Nine mesocosms with a volume of 275 L for the main tank and 70 L for the settling tank were designed following the Jaubert method (Jaubert, 2013). The Jaubert method uses a large quantity of substrate to promote biological water filtration through the activity of an active bacterial community. Here, the substrate consisted of sand and coral stones taken from Vairao lagoon (Tahiti, French Polynesia). After sterilizing the sand and rocks through successive treatments with fresh water and UV radiation, we added 65 L of sand and 17 kg of rocks (10 kg in the main tank and 7 kg in the settling tank) to each mesocosm. Following the addition of sand and rocks, we calculated the available water volume in each tank to be 270 L. The parameters of the tanks were maintained at the following values, the temperature was 27 ± 0.3°C, the pH was 8 ± 0.1, and the salinity was 36.1 ± 0.2 psu. The lighting was turned on for 8 hours per day. After one year of cycling, allowing for the establishment of denitrifying bacteria in the substrates, we introduced marine organisms. In each mesocosm six giant clams Tridacna maxima (shell length: 4 ± 1 cm) were added. To rid them of potential parasites, the clams underwent a one-minute freshwater bath before being placed in the mesocosms. After observing several cases of fireworm predation during the acclimation phases which appear to originate from an initial batch of giant clams that had not undergone parasite removal, we decided to suspend all the animals to prevent direct contact with the substrate ( Figure 1 ). The concentrations of plastics per mesocosm are shown in Table 1 .

Bivalves have already proven to be reliable bio-indicators for assessing the pollution status of environments (Ding et al., 2021; Prokić et al., 2021; Multisanti et al., 2022). Sentinel species must meet the following criteria: wide distribution range, a well-known biology, sedentary nature, ability to provide early alerts, key function in the ecosystem, homogeneous response to pollutants, and the existence of identifiable toxic effects associated with the degree of pollution (Li et al., 2019). By meeting several of these criteria, Tridacna maxima, a tropical giant clam, may be an ideal sentinel species. This sessile filter-feeder species is widely distributed in French Polynesia (Andréfouët et al., 2005; Gilbert et al., 2005). Tridacna maxima is already employed as a sentinel for Metallic Trace Elements (MTE) in Polynesia (Godéré et al., 2023). Furthermore, research teams have demonstrated that in polluted areas, the quantity of zooxanthellae in the mantle and the frequency of giant clam valve movements, as monitored by valvometry, serve as reliable indicators of environmental pollution by contaminants such as lead (Yaqin et al., 2020; Killam et al., 2023). It has also been demonstrated that this species is capable of bioaccumulating microplastics (Arossa et al., 2019). Negative effects, such as oxidative stress and physiological alteration resulting from plastics, have been observed in other bivalve species exposed to contaminated environments (Sureda et al., 2018). Moreover, the giant clam requires a limited amount of nutrient, which aligns with the Jaubert method used here (i.e., low water renewal).

The lowest concentrations in this range (i.e., 0.0019 g/L) were estimated based on the quantities of waste produced by pearl farming in certain Polynesian atolls (DRM, 2015). However, the distribution of plastics in lagoons can be highly heterogeneous due to complex lagoon hydrodynamics (Bruyère et al., 2024). As a result, local plastic concentrations may be significantly higher than global estimates at the lagoon scale, with certain plastic accumulation zones acting as hotspots for chemical release (Gardon et al., 2021). To encompass a broad spectrum of potential plastic concentrations in pearl farming lagoons, we selected a range from 0.0019 to 4.0500 g/L, using a base-3 logarithmic interval between each concentration ( Table 1 ). This range includes plastic concentrations typically tested in toxicological studies on plastic leaching (Ke et al., 2019; Gardon et al., 2020a; Klein et al., 2021; Wang et al., 2023).

The plastics used in this experiment were supplied by a pearl farm on the Mangareva atoll in the Gambier Archipelago and consisted of standard new ropes and collectors commonly used in pearl farms across French Polynesia. The collectors are spongy plastic structures used for recruiting oyster spat. Previous RAMAN analyses confirmed that these materials are composed of polyethylene (PE) and polypropylene (PP) (Gardon et al., 2020a).

