The survey covered 16 stations with different depths in the PRE and NSCS, producing 52 samples of FL and PA microbial communities (Supplementary Figure S1; Supplementary Table 1). Sampling stations were categorized into the Pearl River Estuary (PRE) and the northern South China Sea (NSCS) based on natural bathymetric differences, with PRE stations located in shallow estuarine waters and NSCS stations located in deeper offshore waters. Locations in the PRE were marked with ‘P’ and those in the NSCS with ‘S’. A CTD system (SBE 9–11 Plus, Sea-Bird Electronics, USA) with 12 L Niskin bottles was used to measure hydrographic parameters such as depth, temperature, and salinity. The research included samples taken from various depth layers, with depths ranging from 5 to 25 m in the PRE and 5 to 67 m in the NSCS. Detailed related site data can be referred to the previous article of our team (Chen et al., 2022).

Seawater ranging from 2 to 4 liters was filtered in sequence using 2.7 μm glass microfiber membranes followed by 0.22 μm cellulose nitrate membranes to distinguish PA from FL fractions. Particles and their communities that are trapped by 2.7 μm prefilters are identified as the PA community in this study, while those that pass through these filters but are stopped by 0.22 μm filters are identified as FL communities (Cui et al., 2020). Therefore, FL eukaryotes mainly consist of small eukaryotes that are not bound to large particles and their FL life stages. The filter membrane was quickly frozen using liquid nitrogen within 20 min for DNA extraction, and the filtrate was kept at −20 °C for nutritional analysis. Ammonium levels were measured using the indophenol blue spectrophotometric technique (Sims et al., 1995). A continuous flow autoanalyzer was employed to measure nitrate, nitrite, phosphate, and silicate (Technicon AA3, Germany). All procedures were conducted in accordance with institutional and national guidelines, no endangered species were involved.

With slight modifications, DNA was extracted from filters using the FastDNA SPIN Kit for Soil (MP Biomedicals, USA) (Tournier et al., 2015). The steps included cryopreserving the filters, lysing them, precipitating proteins, and purifying through Spin columns. Using the Illumina MiSeq platform, amplicon sequencing was conducted targeting the universal 16S/18S rRNA genes with the primer pair 515F/926R (5′-GTGYCAGCMGCCGCGGTAA-3′)/(5′-CCGYCAATTYMTTTRAGTTT-3′) (Walters et al., 2015). Cutadapt was used to trim low-quality sequences (average quality < 20) and remove primer regions (Martin, 2011). The DADA2 plugin was used to truncate, denoise, and remove chimeras from the trimmed sequences. Taxonomic classification of ASVs was performed using the SILVA 138.1 database with the classify-sklearn plugin in QIIME2 (Bolyen et al., 2018; Quast et al., 2013).

The universal primer pair 515F/926R amplifies small subunit (SSU) rRNA genes from bacteria, archaea, and eukaryotes. After quality control and ASV inference, taxonomic assignment was performed against the SILVA SSU rRNA database. ASVs classified as Eukaryotes were extracted for downstream analyses. To ensure the reliability of eukaryotic community profiles, only ASVs with confident taxonomic assignments (bootstrap ≥70%) were retained, while non-eukaryotic, ambiguous, or low-abundance ASVs occurring in fewer than 2 samples were removed. Previous evaluations have demonstrated that recovery of 18S rRNA sequences from 515F/926R amplicons depends primarily on downstream bioinformatic filtering rather than primer bias itself (Parada et al., 2016). Accordingly, although differential amplification efficiencies across domains, particularly for eukaryotic taxa, cannot be entirely excluded, primer bias is unlikely to be the primary source of systematic variation in the resulting community profiles (Parada et al., 2016). Furthermore, this primer pair offers high coverage and minimal bias in marine systems, ensuring robust comparability across domains (McNichol et al., 2021). Serial filtration protocols minimized size-related artifacts in community assembly and network inferences (Levine et al., 2024).

To ensure uniform sequencing depth for subsequent statistical and ecological analyses, ASV tables were rarefied in R using the rrarefy function in the vegan package. Rarefaction was performed separately for each domain by randomly subsampling each sample to the minimum sequencing depth within that dataset: 668 reads per sample for archaea, 77,367 reads per sample for bacteria, and 2,598 reads per sample for eukaryotes.

The sequence data were standardized to a uniform sequencing depth across all samples. We created the sampling map using Ocean Data View.¹ Community assembly was quantified using the iCAMP phylogenetic-bin-based null model with 1,000 randomizations, using standard thresholds for βNRI (±1.96) and RCbray (±0.95) to assign ecological processes (Ning et al., 2020). To better interpret and present data, Chiplot created various visualizations, including stacked bar charts, non-metric multidimensional scaling (NMDS), canonical correspondence analysis (CCA), phylogenetic trees, and heatmaps (Li et al., 2023a). Prior to CCA, environmental variables (depth, temperature, salinity, and nutrient concentrations) were examined for multicollinearity using variance inflation factors (VIFs), and no severe collinearity was detected (all VIF < 10). Therefore, all variables were retained in the final CCA model. In R (vegan v2.6–4), variance partitioning analysis (VPA) was utilized to pinpoint the primary environmental influences (Lyu et al., 2023). Cross-domain interaction networks were constructed in Wekemo Bioincloud, retaining the top 1% abundant genera and filtering edges by Pearson correlation (|r| > 0.8, FDR-adjusted p < 0.05), with isolated nodes removed (Gao et al., 2024). The core taxa were defined as those with a relative abundance >0.5% in at least one sample, and keystone roles were assigned based on Zi and Pi thresholds (Zi > 2.5 for module hubs, Pi > 0.62 for connectors, Zi > 2.5 and Pi > 0.62 for network hubs).

