2.1 Experimental design and Z. marina sampling

The Kiel Outdoor Benthocosms (KOB), situated on an aluminum float (ALU-BAU, Büdelsdorf, Germany), moored to the dock of GEOMAR Helmholtz Centre for Ocean Research in the inner Kiel Fjord, was used for this experiment from November 25, 2015, to August 26, 2016. The mooring was made up of sledges with wheels that could slide along rails fastened to the pier's supports in order to accommodate for changes in sea level. Six 3000 L benthocosm tanks (Heers & Brockstedt Umwelttechnik, D-24539 Büdelsdorf) with internal dimensions of 2 m × 2 m × 0.9 m and a 5 cm thick styrofoam layer between inner and outer walls, to achieve the best thermal insulation, were housed inside the KOB. These six tanks, each of which divided into two subunits comprising 12 experimental units (1500 L), were labeled A through F and ran in a factorial “ANOVA mode” with present vs. future climatic scenarios (3–6 replicates). Surface seawater from the Kiel Fjord flows through continuously at a rate of 1800 L per day, providing nutrients, organic matter, and daily variations in abiotic parameters like temperature, salinity, pH, and oxygen that are similar to those found in nature. The system includes a wave generator that prevents stratification by causing water movement and it has an open top for natural light supply. A GHL feedback system (GHL Advanced Technology) was specifically installed in each tank, and it worked in tandem with heaters and coolers to maintain the desired temperatures. Wahl et al. (2015) provide complete details on the benthocosms materials, conditions, control, and registering elements (temperature, salinity, oxygen, etc.). The experimental community was created in accordance with the natural shallow-water Baltic Sea system's community structure, which is dominated by Fucus vesiculosus beds and Z. marina meadows. Therefore, the tanks were set up to include 14 common organisms found in such ecosystems, which, among others, included the gastropod Littorina littorea, the isopod Idotea balthica, amphipods of the genus Gammarus, the decapod Palaemon sp., and the bivalve Mytilus edulis. The input of fjord water allowed other organisms to enter the tanks as well. Those were mostly filamentous algae, polychaetes, a few different species of bivalves (including Limecola balthica, Mya sp., and Mysella sp.), and gastropods (like Hydrobia ulvae and Retusa sp.), and the round goby (Neogobius melanostomus). See Sawall et al. (2021) for more details.

Zostera marina was sampled in November 2015 in the Kiel Fjord (Falckenstein: 54°24′24.7″ N, 10°11′38.7″ E), Baltic Sea, along a 100 m transect at 2–3 m depth within a meadow. Around 1–3 shoots (leaves, rhizomes and roots) per ramet were collected and kept in large coolers with fjord water until transported to the KOB within 2 h after collection. Six shoots of similar size (shoot length: 6–10 cm; rhizome length: 3–5 cm), from different plants, were planted, in sediment previously collected from the Fjord, in 5 L boxes. Every tank received 18 of those boxes with six plants each and 10 beakers with two plants each which resulted in a total of 134 plants. During an acclimation period of 7 days, all the shoots were maintained under ambient fjord temperature.

For this particular experiment, two different temperature regimes were applied, with each treatment replicated in six independent tanks: (i) a control regime without simulated temperature rising, but natural stochastic temperature variability (tanks B1, B2, D1, D2, F1, F2 = Ambient); (b) a warming regime, chosen considering a future warming scenario for the Baltic Sea (Gattuso et al., 2014; Gräwe et al., 2013) (tanks A1, A2, C1, C2, E1, E2 = Warm). The temperature regime was set to an increase in around 3.6°C above naturally fluctuating fjord temperatures (Ambient). The target temperature of the Warming treatment was approached by daily temperature increments of 0.3°C over 12 consecutive days, in three different periods. Temperature details for this experiment are represented in figure 1 of Sawall et al. (2021). The number of shoots in each tank was registered a few days before each sampling time.