After two months of experimentation, we collected 500 mL of seawater from each mesocosm. The samples were subjected to chemical characterization in duplicate at CEDRE (Center for Documentation, Research, and Experimentation on Accidental Water Pollution, Brest, France) employing gas chromatography (HP 7890N, multifunction injection Combipal MPS2, Gerstel) coupled with tandem mass spectrometry (GC/MSMS). The interface temperature was maintained at 300°C, with a pre-programmed injection ranging from 50°C (0.5 min) to 280°C (6 min) at a rate of 15°C/min. This setup was integrated with a temperature-programmed injector (Cooling Injection Device, Gerstel: -10°C (0.05 min) to 300°C (10 min) at 12°C/s). The oven temperature program started at 70°C (0.5 min) to 150°C at 20°C/min, followed by an increase to 320°C (5 min) at 7°C/min. Helium served as the carrier gas, and the capillary column employed was an RXi 5-ms (Restek, Bellefonte, USA). The chromatograph was linked to a tandem mass spectrometry detector (Agilent 7000 Triple Quad).

Quantitative analysis of Polycyclic Aromatic Hydrocarbons (PAHs), Polychlorinated Biphenyls (PCBs), Polybrominated Diphenyl Ethers (PBDEs), pesticides, and additives was carried out using internal standard calibration in Multiple Reaction Monitoring (MRM) mode, incorporating two transitions for each compound (Quantifier/Qualifier). The acquisition frequency for each fragment was set at 2 cycles per second. MassHunter Workstation V10.0 software (Agilent Technologies) used to reprocess of chromatograms in the analysis of organic compounds.

Prior to introducing the plastic into the mesocosms, two liters of water from each mesocosm were collected, filtered through 0.22 µm pore size polycarbonate membrane filters, and stored in lysis buffer (2 % CTAB, 1.4 M NaCI, 100 mm Tris-HCl pH 8, 20 mM EDTA pH 8 and 200 μg/mL) at -80°C after grinding and digestion with proteinase K. One giant clam from each mesocosm was also sacrificed at T0, a portion of their muscle was extracted and frozen at -20°C while their gills were preserved in lysis buffer CTAB 2 % at -80°C. Growth rate, a biological parameter of bivalves known to be impacted by the presence of plastic in the environment (Détrée and Gallardo-Escárate, 2018), was measured by recording the shell length of all the giant clams at different time points during the experiment using a caliper. The growth rate was calculated as the difference between initial and final shell lengths divided by the number of days in the experiment. The plastics were then randomly immersed in the mesocosms, making the initial time of the experiment (T0). Every fifteen days (i.e. T1, T2, T3), and continuing until the end of the experiment, two liters of water from each mesocosm were collected and processed as described above. After 60 days of plastic immersion, the experiment was stopped (i.e. T4). All remaining giant clams were sacrificed, and their gills and muscles were collected and preserved as previously. Samples were also collected from the macroplastics (ropes and collectors) and the water in each mesocosm. All samples were stored at -80°C in 2 % CTAB lysis buffer awaiting DNA extraction.

The presence of plastic in the environment can result in adverse health impacts on the energy balance of marine organisms such as pearl oyster Pinctada margaritifera (Gardon et al., 2018). One of the proxies chosen to assess the health of the giant clams was the quantification of their muscle glycogen reserves. After thawing the giant clam muscles, 100 mg were sampled, transferred to Eppendorf tubes to which 500 µL of Trichloroacetic Acid (TCA) and 3 stainless steel beads were added. The samples were ground for 5 minutes with a grinder set to 30 beats per second. Once the sample was ground, 2.5 mL of 15 % TCA was added. The sample was then incubated at 4°C for one hour. The tube was then centrifuged for 10 minutes at 3 000 g and at a temperature of 4°C. After centrifugation, 1.5 mL of the supernatant was recovered from the tube and supplemented with 12 mL of 96 % ethanol. The samples were then stored at 4°C for 12 hours with agitation (shaking platform). The next day, the tubes were centrifuged at 4°C for 30 minutes at 4 000 g. The supernatant was removed, and 200 µL of pure water was added to the pellet before a brief vortex to re-solubilize the pellet (containing glycogen). Glycogen quantification was performed through a colorimetric reaction with anthrone-sulfuric acid. The absorbance of the resulting colored complex was measured at a wavelength of 620 nm (FLUOstar, BMG Labtech).

