Seawater samples were collected during the Bohaike Research Vessel cruise from August 17 to 22 in 2023, which investigated the waterways extending outward from the southern CRE trough to the East China Sea shelf (122.15°E–123.08°E, 29.62°N–31.02°N, Supplementary Table S2). Based on the situation of SF (121.9°E–122.5°E) and PF (122.5°E–123.5°E) delineated from multi-year observational data in and off the CRE during the summers of 1988–2016 by JianZhong Ge (Li W. et al., 2021), we set up a total of thirteen stations across the CDF system (Supplementary Table S2). In the study area, four stations are set up on the nearshore water mass of SF, six stations on the offshore water masses of PF, and three stations between SF and PF. A total of 66 seawater samples were collected from three layers of the water columns (Figure 1). The surface samples were collected at a depth of 0–2 m beneath the sea surface. The samples collected at a depth of 3–4 m above the seabed were considered as the bottom samples. The seawater sample of the middle layer was taken only when the station depth exceeded 30 m. Briefly, all seawater samples were collected using a conductivity-temperature-depth (CTD) rosette sampler (Sea-Bird Electronics, SBE 32, USA) equipped with 8 L Niskin bottles. Meanwhile, in each depth profile, environmental parameters such as salinity, temperature, turbidity, pH and dissolved oxygen (DO) were monitored in situ using EXO3 Multiparameter Sonde (YSI, Yellow Springs, Ohio, USA). We identified the cluster patterns of the SF, BE, and PF water masses and reassembled the stations among the surface, middle, and bottom water layers (Figure 1), based on turbidity, salinity, and the other optimal environmental parameters (Supplementary Table S3). Dissolved inorganic nitrogen (DIN: NO3- + NO2- + NH4+), dissolved inorganic phosphate (DIP), and dissolved silicate (DSi) were analyzed using a SmartChem automated discrete chemistry analyzer (SmartChem200, Alliance, France) at the Third Institute of Oceanography, Ministry of Natural Resources (TIO). At each water layer, a total of 2 L of seawater was collected in duplicate and then filtered immediately for the collection of microbial cells. Filtration was performed sequentially through 3 μm and then 0.22 μm Millipore polycarbonate filters (Millipore Corporation, Billerica, MA, USA) to obtain PA and FL microorganisms for high-throughput sequencing, respectively.

To extract total microbial genomic DNA from both 3 μm and 0.22 μm filters, we utilized the DNeasy^Ⓡ PowerWater^Ⓡ kit (Qiagen, Germany) following the manufacturer instructions. The DNA quality was detected through 1.6% agarose gel electrophoresis, and its concentration was quantified using a NanoDrop2000 spectrophotometer (Thermo Scientific, United States). The amplified DNA library preparation was conducted using the NEXTFLEX Rapid DNA-Seq Kit for Illumina^Ⓡ (Bioo Scientific, USA) following the standard protocol. The ~390 base pairs (bp) in the hypervariable V4 region of the prokaryotic 16S rRNA gene were amplified with a universal primer pair 515F (5′-GTG YCA GCM GCC GCG GTA A-3′) and 806R (5′-GGA CTA CNV GGG TWT CTA AT-3′) by a T100 Thermal Cycler PCR thermocycler (BIO-RAD, USA). Each PCR reaction mixture was programmed to a 20 μL system, including 10 μL of 2 × Pro Taq Master Mix, 0.8 μL of each forward and reverse primer (5 μM), a volume containing 10 ng of template DNA, and nuclease-free water to a final volume of 20 μL. The PCR amplification cycling procedure was comprised of initial pre-denaturation at 94°C for 5 min followed by 27 cycles of denaturing at 95°C for 30 s, annealing at 55°C for 30 s, extension at 72°C for 45 s, single extension at 72°C for 10 min and end at 4°C. The PCR products were recovered using a 2% agarose gel and later purified using a DNA gel recovery kit (PCR Clean-Up Kit, China). The purified amplicons were pooled in equimolar concentrations and subjected to paired-end sequencing on a PE300 Illumina Nextseq2000 platform (Majorbio Bio-Pharm Technology Co. Ltd., Shanghai, China).