In order to study, the bacterial community associated with Z. marina roots and adjacent sediment, and compare it between ambient and warming treatments, six replicates (one sample per tank) were collected from the ambient and warming treatments. The sampling was done in four different timings (Mach, April, May, and June) which corresponded to 4 different ambient and warming temperatures (Figure 1). Sediment and plant samples were also collected from the natural Z. marina meadow in the Falckenstein Fjord, in the beginning and end of experiment, to be used as control (Figure 1). All the roots collected for this experiment were taken from living and healthy Z. marina shoots.

Seagrass roots were excised from the plant with sterilized blades and the adjacent sediment was scooped using a 2 mL tube. Samples were stored using the DNA/RNA shield from Zymo Research and stored at −20°C until extraction.

2.2 Abiotic parameters monitoring

Dissolved oxygen was measured continually and logged every 60 s by a Hach-Lange LDO optode in each of the experimental tanks (A, B, C, D, and E) (Hach-Lange, 4H.Jena Engineering GmbH). A GHL Profilux 3ex was used to measure and regulate the temperature, and data were logged every 10 min (the same was done in the Fjord). Around 3 h after daylight, salinity was manually measured using a WTW multimeter (Multi WTW Cond 3110 1 Tetra Con 325). 20 mL of water were taken every month (from December to April) or every 2 months (from April to August) for inorganic nutrient analysis (PO₄, NH₄, NO₂, NO₃). This water was filtered using a 0.45 μm cellulose-acetate membrane filter (Sartorius) and the filters were stored at −20°C until further analysis using an autoanalyzer (Seal Analytical) following basic protocols (Lewandowska et al., 2014).

2.3 DNA extraction and bacterial 16S amplicon sequencing

DNA was extracted with the DNeasy PowerSoil Kit (Qiagen, Benelux BV) following the Vortex Adapter protocol. Extracted DNA was sent to Molecular Research (MR DNA, Shallowater, TX, USA). An Illumina library was prepared by amplifying the 16S rRNA gene V3V4 variable region, using the forward 341F (5′-CCTACGGGNGGCWGCAG-3′) and reverse 805R (5′-GACTACHVGGGTATCTAATCC-3′) primers (Klindworth et al., 2013) and sequenced on an Illumina MiSeq system (2 × 250 bp paired-end) following the manufacturer's guidelines. PCR conditions were as follows: 95°C for 3 min, 25 cycles of 95°C for 30 s, 55°C for 30 s, 72°C for 30 s, and a final elongation of 72°C for 5 min. The samples were pooled in equal proportions based on their molecular weight (calculated based on the size of the amplicon) and DNA concentrations (using Qubit Invitrogen) and purified using calibrated Agencourt AMPure XP beads. DNA libraries were prepared by following Illumina MiSeq DNA library preparation protocol and paired-end (2 bp 250 bp) sequencing performed at MR DNA (Shallowater, TX, United States) on an Illumina MiSeq system following the manufacturer's guidelines.

2.4 Bacterial community sequence analysis

Demultiplexed Illumina raw paired-end sequences were imported into the MR DNA software FASTq Processor where barcodes and both primers were removed, and the fastq files were prepared to be used in the Quantitative Insights into Microbial Ecology 2 (Qiime2 v.2021.4; Bolyen et al., 2019) package for complete microbiome analysis. The resulting files were imported into QIIME2 using the Casava 1.8 paired-end demultiplexed fastq tutorial and were then quality filtered, trimmed according to the quality of demultiplexed sequence reads (forward: trimmed left at 0 sequence base and truncated at position 228rd; reverse: trimmed left at 0 sequence base and truncated at position 220th), dereplicated, denoised, merged, and assessed for chimeras to produce ASVs using the DADA2 plugin (Callahan et al., 2016). Taxonomy was assigned to ASVs using a pretrained naive Bayes classifier. The classifier was trained on Silva 138 99% OTUs full-length dataset using the QIIME2 feature classifier plugin (Bokulich et al., 2018). The feature table was then filtered to exclude ASVs assigned to chloroplasts, mitochondria and singletons and doubletons. For statistical purposes, the ASV table was rarefied to the minimum number of sequences (3619). We used rarefied data to perform PERMANOVAs and multivariate analysis, and nonrarefied data to perform any differential/discriminant analyses and make relative abundance bar plots.