The normality of the data was tested using a Shapiro-Wilk test, and homoscedasticity was assessed using a Levene test on the growth rate and glycogen concentration in the muscle tissues of giant clam individuals. As normality conditions were not met in our datasets, we conducted a Kruskal-Wallis test to determine if there were significant differences in the growth rate and glycogen concentration among the giant clams subjected to various plastic concentrations. The significance level chosen for these analyses was a P-value < 0.05. All these analyses were performed using R v4.0.5.

Bacterial assemblages were assessed for three compartments: the surface of plastics, the water column and the giant clams (Tridacna maxima). Samples for DNA extraction consisted of polycarbonate PCTE membrane filters of 2 L of water, 50 to 300 mg of gills tissue, and plastic smears respectively.

DNA was extracted using CTAB (cetyl trimethylammonium bromide)-based extraction protocol adapted from (Winnepennickx et al., 1993). Briefly, 600 μL of CTAB 2 % buffer was added to each sample tubes (filters, cotton swabs and clam gill tissue) as well as three sterile stainless-steel balls of 3 mm-diameter. Tubes were then immediately placed in a vibrating ball mill (Mixer Mill MM400, Retsch, Haan, Germany) for grinding at 30 Hz for 3 minutes and further incubated overnight at 60°C. After thawing the samples at room temperature, an equal volume of phenol-chloroform-isosoamyl alcohol (25:24:1) was then added to each sample. After gentle inversion of tubes manually samples tubes were centrifuged for 10 minutes at 12 000 g and 400 µl of the aqueous phase transferred to a new tube. DNA was precipitated by adding an equal volume of isopropanol and incubated at room temperature 10 minutes after gentle inversion of tubes. After centrifugation for 20 min at 12 000 g, the supernatant was discarded and the pellet washed with 70 % ethanol. The pellet was air-dried for 10 minutes and resuspended in 50 µl of DNase-free water.

After DNA extraction, the V4 region of the bacterial 16S SSU rRNA gene was amplified using the primers 515F (5’-GTGYCAGCMGCCGCGGTAA-3’) (Parada et al., 2016) and 806R (5’-GGACTACNVGGGTWTCTAAT-3’) (Apprill et al., 2015). PCR products were submitted for 2x250 paired-end sequencing on an Illumina MiSeq platform at the Genome Quebec platform, Montréal (Canada). Samples with fewer than 10 000 reads were not considered for subsequent analyses. Sequences have been deposited in the ENA database (BioProject: PRJEB74057).

The fastq files generated by Illumina MiSeq were processed using QIIME2 v2022.11 (Bolyen et al., 2019) and DADA2 (Callahan et al., 2016). Primers were removed from the demultiplexed sequences using Cutadapt (Martin, 2011) (via q2-cutadapt). DADA2 parameters were selected using the FIGARO tool [44]: trim Position [221,229] and max Expected Error [1,1]. The following steps of the DADA2 pipeline were performed using the default parameters (via q2-dada2). Amplicon sequence variants (ASVs) clustering was performed using dbOTU3 (Olesen et al., 2017) with the default settings (via q2-dbOTU3). Taxonomy was assigned to ASVs using the q2‐feature‐classifier (Bokulich et al., 2018) classify‐sklearn naive Bayes taxonomy classifier against the SILVA138 (Quast et al., 2013) pre-trained classifier for the V4 region of the 16S gene.

Eukaryotic, mitochondrial, chloroplast and unidentified sequences were removed from the dataset. ASVs with a relative abundance greater than 0.5 % and present in fewer than three samples were filtered out using the phyloseq v1.42.0 R package and a custom function. First, we estimated the Bray-Curtis distance matrix from the relative abundance dataset (samples in rows, ASVs in columns) based on the recommendation of Laporte et al. (2021) for eDNA metabarcoding. Then, we performed a Principal Coordinate Analysis (PCoA) using the pcoa function available in ape v.5.7-1 R package with Lingoes correction of negative eigenvalues, as recommended by Legendre and Anderson (1999). The percentage of variation and significance of each PCoA axis were assessed using the pcoa.sig function available in the R package PCPS v1.0.7.

We conducted a distance-based Redundancy Analysis (db-RDA), a statistical tool developed by Legendre and Anderson (1999), to assess the influence of each experimental factor - i.e. time of the experiment, type of the sample, plastic concentration - on community assemblage variation. The following ‘time of the experiment’ and ‘type of the sample’ were transformed into dummy variables using the R package fastDummies v.1.7.3. Collinearity among explanatory variables was assessed before running the models. Both an empty model and a complete model were built to select the significant variables to retain, using the ordistep function (with the direction set to “both”), available in the R package vegan v2.6-4. For the final db-RDA model, significance of each variable and axis was assessed using the anova function with 10 000 permutations. The global adjusted R² value and significance were reported for the final model. All statistical analyses were conducted in R 4.2.3 (R Core Team, 2022).