The raw sequences data of 16S rRNA genes were processed using QIIME2 (version 2022.2) for bioinformatic analysis (Bolyen et al., 2019). We used Cutadapt (version 4.3) to demultiplex and remove both adapt and primer sequences from raw paired-end reads (Martin, 2011). The DADA2 (version 1.18.0) plugin (q2-dada2) in QIIME2 pipeline was employed with recommended parameters to successively filter, denoise, merge, and then eliminate singletons and chimeras (Callahan et al., 2016). Reads with low-quality scores (below 20%) and those shorter than 270 bp were filtered to generate Amplicon Sequence Variants (ASVs) feature tables and representative sequences. Each representative sequence was taxonomically classified using a trained Naive Baye Classifier (Quast et al., 2013; Ii et al., 2021). The classifier was trained on 515F/806R amplified fragments using the Sliva 16S rRNA gene database (version 138.1). For downstream analyses, we excluded ASVs identified as eukaryotes, mitochondria, and chloroplast. To address inconsistent sequencing depth, the ASVs table was rarefied at a minimum threshold based on the lowest read number (21,056) using QIIME2 (q2-feature table plugin). To visualize the community composition of microorganisms with different lifestyles, we selected the taxonomically annotated results in relative abundance > 1 % at both phylum and family taxonomic levels.

All data analysis and statistical analysis were conducted and visualized with R Studio software (R Core team, 2024).

To characterize lifestyle preferences, we calculated the Particle-Associated Niche index (PAN index) based on the weighted abundance average of ASVs from PA and FL prokaryotes (Salazar et al., 2015; Mestre et al., 2018; Ma et al., 2024). The PAN index ranged from 0 to 1. When values approaching 0, it indicated a free-living lifestyle, values close to 1 demonstrated a particle-associated lifestyle, and values around 0.5 reflected a transition between PA and FL lifestyles.

We stimulated the neutral community model (NCM) using the MicEco (version 0.9.19) package in R to quantify deterministic and stochastic processes in PA and FL prokaryotic assembly process (Sloan et al., 2006). To partially elucidate the assembly mechanisms in deterministic and stochastic processes, we computed a null model analysis to clarify the specific ecological mechanisms using the iCAMP package (Zhou and Ning, 2017; Ning et al., 2019) in R. The ecological processes included heterogeneous selection (HeS), homogeneous selection (HoS), homogeneous dispersal (HD), dispersal limitation (DL), and drift (DR). The observed ASVs derived from the same ancestor after truncating the phylogenetic tree at a certain phylogenetic distance (as short as necessary) to the root point were grouped into the same group bin. In accordance with the phylogenetic bins, we identified the assembly processes driving the turnovers of each bin (Ning et al., 2020, 2024). The most appropriate phylogenetic tree models for bin clusters were modeled using the optimal algorithms in MEGA (version 11) (Tamura et al., 2021). Eventually, the conducted phylogenetic trees were visualized individually using iTOL (Interactive Tree Of Lifel) for embellishment, annotation and management (Letunic and Bork, 2024).

The frontal effects within the dual fronts system combined with secondary circulation collectively shape the spatial distribution patterns of environmental parameters across the SF, BE, and PF water masses. The drastically decreasing suspended particulate concentration progressively improved light conditions on the offshore side of the sediment front, exhibiting as low turbidity (Supplementary Table S3). From the nearshore SF water mass toward the offshore PF water mass, nutrients (e.g., DIN and DSi) and temperature steadily declined with increasing salinity and pH. A distinct water mass characterized by high salinity, low temperature, and pH developed on the offshore side of the plume front (Supplementary Table S3). Meanwhile, the physicochemical characteristics of the BE water mass exhibit an intermediate between the SF and PF water masses (Supplementary Table S3). Additionally, the relative P-limitation and relative Si-limitation widely existed across three water masses (Table 1). Within the region influenced by CDW, the SF water mass was enriched with sufficient nutrients (DIN: 27.02 ± 12.06 μM, DSi: 28.08 ± 8.55 μM, DIP: 0.28 ± 0.03 μM). Neither relative nor absolute N-limitation was observed except for a large proportion of absolute P-limitation (50%, Table 1) in the SF water mass. Moreover, we found a portion of 14.29% stations experienced the absolute Si-limitation in the BE water masses. The PF water mass displayed relatively high nutrient levels, such as DIN (9.67–11.37 μM) in the surface water at the YEC8 station and DSi (9.99 μM) at the YEC9 station. Nevertheless, the PF water mass still experienced nutritional stress effects (Table 1).

