg accumulation in fish and provide a sustainable method for extracting Hg from the marine environment.