By considering each type of sample separately (i.e. plastics, water and the giant clams), we conducted a linear regression analysis to test the relationship between the relative abundance of each ASV and the gradient of plastic concentration. From this analysis, we selected only the ASVs showing a significant correlation with the cor.test function (P-value < 0.01) and a high R² (R² > 0.4). These ASVs will be referred to as candidate bacteria for the plastic response. We extracted the genus of each candidate to evaluate the representativity of the genus and to determine if some candidate’s genus were shared among different type of samples. The results were represented using the ggplot2 v.3.4.2 R package. To evaluate the number of ASVs shared among all samples and those unique to each sample, we used the UpSet R v.1.4.0 R package (Conway et al., 2017).

The pairwise distances of the DNA sequences associated with the selected candidate bacteria were calculated using the “K80” model with the dist.dna function available in the ape v.5.7-1 R package. The dendrogram was then constructed using the hclust and as.dendrogram functions from the stats v4.2.2 R package. We used FigTree v1.4.4, a graphical viewer, to visualize the tree (http://tree.bio.ed.ac.uk/software/figtree/).

The overall dataset comprises 104 samples (16, 42 and 46 samples of plastic, water and T. maxima, respectively) with sufficient sequencing depth (> 10 000 reads) ( Supplementary Figure S1 ) including a total of 10 710 bacterial ASVs. To keep only representative ASVs for describing bacterial community, we selected 1 716 ASV present in relatively high abundance (i.e. relative abundance > 0.5) and in at least three out of 104 samples. A Principal Coordinates Analysis (PCoA) was then performed to reflect the overall variation of the bacterial community ( Figure 2 ). On the first significant Principal Coordinate axis (PCo1; 19.62 %), water samples were delineated from Tridacna maxima samples, while on PCo2 (12.38 %) plastic samples were distinct from water samples. Notably, we also observed that T. maxima samples formed two separate groups along PCo1 and PCo2 ( Figure 2 ). The first group of T. maxima individuals, located on the right of PCo1 (PCo1 > 0.4 and PCo2 < 0.00), corresponds to samples with low bacterial diversity and two dominant ASVs (ASV1 and ASV2: both from the Endozoicomonadaceae family) ( Figure 2 , Supplementary Figure S2 ). The second group (PCoA1 < 0.4 and PCo2 > 0.00), located on the left of the PCo1 axis, comprises T. maxima samples with more diverse and evenly distributed bacterial communities ( Figure 2 , Supplementary Figure S2 ). These two groups include giant clams from different plastic concentrations and time points (i.e. T0, T4), indicating that neither plastic concentration nor time explains observed the pattern. Kruskal-Wallis statistical tests on growth rate and muscle glycogen concentration (i.e. proxies for the physiological status of our animals) between these groups were not significant (P > 0.01, Supplementary Figures S3, S4 ). Similarly, the correlation between bacterial diversity and growth rate was non-significant (R² = -0.22, P-value = 0.25) as was the correlation with muscle glycogen concentration (R² = 0.12, P-value = 0.56).

We tested the influence of plastic concentration on a broad scale using a distanced-based Redundancy Analysis (db-RDA) on the six significant PCoA axes ( Supplementary Figure S5 ). The analysis revealed that plastic concentration did not significantly influence the variation in community assemblage (P-value = 0.418). In contrast, and consistent with the PCoA results (i.e. plastics, water column and the giant clams; Figure 2 ), the majority of the variation was significantly influenced by the type of sample (adjusted R² = 51.8 %; P-value < 0.001). We also checked whether the type of plastic used in pearl farms (i.e. ropes and collectors) influenced community variation and found no significant difference in community assemblage between the two types of plastic (P-value = 0.3). To investigate the influence of the temporal scale on the bacterial community, we retained only the water samples, as they were collected at five different time points (i.e. T0, T1, T2, T3, T4). Sampling time significantly influenced bacterial community assemblages (P-value < 0.001), accounting for up to 19.4 % of the global variation. Each sampling time was significant, suggesting regular changes in the bacterial communities over time.