We obtained a high-quality dataset comprising 1,075,205 sequences from PA and FL water samples. After denoising and quality control, paired-end joining, filtering chimeras, and removing mitochondria and chloroplast, a total of 2,602 ASVs were generated. The sequencing data achieved a coverage of 99.99 to 100% for prokaryotic species in all samples (Supplementary Table S1), which meant we had effectively captured the prokaryotic community diversity.

Across all water masses, the PA prokaryotic community maintained relatively higher richness and evenness than the FL prokaryotic community in the same water mass reflected by Shannon, phylogenetic diversity index (Faith_PD) and Evenness index (Figure 2). The statistical comparisons in the Shannon and Faith_PD index displayed pronounced differences between PA and FL prokaryotic community structures across three water masses (P ≤ 0.001), except the Shannon index in the SF water mass (Figure 2). For instance, the PA prokaryotes consistently maintained an elevated Shannon index (9.5–11.35) relative to the FL prokaryotes (8.8–11). The mean Shannon indices for the SF, BE, and PF water masses were 10.166, 10.446, and 10.084, respectively (Supplementary Table S5), indicating a relatively high biodiversity of the BE water mass. Notably, communities in the PF water mass manifested the constrained Shannon index (PA: 9.574–10.033; FL: 8.994–10.265). The Faith_PD further validated statistically distinct differences in communities between the BE and PF water masses (P ≤ 0.05, Figure 2). No significant differences in community evenness were detected across three water masses (Figure 2). The PA community in the BE water mass displayed marginally higher evenness compared to that of the PF water mass (P ≤ 0.05).

NMDS analysis was performed to elucidate prokaryotic community structure based on an unweighted-Unifrac distance matrix, stratified by lifestyle preferences and position in the frontal zone (Figure 3A). The results revealed the prokaryotic community exhibited a distinct clustering trend corresponding to lifestyle preferences (confidence <0.9), supported by the ADONIS permutation-based statistical test (P ≤ 0.001). In addition, a distinct partitioning was found by the frontal hydrographic activities between PA and FL prokaryotes (Figure 3A). The remarkable dissimilarity in the SF, BE and PF water masses was further confirmed by the Pairwise Samples t-test (P ≤ 0.001, Figure 3B). Especially, the BE community retained a partial similarity with both SF and PF communities (Figure 3A).

A total of 69 phyla, 160 classes, 385 orders, 562 families, and 872 genera were identified (Supplementary Table S6). The prokaryotic community in the CDF was predominantly composed of Proteobacteria (48.27%), Bacteroidota (14.18%), Actinobacteriota (8.83%), and Cyanobacteria (5.93%, Figure 4A). The most predominant phylum (Figure 4A), Proteobacteria, showed high relative abundance in FL and PA communities with 52.56% and 45.49%, respectively (Supplementary Table S6). The class Alphaproteobacteria maintained consistent hierarchical dominance across the SF, BE, and PF water masses (Figure 4B). Except for the predominant taxa above, Acidobacteriota surpassed Bacteroidota in relative abundance in each sample across the water column at stations YE9 and YE10 (Supplementary Figure S2). The Campilobacterota exhibits higher relative abundance in the middle sample at YE12 station than in the other samples (Supplementary Figure S2). The prokaryotic community in vertical stratification (e.g., the surface, middle, and bottom water) failed to form a clustering pattern across different stations (Figure 3). Consequently, the significant differences in community combination and distribution patterns were observed in FL and PA samples merged from different depths at each station in the further analysis of community composition (Figure 4).