The microbial diversity and richness of all samples were estimated using alpha diversity metrics—Shannon index and observed ASVs—using the diversity plugin within QIIME2. Boxplots were generated for a visualization of the alpha diversity differences among groups of samples (grouped according to type of sample, treatment, and sampling time—detailed below). Univariate PERMANOVA was used to determine significant differences (as in Hartmann et al., 2015) among groups of samples. Matrices were constructed using Euclidean distance, and the number of permutations was set to 999. Monte Carlo tests were done when the number of unique permutations was less than 400 (Metropolis & Ulam, 1949).

Bacterial community composition and structure (beta diversity) were assessed through ASV square root transformations, after which a resemblance matrix was compiled using the Bray–Curtis dissimilarity algorithm. Statistical analyses were done using a three-factor PERMANOVA: Type of Sample (fixed factor with two levels—roots and sediment), Temperature treatment (fixed factor with three levels—ambient, warm and control [fjord]), and sampling time (fixed factor with four levels—1, 2, 3, and 4). All the factors and respective interactions were tested. Monte Carlo test was done when the number of unique permutations was less than 400.

Community differentiations were visualized with canonical analysis of principal coordinates (CAP), using the interaction of all three factors (Type of Sample × Treatment × Sampling Time) as a priori factor. For a better visualization of the differences among samples, the control samples (Fjord) were not included in the CAP (but were included in the PERMANOVA statistical tests). Abiotic factors were overlaid as vectors in the CAP plot to show which were the main drivers of the clustering pattern. All the statistical tests and multivariate analysis were performed in PRIMER-E C PERMANOVA v.6 (Clarke & Gorley, 2006).

While using the total bacterial community to produce the relative abundance bar plots, we noticed that some of the most abundant taxa from a given treatment (considering the interaction of the three factors Type of Sample × Temperature Treatment × Sampling Time) were only present in one or a few replicates and were not representative of the complete set of replicates. That would lead us to biased results that would not reflect the consensus community of that certain treatment. Core community usually refers to any group of microbial taxa shared by a certain number of samples from a host species, environment of interest, or treatment (Neu et al., 2021; Risely, 2020). Therefore, we have chosen the core strategy to characterize the bacterial community of each treatment, maintaining only those ASVs that were present in at least 80% of the replicates from that treatment. That way, we would avoid drawing inferences based on abundant but unrepresentative taxa. The core community of each group/treatment of samples (e.g., Roots/Ambient/Sampling time 1) was calculated using the core features function in Qiime 2 (Bolyen et al., 2019) and using 0.80 as the minimum fraction. After each calculation per treatment, all the “core tables” were merged. This “core” dataset was used for bacterial community composition analysis through relative abundance barplots and any comparison and classification analysis (i.e., LEfSe, including for the PiCRUSt analysis).

It is important to highlight that this dataset we called “core” is not the same as the so called “core community” that are constant across seasons, environmental stress, geographic location, etc., for a certain species. Instead, this “core” dataset identifies the main changes from one treatment to another for all or most of the replicates (making it consistent).

Relative abundance barplots were made to visualize the differences in taxa (Family level) between treatments. To assess which taxa strongly discriminate between Ambient and Warm treatments, per sampling point (roots and sediments separately) and to have a clearer idea of the persistently high temperatures effect on the microbiome, LEfSe analysis (Linear discriminant analysis Effect Size) were performed using the online tool Microbiome Analyst (https://www.microbiomeanalyst.ca/; Chong et al., 2020; Dhariwal et al., 2017). LEfSe employs Kruskal–Wallis rank sum test to detect features with significant differential abundance, followed by Linear Discriminant Analysis to evaluate the relevance or effect size of differential abundant features. We set the p-value cutoff to 0.1, using the FDR-adjusted data and 2.0 Log LDA score. Similarly, LEfSe analysis was done to access the main differences between the control/fjord samples and the corresponding (time 1 and 4) ambient sediments and roots.

PERMANOVA analysis and visualization of microbial community structure among distinct treatments (CAP) were done with both datasets (total community and core community) to check discrepancies, and the results were fundamentally the same.