We identified 1 180, 1 251 and 820 ASVs specifically associated with plastics, water, and giant clams, respectively ( Figure 3 ). We emphasized the specificity of the bacterial community associated with each compartment, as at least 30.2 % of the ASVs were unique to one of the compartments. The strongest continuity in community structure was found between water and plastic samples, with 514 shared ASVs, while the lowest continuity was found between Tridacna maxima and water samples, with only 80 shared ASV ( Figure 3 ).

Except for the mesocosm with the highest plastic concentration (4.05 g/L), chemical analyses revealed that the concentrations of the tested chemical components were below the limits of detection (LD) or quantification (LQ) ( Table 2 ). In the mesocosm with 4.05 g of plastic, gas chromatography coupled with tandem mass spectrometry (GC/MSMS) indicated the presence of several pesticides such as Beta-BHC, Delta-BHC, acetochlor, metolachlor, and chlorpyrifos. The sum of pesticides reached 20.69 ng/L and 1.13 ng/L in the first and second replicates, respectively. Nonylphenols, belonging to the alkylphenols class, were also detected in one of the two replicates at a level of 13.55 ng/L of water in the same mesocosm ( Table 2 ).

For each type of sample (i.e. plastic, water, and T. maxima), the linear regression between the relative abundance of an ASV and plastic concentration detected a total of 33 candidates ASVs (R² > 0.4 or R² < -0.4 and P-value < 0.01). Candidate ASVs were also required to be present in at least three samples from the same environment. All the selected candidates had a positive R², revealing a consistently positive relationship between relative abundance and plastic concentration ( Supplementary Figure S6 ). Comparing the bacterial abundances in mesocosms with plastic concentrations ranging from 0 to 1.35 g/L, abundance tended to be similar; however, at 4.05 g/L, the abundance was higher, suggesting that the linear regression is largely driven by this unique extreme point in the distribution, regardless of the sample type ( Supplementary Figure S6 ).

The mean abundances of all candidate bacteria obtained for plastic support, water and T. maxima samples at low plastic concentrations (ranges from 0 to 1.35 g/L) were 0.0007, 0.0018, 0.0001 while at high concentrations (4.05 g/L) these abundances were 0.0063, 0.0069, 0.0019 respectively. This represents 9, 3.8 and 19-fold shift in the abundance of candidate bacteria on plastic, water and T. maxima, respectively, with T. maxima showing the highest shift ( Figure 4 ). These differences in bacterial candidate abundance between concentrations of 0 to 1.35 g/L and the highest concentration at 4.05 g/L of plastic were significant across all three sample types (Wilcoxon test; P-value < 0.01).

Among these 33 candidates ASVs, some are specific to a particular type of sample. Specifically, four candidates ASVs are exclusively present in clams, nine are specific to the seawater compartment, and 18 are specific to the plastic. Notably, two candidates ASVs shared between water and plastic ( Figure 5 ) corresponding to Sulfurospirillum (Class: Campylobacteria) and Filomicrobium (Class: Alphaproteobacteria) ( Table 3 ).

Highlighting the genetic similarity across ASVs candidates reveals that the 33 identified bacterial genera are distributed into five classes, namely Campylobacteria, Bacteroidia, Planctomycetes, Alphaproteobacteria, and Gammaproteobacteria ( Table 3 , Figure 6 ). The Campylobacteria class is represented by a single genus, Sulfurospirillum, while other classes, such as Alphaproteobacteria, are overrepresented, encompassing at least four candidate families of bacteria associated with plastic concentration.

However, the identification of candidate ASVs in one type of sample does not imply their absence in other environments. Table 3 shows that most of the candidate ASVs identified in one environment are also present in other environments without a significant impact on their abundance. In total, 11 ASVs (ASV635, ASV355, ASV905, ASV189, ASV1359, ASV513, ASV44, ASV703, ASV914, ASV730, ASV620) are shared among the three compartments (i.e., plastic support, water, and T. maxima). Additionally, 15 candidate ASVs from plastic substrate or water are shared between these two same environments, further emphasizing the bacterial continuity between them. Only five candidate ASVs are present exclusively in a single environment, namely ASV1045 (T. maxima), ASV492 (T. maxima), ASV1661 (water), ASV190 (water) and ASV1012 (plastic).