Further analysis revealed distinct community structures between PA and FL across three water masses (Figure 4B). The PA prokaryotic community included Planctomycetota (5.63%), Acidobacteriota (1.22%), Verrucomicrobiota (1.16%), and Desulfobacterota (1.07%). Thermoplasmatota (1.63%) was the main phylum of the FL community (Figure 4B). The relative abundance of Proteobacteria subsequently declined from the SF to PF water mass in the PA community. At the same time, Cyanobacteria exhibited an elevated tendency of relative abundance in the FL community (Figure 4B). Meanwhile, the relative abundance of Actinobacteriota decreased with the increasing trend of Planctomycetota in the PA community (Figure 4B). The apparent high relative abundance of the prokaryotic phyla except the top 20 phylum indicated there were many kinds of rare prokaryotic taxa in the PA community, particularly in the PF water mass (Figure 4B). In the SF water mass, the sub-dominant group, including Actinobacteriota, Bacteroidota, Crenarchaeota, and Acidobacteriota, respectively occupied 12.52%, 8.70%, 2.92%, and 2.89% relative abundances, respectively (Supplementary Figure S2). Meanwhile, Bacteroidota (15.05%) and Cyanobacteria (7.34%) were the sub-dominant groups in the PF water mass (Supplementary Figure S2). Prokaryotic community within the BE water mass showed intermediate compositional characteristics between the SF and PF water masses (Supplementary Figure S2), such as Bacteroidota (17.71%), Actinobacteriota (9.3%), Cyanobacteria (7.06%) and Crenarchaeota (5.60%, Supplementary Table S6).

To investigate the dynamical adjustment of prokaryotes in response to environmental fluctuations, we combined the community composition with niche preferences defined as the PAN index. More apparent differences in prokaryotes enrichment and contrasting particle preference were shown across distinct water masses at the family level (Figure 5). Most prokaryotes in the BE water mass favored attaching at particles (PAN index >0.5), while those in other water masses showed a transitional tendency between particle-associated and free-living lifestyle (PAN index close to 0.5, Figure 5). Rhodobacteraceae, Flavobacteriaceae and Halieaceae (15.58%, 8.52%, and 3.73%, Supplementary Table S7) were the dominant prokaryotic taxa distributing across three water masses (Supplementary Figure S3). Corresponding to dynamic environmental fluctuations (e.g., turbidity, salinity, and nutrients), prokaryotic communities experienced the compositional transitions across three water masses (Supplementary Table S3). For instance, only in the PF water mass were Oleiphilaceae, Hyphomonadaceae, Pseudoalteromonadaceae, and Cyanobiaceae observed to have a relatively high relative abundance (Supplementary Figure S3). Pseudoalteromonadaceae, Alteromonadaceae, Microtrichaceae, and Sphingomonadaceae uniquely enriched in the FL community of the SF water mass (Supplementary Figure S3). Moreover, in the BE and PF water masses, Rhodobacteraceae, Flavobacteriaceae, Actinomarinaceae, Alteromonadaceae, Cyanobiaceae and Nitrosopumilaceae consistently switched strategies between particle-associated lifestyle and freely living in the aquatic environments, while displaying stronger preferences for free-living lifestyle in the SF water mass (Figure 5). Stappiaceae, Pseudomonadales, and Saprospiraceae showed their preferences for particle-associated lifestyle across all water masses (Figure 5). The PAN index of major components in the FL community (Supplementary Figure S3), such as AEGEAN-169-marine-group, SAR11, SAR116, and SAR86, implied a preference for disassociating from particles (Figure 5). Additionally, Marine_Group_II (MGII), the dominant archaea, emerged with different lifestyle preferences across three water masses (Figure 5).

The mantel test results indicated turbidity was the primary influence factor in the prokaryotic community (Figure 6). Salinity, DIN, and DIP collectively played vital roles in shaping the community structure of both the SF and PF water masses (Figure 6A, Supplementary Table S4). Specifically, the prokaryotic community in the SF water mass exhibited prominent relationships with turbidity (r ≤ 0.4, P ≤ 0.01) and moderate relationships with DIP (r ≤ 0.2, P ≤ 0.05). DSi emerged as the dominant impact factor of the BE water mass (r ≤ 0.6, P ≤ 0.01), followed by salinity, turbidity, and DIN (r ≤ 0.2, P ≤ 0.01). Similar to the BE water mass, salinity, turbidity, DSi, and DIN were key factors impacting the community structure of the PF water mass (Figure 6A). In the SF water mass, the FL community was more responsive to turbidity (r ≤ 0.6, P ≤ 0.01) and DSi (r ≤ 0.4, P ≤ 0.05) compared to the PA community (Figures 6C, D). The PA community structures experienced more pronounced influences by environmental parameters in both BE and PF water masses than the FL community (Figures 6C, D). For instance, PA prokaryotes in the BE water mass displayed remarkable associations with salinity (r ≤ 0.4, P ≤ 0.01), whereas FL prokaryotes lacked substantial associations (Figures 6C, D). Similarly, the PA community in the PF water mass maintained strong associations to environments, in contrast to the FL community which lacked such associations (Figures 6C, D). Additionally, the abundant species in the prokaryotic community of different water masses varied in correlation with environmental parameters (Figure 6B). In the SF water mass, the abundant species showed strong positive correlations with DIN (|r| ≤ 0.6, P ≤ 0.001), DSi (|r| ≤ 0.8, P ≤ 0.001), and turbidity (|r| ≤ 0.6, P ≤ 0.01). Meanwhile, they also had a negative correlation with salinity (|r| ≤ 0.4, P ≤ 0.01), such as Anaerolineaceae, SAR11, Nitrosomonadaceae and Woeseiaceae (Figure 6B).