2.5 Predictive functional analysis

Core ASVs table was analyzed using PiCRUSt2 (Phylogenetic Investigation of Communities by Reconstruction of Observed States; Douglas et al., 2020). A cutoff nearest-sequenced taxon index value of 0.15 was employed to remove unreliable predictions, as recommended by Langille et al. (2013). The metabolic functions of these pathways were referenced against the MetaCyc database (Caspi et al., 2018) for description.

The predicted pathway abundances were statistically analyzed, for roots and sediment separately, through a two-factor PERMANOVA: Temperature treatment and sampling time (and their interaction). The parameters used were the same as for the ASV abundance and bacterial community composition (see above). Also, as done for the community composition analysis (see above), CAP plots were prepared to visualize samples' clustering following the abundance of each metabolic pathway predicted for each sample.

To assess functional differentiation among the different treatments (per type of sample), LEfSe was performed, using the online tool Microbiome Analyst (https://www.microbiomeanalyst.ca/; Chong et al., 2020; Dhariwal et al., 2017) using the same parameters as for the bacterial community analysis (see above).

3.1 Bacterial community diversity and structure

To assess bacterial community diversity and evenness (alpha diversity), we calculated the Shannon diversity index and the number of observed ASVs for each sample. Overall, sediments had a consistently higher diversity than root communities regardless of the factor considered (Figure 3a,b). Differences between the two sample types ranged from a to b. Within root communities, both diversity indices showed no differences among the different sampling times (1, 2, 3, and 4) within each temperature setting/treatment (Ambient, Warming, and Fjord [Control]; Figure 3a). Similarly, no differences were found among the different temperature treatments for the same sampling time (Figure 3a). For the sediments, the Shannon index did not reveal any difference among samples (Figure 3a). However, the number of ASVs within the Ambient setting was higher for samples collected during sampling time 4 compared to sampling time 2 (Figure 3b). In the Warm treatment, samples from the sampling times 2 and 3 exhibited a lower diversity as compared to sampling time 4 (Figure 3b). Within the Fjord, no diversity differences were found among the sampling times.

The increase in seawater temperature affected the structure of microbial communities depending on both sample type (sediment or root) as well as sampling time. Among the single factors, the sample type (p = .001, Table 1, Table S1, Figure 4) accounted for most variation among samples (Table 1). Temperature treatment and sampling time as single factors contributed equally to community structure by around 10% (Table 1).

When comparing the different temperature regimes (Ambient, Warm, and Fjord control) for each sampling time (roots and sediments separately), bacterial communities were always statistically different, except when comparing those associated with roots sampled in sampling time 1 and sediments sampled in the first and second dates (Table S2a,b).

As for the bacterial community changes across the different sampling times, for each temperature treatment, the microbiome associated with roots and sediments at ambient temperature did not significantly change from sampling time 1 to sampling time 2 (Table S3a). However, at warm treatment, only the microbiome associated with the roots sampled in sampling time 4 was different from the remaining (Table S3b) but for the sediments, as before, bacterial communities were always statistically different except between sampling 1 and 2 (Table S3b). The samples collected in the Fjord have also shown differences in the bacterial community associated with roots and sediments sampled in the two different sampling times (Table S3c).

The PERMANOVA outcomes that are shown in the CAP display a clear and strong separation between roots and sediments (Figure 4) but, in addition we can see a separation by sampling point reflecting a separation driven by actual temperature (Figure 4), which is especially clear in the sediments.

The abiotic factors measured in this study were overlayed in our multivariate analysis, showing that temperature and oxygen had the strongest correlation to the multivariate patterns separating bacterial communities at different temperatures (Figure 4).

3.2 Microbial community composition and main shifts across temperature treatments and temperature gradients

To ensure a reliable analysis of bacterial community composition for each treatment, we removed the ASVs that were not present in at least 80% of the replicates.

3.2.1 Bacterial community associated with Z. marina roots

Overall, the bacterial communities associated with Z. marina roots were distinct from those at the collection site (Fjord). However, samples from the Ambient treatment at sampling time 1 and 4 share a higher number of taxa with the correspondent Fjord samples (for sampling time 1 no discriminant features were found between these two treatments), including two of the most abundant taxa (Spirochaetaceae and Gammaproteobacteria for sampling time 1 and Gammaproteobacteria and Sulfurimonadaceae for sampling time 4; Figure 5a,c). For the fjord roots, sampled at the end of the experiment, Arcobacteraceae and Bacteroidetes BD2-2 were found as discriminant when comparing to ambient samples from that same period (Figure S1a).