The neutral community model analysis revealed an excellent fit between the ASV occurrence frequencies and the neutral model in the CDF (R² = 0.8861, Supplementary Figure S5). The value of the nice fit also indicated an overall ecological model close to the neutral theoretical pattern (Supplementary Figure S5). Additionally, the PA community showed stronger alignment to neutral (RPA2 = 0.8311) compared to FL prokaryotes (RFL2 = 0.7637), suggesting the stochastic processes predominantly governed the assembly process of the PA community (Figure 7A). Meanwhile, the consistent patterns emerged across three distinct water masses (Figure 7A). For instance, the PA communities manifested a superior neutral model fit in the SF (RSF-PA2 = 0.5565), BE (RBE-PA2 = 0.4781), and PF (RPF-PA2 = 0.7271) water masses, which exceeded the corresponding fit in the FL communities (RSF-FL2 = 0.4134; RBE-FL2 = 0.399; RPF-FL2 = 0.673). The estimated dispersal extent of the FL community was lower than that of the PA community (Nm_FL = 13,772; Nm_PA = 17,204). Such difference implied the relationship in species migration with diverse lifestyles and enhanced the dispersal capability of the PA prokaryotes (Figure 7A).

The null model analysis quantified the relative contributions of assembly processes. The dispersal limitation (32.80%–34.39%) of stochastic processes and homogeneous selection (30.39%–31.15%) of deterministic processes predominantly controlled the community assembly in the CDF (Figure 7C). The stochastic process was the dominant mechanism in both PA and FL community assemblies (59.9% and 60.16%, Figure 7C), in accordance with the neutral community model (Figure 7A). As the water mass extended further from the CRE, the assembly mechanisms with the most and least contribution varied in different water masses (Figure 7B). As an example, the homogeneous selection functioned as the dominant mechanism for the PA community in the BE water mass and the FL community in the SF water mass. However, the dispersal limitation was the mechanism with the most contribution in other communities. Additionally, the heterogeneous selection represented the least significant ecological process in the FL community assembly (8.69%) (Figure 7C). In contrast, the homogeneous dispersal had the lowest influence on the PA community (8.19%), even less than the heterogeneous selection (9.71%) (Figure 7C). Particularly, the heterogeneous selection and ecological drift exhibited distinctive relative contributions to community assembly across different water masses (Figure 7B). In the PF water mass, the FL community experienced a pronounced effect from heterogeneous selection in comparison with other water masses (Figure 7B). The pronounced contribution of the heterogeneous selection could highlight the broader niche widths of indicator species in the driving mechanism. Meanwhile, ecological drift progressively diminished its contribution to the assembly of the PA community (20.03%–17.44%–12.24%) (Figure 7B). Regarding the FL community, the contribution of ecological drift peaked in the BE water mass (24.28%) (Figure 7B).

Based on the phylogenetic signal thresholds, prokaryotic ASVs from PA and FL communities were classified into 266 and 152 phylogenetic bins, respectively. These bins accounted for 86.97% and 97.91% relative abundance in the PA and FL communities (Figure 8), respectively. By separately cross-referencing the top 25 family from the PA and FL prokaryotic communities, we identified 278 bins belonging to the top 25 family (Figure 8). A total of 162 PA-bins contributed 50.27% to meta-community assembly, predominantly driven by Proteobacteria (23.92%) and Bacteroidota (14.18%, Figure 8A), and these two dominant participants drove dispersal limitation (35.06%) and homogeneous selection (32.19%, Figure 8A), respectively. SAR324 (0.66%) and Crenarchaeota (0.55%) played crucial roles in ecological drift. In contrast, 116 FL-bins exhibited higher contribution proportion (91.10%, Figure 8B). The coordinated effects from Cyanobacteria (83.89%), Crenarchaeota (36.93%), Actinobacteriota (35.81%) and Thermoplasmatota (34.64%) enhanced homogeneous selection (Figure 8B). Specific assembly mechanisms in the sub-community and meta-community were reflected at family level. The PA meta-community was primarily influenced by dispersal limitation, whereas homogeneous selection dominated the FL meta-community assembly (Supplementary Figure S6). Lifestyle transitions between PA and FL dynamically altered the specific mechanism and contribution in the sub-community of assembly drivers. For instance, Rhodobacteraceae, a dominant ecological driving species with high contribution in community assembly of both PA and FL lifestyles, showed the reduced contributions in the FL meta-community of the SF water mass (Supplementary Figure S6b). Sphingomonadaceae and Pseudomonadaceae, freely living in the aquatic environments, drove heterogeneous and homogeneous selection in their sub-community (Supplementary Figure S6b), respectively. While these two contributors transited lifestyle to PA, they shifted synergistically to promote dispersal limitation together with Microtrichaceae rather than the deterministic process (Supplementary Figure S6a). Additionally, the assembly of Moraxellaceae sub-community in both BE and PF water masses was solely controlled by ecological drift (Supplementary Figure S6a). Intriguingly, Cyanobacteria (3.10%) demonstrated divergent assembly patterns owing to lifestyle transitions. FL-Cyanobacteria exhibited a weakened contribution in the PF water mass (Figure 8B), compared to the other water mass. Whereas, PA-Cyanobacteria activated heterogeneous selection and dispersal limitation (Figure 8A). A similar assembly transition was also observed in the sub-community assembly of SAR11, NS9, and SAR86 clades mediated by three water masses (Supplementary Figure S6b).

The hydrological conditions and lateral variation in environmental gradients in the CDF system cause significant differences in community structure between the SF and PF water masses (Figures 2, 3), and form the unique distribution patterns and lifestyle preferences in the SF, BE, and PF water masses (Figures 4, 5).

The sediment front and plume front restrict the material transport and energy exchange, resulting in the constraints of community structure similarity between the SF and PF water masses (Figure 3A). Simultaneously, the instability of the dual fronts induces the exchanges of water masses across the front, leading to a transitional water mass defined as a critical environmental filter (the BE water mass). While the ecological gradients in the BE water mass enhance the ecosystems' isolation, they still maintain a certain similarity with the other water masses (Figures 3A, B). Such divergences in community structure are particularly evident in the unique combinations of abundant prokaryotes (cumulative relative abundance >90%, Figure 6B). The coexistent prokaryotic taxa in all water masses include Rhodobacteraceae, Flavobacteriaceae, Cyanobiaceae, Actinomarinaceae and Nitrosopumilaceae (Supplementary Figure S3), consistent with previous studies in the CRE (Ye et al., 2016; Wu et al., 2019; Jiang et al., 2022; Jin et al., 2024). These common marine species play pivotal roles in marine ecosystems, participating in the biogeochemical cycling of carbon, nitrogen, and sulfur (Wu et al., 2013; Simon et al., 2017; Liu et al., 2020). Apart from the coexistent prokaryotes, abundant prokaryotic taxa in the SF water mass show pronounced correlations with turbidity (Figure 6B). This physical parameter from suspended sediment indirectly reflects the correlations with inorganic materials and organic matter, such as chronic nitrogen eutrophication in the SF water mass (Hubeny et al., 2017; Yang et al., 2022; Matos et al., 2024). As an example, Pseudomonadaceae possess the capability to degrade terrestrial organic matter (Girard et al., 2023), which can utilize substantial organic matter carried by riverine particles in the SF water mass (Zhou and Ning, 2017; Yang et al., 2022).