In the Ambient treatment, there was a decrease in the number of characterizing taxa at the sampling time 2, Flavobacteriales and Gamaproteobacteria dominated the total taxa relative abundance (Figure 5a). However, by sampling time 3 the bacterial community reverted to a bacterial community more similar to sampling time 1, sharing three of the most abundant taxa (Spirochaetaceae, Gammaproteobacteria, and Bacteroidetes BD2-2; Figure 5a). At sampling time 4, the Gammaproteobacteria was still the most abundant taxon, but there was also an increase of members from the Sulfurimonadaceae and Desulfocapsaceae families, which were within the three most abundant for this treatment (Figure 5a).

In the Warm treatment, there was an increase in the Spirochaetaceae (the most abundant taxa for the three first sampling times), along with temperature (Figure 5b). At the highest temperature (sampling point 4) the most abundant taxa were identified as Marinilabiliaceae. In contrast to the ambient treatment, at least one taxon related to the sulfur cycle was among the most abundant across the experiment in the Warm treatment. In the warmest sampling time, two, out of four, different families of sulfate-reducing bacteria (Desulfocapsaceae and Desulfobacteraceae; Figure 5b) were among the most abundant.

When comparing corresponding sampling times between the two temperature treatments (Ambient and Warm), we found that the higher the temperature the greater the number of biomarkers/discriminant bacteria between the homologous sampling times (Figure 5d–g). At sampling times 1, Gammaproteobacteria was the discriminant taxa for the Ambient treatment while Sulfurimonas was the discriminant taxa for the Warm treatment (Figure 5d). Two different Spirochaeta species were discriminant for the second sampling in the Warm treatment, while Flavobacteria and Gammaproteobacteria are more abundant in the Ambient treatment (Figure 5e). Two ASVs identified as Gammaproteobacteria were found to be discriminant for the third sampling at Ambient treatment, along with Bacteroidetes BD2-2 and Cohaesibacter, but the Warm treatment was still represented by a Sprirochaeta (Figure 5f). In the last sampling time, Sulfurimonas and Gammaproteobacteria were significantly more abundant at the Ambient temperature while the Warm treatment was discriminated by two ASVs identified as Marinilabiliaceae and a Desulfobacteraceae (Figure 5g).

3.2.2 Bacterial community associated with bulk sediments

The sediment-associated microbiome was more stable compared with the root-associated community, and this was observed across all three treatments (Fjord/Control, Ambient, and Warm; Figure 6a–c). Although sharing the most abundant taxa, the ambient sediments collected in the first sampling day showed a lower number of taxa when compared to the first fjord sediments (Figure 6a,c). Also, winter fjord samples showed a higher number of discriminant taxa (e.g., Actinomarinales, Bacteroidales SB-5, Thiogranum, Lutimonas; Figure S1b). Comparing the latest sampling time, the major taxonomic groups changed and only Flavobacteriaceae was shared among the most abundant taxa for ambient and fjord sediments (Figure 6a,c). Actinomarinales, Chromatiaceae, Bacteroidetes VC2.1 Bac.22, Thiomicrospiraceae, and Eudoraea were within the discriminant taxa for fjord sediments and Microscillaceae, Oscilatoriaceae, and Candidatus Thiobius differentiate the ambient samples (Figure S1c).

The microbiomes of the Ambient temperature treatment shared the same four most abundant taxa (Chromatiaceae, Desulfosarcinaceae, Sandarinaceae, and Flavobacteriaceae; Figure 6a), until the fourth sampling when a major shift in the microbiome occurs, and the most abundant taxa are members of the Gammaproteobacteria, Flavobacteriaceae, Microscillaceae, and Oscillatoriaceae (Figure 6a).