The environmental selection and niche competition in distinct microhabitats motivate PA and FL prokaryotes to change their evolutionary trajectories. To minimize competition with other prokaryotes, prokaryotes with different lifestyles can adjust their physiological and morphological traits, genomic composition, physiological function, and survival strategies (Badger et al., 2006; Guider et al., 2024). Compared to the oligotrophic aquatic environments where FL prokaryotes reside, the available suspended particulate matter provides critical substrates for survival and proliferation (Wen et al., 2024; Yu et al., 2025). The dual frontal effects can exacerbate the divergency of PA and FL communities in structure diversity and taxonomic combinations. Previous study have observed the genome of FL-MGII contained a large number of genes encoding rhodopsin and opsin, while PA-MGII possessed functional genes encoding cell adhesion proteins (Tully, 2019). The FL-MGII with photosynthetic strategies could reduce the dependence on particulate organic matter in BE and PF water masses with improved light condition by flocculation (Figure 5). However, the hydrodynamical activities lead to the decoupling of the correlation between FL-MGIII and photosynthetic autotrophs, resulting in the maladaptation to FL lifestyle (Figure 5). Therefore, the MGII in the SF water mass favoring a particle-associated lifestyle may benefit from its functional genes encoding cell adhesion proteins (Orsi et al., 2015; Wang et al., 2020). The frontal effect can also provide convenient access and functional microhabitats for prokaryotes participating in the biogeochemical cycles, whose metabolic activity often requires an anoxic microenvironment in particulates. For example, Desulfobacterota, including Desulfobulbaceae and Desulfuromonadaceae families, are capable of participating in sulfate reduction and denitrification under anaerobic conditions (Waite et al., 2020; Katie E. Hillyer et al., 2023). Consequently, Desulfobacterota usually occurs and accumulates in the sediment-water interface and marine sediments. The relative abundance of Desulfobacterota in the SF water masses (1.07%, Figure 4B) benefits from the tidal mixing and resuspension process. The resuspended particles induced by tidal mixing are considered as relative anoxic microenvironments, supporting the colonization and proper progression of biological activities for Desulfobacterota (Hallstrøm et al., 2022; Robicheau et al., 2023).

Our data demonstrated that stochastic dispersal limitation and deterministic homogeneous selection dominated the assembly process of prokaryotic communities in the CDF (Figure 7, Supplementary Figure S5). Collaboration between intricate turbulent hydrodynamic conditions and lifestyle preferences shifts accounts for the main assembly mechanisms in the CDF system. The current studies have shown the stochastic dispersal rate of the community accelerates with hydrological mixing (Meyerhof et al., 2016; Zhou et al., 2021), which can be applied to the tidal-induced mixing in the SF water mass and the convergency of secondary circulations in the PF water mass. These complicated hydrological activities elevate the contribution of the stochastic process by strengthening the stochastic dispersal rate (Figure 7). The elevated dominance even surpasses the ecological selection imposed by environmental heterogeneity in each water mass, consistent with previous research (Lian et al., 2023; Jin et al., 2024). Consequently, dispersal limitation emerges as the principal mechanism governing prokaryotic community assembly in the dual fronts system (Figure 7).

Hydrological mixing processes in the frontal zone significantly modulate the alterations in water mass (e.g., nutrient gradients and combination and source of organic particulate matter), playing pivotal roles in driving community assembly. In the sediment front, sufficient nutrients (Supplementary Table S3) create favorable niches for prokaryote enrichment, increasing the contribution of homogeneous selection to FL community assembly (Figure 7B). In the plume front, the oligotrophic aquatic environment imposes stringent constraints and selective stress on FL prokaryotic proliferation, which stimulates the compositional divergence in the prokaryotic community. Eventually, intense environmental stress results in a stronger influence on heterogeneous selection (Figure 7B). The elevated contribution of heterogeneous selection also suggests there are indicative species in the PF community, such as Hyphomonadaceae with high contributions to community assembly (Supplementary Figure S2). Conversely, organic particulate matter and algal-derived polymers accumulating in the BE water mass mitigate environmental pressures on PA prokaryotes (Li W. et al., 2021; Jin et al., 2024). Therefore, the PF-induced convergence effect can facilitate stable microenvironmental filtering to drive particle-mediated community assembly.

Lifestyle preferences additionally exert profound influences on community assembly in the CDF system, which manifests as the reconfiguration of driving species composition in the sub-community and meta-community. The driving mechanism of lifestyle preferences in assembly mechanisms primarily relates to biodiversity gradients of both meta-community and sub-community. FL prokaryotes, detaching from particles and inhabiting aquatic environments, occupy broader ecological niches and higher spatial turnover rates than PA prokaryotes (Chun et al., 2021; Leu et al., 2022; Zhang et al., 2023). In contrast, the PA prokaryotes are constrained by particle-mediated dispersal within the dual-front system (Figure 7). From a macroscopic perspective, the meta-community theory can account for the different assembly mechanisms of lifestyle preferences (Bjornstad et al., 2009). Such as the declined environmental selection intensity by increasing biomass and community size, eventually modulates the ecological processes in meta-community assembly (Ning et al., 2019; Niu et al., 2022; Lerch et al., 2023). In consequence, the larger meta-community size of the PA prokaryotic community compared to the FL community (Figure 2) alleviates the impacts on the dispersal limitation of the PA community assembly. Such consequences can explain the contrary results based on the neutral community model (Figure 7A). Although the meta-community in the PF water mass experiences a higher limitation in dispersal than the BE water mass, the FL and PA meta-community in the PF water mass disperse wider (Figure 7A). The prokaryotic community in the PF water mass possesses a more extensive meta-community than the other water mass, which spans vertically across the plume front (Figure 1). In consequence, the larger meta-community size in the PF water mass alleviates the impacts on dispersal limitation. Eventually, the counteracting effect explains the significant influence of dispersal limitation, but low dispersal extent of the meta-community mentioned above (Figure 7A). From a microscopic perspective, the relative abundance of specific prokaryotic taxa is the main reason to drive the assembly processes of the sub-community. For instance, the dominance in relative abundance and exceptional taxonomic diversity (Figures 8A, B) directly make Proteobacteria a major driver of community assembly in the CDF (Vellend et al., 2014; Evans et al., 2017; Xun et al., 2019). Including Alphaproteobacteria and Gammaproteobacteria, these families play pivotal ecological roles in community assembly when associated with particulates or freely living in the aquatic environment (Supplementary Figure S3). Moreover, the community with high abundance is easily prone to induce disproportionate richness, resulting in ecological drift and dispersal limitation collectively driving the assembly of Proteobacteria sub-community (e.g., halotolerant Rhodobacteraceae, Figure 8).

The intensive hydrodynamic activities stimulate the attachment-detachment on particulates of prokaryotes by providing mechanical energy and constraining the availability of substrate. Eventually, this synergistic mechanism between hydrodynamic activities and lifestyle preferences exerts a vital influence on the community assembly processes. For instance, localized accumulation in both SF and BE water masses causes the disproportionate distribution of particulates across the water column (Supplementary Table S3), causing pronounced disparities in abundance (Evans et al., 2017; Xun et al., 2019). Eventually, such structural imbalances produce a high contribution in ecological drift progressively from the SF water mass to the PF water mass (Figure 8B). The increasing contributions in ecological drift suggest species extinction, immigration, and recolonization events substantially influence the community assembly processes (Ning et al., 2020). FL prokaryotes exhibit heightened environmental sensitivity, their detachment from particulates intensifies homogeneous dispersal and homogeneous selection in community assembly (Banerjee et al., 2016). Conversely, biofilm-forming PA prokaryotes are buffered from environmental fluctuations, reducing the importance of stochastic dispersal and increasing their susceptibility to dispersal limitation (Banerjee et al., 2016), such as Sphingomonadaceae (Supplementary Figures S6a, b). In the SF water mass, a rapid increase in suspended particulate matter reduces light penetration and restricts the survival conditions of Sphingomonadaceae (Ivanova et al., 2020; Zhou et al., 2020; Li W. et al., 2021). In consequence, Sphingomonadaceae taxa are unable to thrive in the FL community (Figure 5), and merely attach to particulates to convert organic substrates into essential biomolecules. The metabolic function traits of Sphingomonadaceae increase its contribution to dispersal limitation in PA community assembly (Supplementary Figure S6). However, the tidal-reduced resuspension processes in the SF water mass drive prokaryotes to shift lifestyle to freely floating, eventually diminishing the importance of dispersal limitation (Ohore et al., 2022). In the SF water mass, the particle-associated lifestyle of Cyanobacteria restricts cell mobility and reduces the susceptibility to environmental heterogeneity. In consequence, the tendency of particle-associated lifestyle weakens the ability of PA-Cyanobacteria to adjust to selection pressures (Figure 8B) and elevates the relative importance of dispersal limitation and heterogeneous selection (Supplementary Figure S6).