The Warm treatment presented the same most abundant taxa as the ambient treatment in the first sampling time and in the second sampling only the Desulfosarcinaceae were replaced by the Defluviicoccaceae (Figure 6b). In the third sampling point, there was a strong shift in the bacterial community associated with the sediments, where the most abundant taxa were Flavobacteriaceae, Gammaproteobacteria, Microscillaceae, and Cryomorphaceae (Figure 6b). In the last sampling, the bacterial community apparently reverted to the initial most abundant taxa (Chromatiaceae, Desulfosarcinaceae, Sandarinaceae, and Flavobacteriaceae).

Ambient and Warm treatments had significant bacterial biomarkers/discriminants for sampling times 3 and 4 (Figure 6d,e). For sampling time 3 in the Ambient treatment, 17 taxa biomarkers were identified, including several genera belonging to less abundant families that were mostly found among Ambient temperature replicates (e.g., Halioglobus, Arenimonas, Ilumatobacter, and Luteolibacter) (Figure 6d). Within the discriminant taxa were also bacteria from the Sandaracinaceae, Desulfosarcinaceae, and Flavobacteriaceae families (Figure 6d). For the Warm treatment, 7 biomarkers taxa were identified, and those with higher LDA scores belonged to the Gammaproteobacteria and Microscillaceae family (Figure 6d). For the sampling time 4, 20 Ambient biomarkers were found and the ones with the highest LDA score belonged to Gammaproteobacteria and the Oscillatoriaceae, Rhodobacteraceae, Microscillaceae families (Figure 6e). In sediments sampled at the highest temperature, 39 taxa were detected as discriminants, including a great proportion of taxa from the Desulfosarcinaceae, Chromatiaceae, Sandarinaceae, and Flavobacteriaceae families (Figure 6e) as well as several ASVs within less abundant families not found in the Ambient treatment (e.g., Thermoanaerobaculaceae).

3.3 Analysis of the predicted functional profiles—PiCRUSt

We conducted the PERMANOVA analyses separately for each type of sample (sediment and root) due to the strong signal between roots and sediments (p < .001) that would mask the differences among the other factors, the Fjord samples were not included in these statistics as we were mainly interested in the differences among the different treatments.

For both roots and sediments, a significantly different putative functional profile was found among the different sampling times, irrespective of temperature treatment (p = .031//.008, Table 2, Figure 7). The functional profiles were mostly clustered based on the effective temperature at which samples were collected. For the roots, a major cluster with samples from sampling times 3 and 4 (A and W treatments) while the earliest sampling times are more dispersed. However, we only found statistical differences between sampling 3 and the remaining (see Table S4) which would be mostly due to the ambient samples that cluster apart in the CAP plot (Figure 7a). For the sediments, the clustering is more evident for sampling times 1 and 2 (A and W treatments) which did not differ among each other (Figure 7b, p = .407, Table S4). The other cluster represents samples collected at temperatures higher than 15.6°C where only sampling time 3 statistically differs from sampling 2 and 4 (Figure 7b, Table S4).

To identify potential key functions that would allow Z. marina to thrive under higher temperatures, we have only presented those functions that were found significantly more abundant at higher temperatures (considering the clustering above—Figure 7—we have considered sampling times 3 and 4, from both temperature setting, as the warmest treatments) (Figure 8a,b).

For root-associated microbial communities, 11 biomarkers were consistently and significantly more abundant in warmer treatments (temperatures ≥15.6°; Figure 8a). These were: Superpathway of sulfur oxidation, superpathway of menaquinol (−7, 8. 11, 12, 13) biosynthesis, superpathway of demethylmenaquinol-8 biosynthesis, 2-methylcitrate cycle I, Isoprene biosynthesis II, superpathway of phylloquinol biosynthesis and 1,4-dihydroxy-2-naphtoate biosynthesis I.

For the microbial communities of sediments, 13 biomarkers highly abundant among higher temperatures were found, which were: Superpathway of l-alanine biosynthesis, pyridoxal-5 phosphate biosynthesis I, nitrate reduction VI (assimilatory) and I (denitrification), dTDP-N-acetylthomosamine biosynthesis, urea cycle, S-adenosyl-l-methionine cycle I, 2-amino-3-carboxymuconate semialdehyde degradation to 2-oxopentenoate, 2-nitrobenzoate degradation I, heterolactic fermentation, l-methionine salvage cycle III, vitamin E biosynthesis (tocopherois), and S-methyl-5-thio-alpha-d-ribose 1-phosphate degradation (Figure 8b).

4.1 Bacterial communities associated with Z. marina roots

4.2 Bacterial communities associated with bulk sediments

4.3 Predicted functional profiles differentiation between ambient and warm treatments

The overall analysis of predicted functions for roots and sediments microbiomes showed, in contrast to the microbial community structure, minimal differentiation among treatments. This may be due to different bacterial taxa performing similar functions throughout the experiment (Chen et al., 2022). However, specific differences were observed in the functional profiles of the warmest months (sampling points 3 and 4 for both ambient and warm treatments). The superpathway of sulfur oxidation and isoprene biosynthesis were identified as the primary biomarkers for the roots sampled during the warmest months. The former suggests increased sulfur cycle activity in warmer treatments, consistent with the high abundance of sulfur cycling-related bacteria in the core community structure, most of them thermophilic or moderate thermophilic (Nakagawa & Takai, 2008). Furthermore, as previously discussed, elevated SST can lead to the die-off of sensitive marine organisms, including seaweeds and seagrasses, resulting in increased organic matter availability and stimulation of sulfur-related bacteria (Duarte et al., 2018; Rolando et al., 2022).

The latter, isoprene biosynthesis II, is, in some terrestrial plants, linked to the production of this volatile organic compound that protect against thermal shock and oxidative stress (Medori et al., 2012). These compounds can be produced by bacteria to enhance resistance in their host plants (McGenity et al., 2018). In addition, several pathways related to menaquinone biosynthesis were increased in warmer treatments. Menaquinone is a vital component of electron transport chains in many microorganisms and may be involved in energy production (Dairi, 2012). This suggests that the microbiome of Z. marina roots increased its activity in response to an environmental stressor such as warming temperatures, enabling a rapid response (Sarmento et al., 2010).

The pathways that better differentiated the warmer treatments in sediments were nitrate reduction (both assimilatory and denitrification) and heterolactic fermentation. Maximum activity of nitrogen removal processes occurs during warmer months in the Baltic Sea, when sediment organic content is higher (Jäntti et al., 2011). Another study reported a 50% increase in nitrate uptake by Z. marina under extreme warming scenarios (Alexandre et al., 2020; Kaldy, 2014), indicating that this mechanism plays a critical role in eliminating excess nutrients from water and helping plants adapt to chronic warming (Dai et al., 2020; Qu et al., 2022). As previously discussed, Z. marina die-off increased organic matter in the sediments, leading to increased decomposition and fermentative processes.

4.4 Conclusions

This study suggests that the effects of temperature emerged earlier in the superimposed warming treatment compared with the natural seasonal warming. The main differences are most likely due to premature Z. marina die-offs under chronic warming conditions, resulting in an earlier increase in organic matter-degrading bacteria and a higher ratio of sulfate-reducing bacteria compared with sulfide oxidizers. Also, the differentiating predicted pathways were mostly related to biogeochemical cycles and possible production of hydrocarbons that would protect seagrass from temperature stress.

These structural and compositional variations of the associated microbiome can provide insights into the seagrass ecological status, and some taxa/genes/pathways may be used as markers for specific stresses. Conte et al. (2021) proposed that taxa engaged in crucial biogeochemical transformations, particularly those related to nitrogen, sulfur, and carbon cycles, would be the most effective indicators. Any changes in these markers can impact the ecology and physiology of the seagrass host. To further understand these fluctuations in the seagrass holobiont under the influence of climatic stressors, controlled laboratory research, such as in the present study, is essential. Seagrass monitoring and conservation programs should integrate results from such research to identify early warning indicators of the seagrass physiological state and prevent the extinction of vulnerable populations. Additionally, a better understanding of how the microbiome can change under the influence of certain stresses is essential to investigate specific taxa that might function as probiotics and be used to prevent the major effects of climate change or restore seagrass ecosystems.