After quality control, taxonomic filtering, and domain-specific rarefaction, 648 archaeal ASVs, 13,283 bacterial ASVs, and 2,543 eukaryotic ASVs were retained for community composition analyses. Microbial communities showed clear spatial and vertical variation across sampling sites (Figure 1). In archaeal communities, the community composition shifted markedly from the estuary to offshore waters. Thermoproteota dominated in the PRE (61–65%), followed by MGII (25%), while MGIII remained rare. In contrast, MGII was highly abundant in the NSCS (83%), and MGIII was more common in both PA and FL fractions compared with the PRE, suggesting a stronger adaptation of MGII/MGIII to oligotrophic offshore conditions. Bacterial communities were dominated by Proteobacteria, with Actinobacteriota enriched in the NSCS PA fraction and Cyanobacteria more abundant in the FL fraction, consistent with their photoautotrophic lifestyle. Among eukaryotes, Stramenopiles dominated PRE PA (58%), while Archaeplastida prevailed in PRE FL (66%). In NSCS PA, Hacrobia accounted for 39%, much higher than in the PRE (14%), implying regionally specific particle colonization patterns. These results indicated domain-specific community structure shaped by both environmental gradients and particle association.

The NMDS analysis indicated that two primary axes, extending from estuaries to offshore waters and from FL to PA habitats, collectively influence the communities. However, the strength and direction of these influences differ by domain (Supplementary Figure S2). Archaea did not display significant differences between FL and PA states (p > 0.05), but they did show a distinct regional separation (p < 0.01), indicating significant biogeographic differentiation rather than lifestyle-driven differentiation. Bacteria only exhibited segregation between FL and PA states in the NSCS (p < 0.01), whereas eukaryotes demonstrated pronounced differences in both regions (p < 0.01), indicating significant biogeographic differentiation. Overall, these results indicated that geographical factors dominate the differences in archaeal communities, while lifestyle differences shape the variations in bacterial and eukaryotic communities.

Along the PRE to NSCS continuum, archaeal, bacterial, and eukaryotic communities separated mainly along the combined salinity and depth gradient, whereas vectors for nutrients such as nitrate (NO₃⁻), nitrite (NO₂⁻), ammonium (NH₄⁺), phosphate (PO₄³⁻), and silicate (SiO₃²⁻) and temperature (T) indicated clear domain-specific sensitivities (Figures 2a–c). The environmental matrix included eight key hydrographic and nutrient variables, which showed no severe multicollinearity and were therefore all incorporated into the final CCA model. The first two canonical axes explained 29.7, 15.0, and 6.1% of the variance in archaeal, bacterial, and eukaryotic communities, respectively, indicating a progressively weaker coupling with the measured environmental variables from archaea to eukaryotes, with the strongest coupling in archaea and the weakest in eukaryotes. Pairs of FL and PA samples shifted roughly in parallel with the environmental vectors, consistent with a shared response to large-scale conditions, while PA samples were generally more dispersed, reflecting microhabitat heterogeneity of particles.

In the case of archaea, low-salinity estuarine samples and high-salinity, deeper offshore samples were distinctly separated along the first axis. The community structure correlated positively with salinity and depth, and inversely with several nutrients (Figures 2a,d). Variation partitioning attributed approximately 65% of archaeal variation to the measured environment, confirming a strong deterministic influence dominated by temperature at 0.11, phosphate at 0.05 and nitrate at 0.04. The interaction term was 0.22, indicating a synergistic effect between nutrients and physical factors. The FL and PA communities tended to converge offshore but remained distinct in the estuary, reflecting lifestyle-modulated environmental filtering.

In bacteria, both salinity and depth contributed to community differentiation and ammonium had an impact at NSCS (Figures 2b,e). The measured environment accounted for approximately 27% of the variation with modest unique effects from temperature (0.06), and minor contributions from nutrients (0.01). The substantial unexplained fraction (0.73) suggested that processes not captured by the measured environmental variables, such as dispersal limitation, stochasticity, or biotic interactions (e.g., grazing or viral lysis), may contribute to bacterial community variation. The significant FL-PA community differences observed in PRE are sufficient to support that bacteria adopt diverse lifestyles to achieve adaptation and survival when responding to variable marine environmental conditions.

For eukaryotes, separation in ordination space was weakest, and environmental vectors clustered near the origin (Figures 2c,f). Variation partitioning attributed only about 4% of the variation to the measured environment, suggesting dominance of unmeasured or stochastic processes. Unique and shared fractions were small, indicating that unquantified or stochastic processes, such as potential micro food web interactions, temporal lags, or particle size structure, likely play dominant roles.

Overall, the three domains exhibited an asymmetric response to environmental gradients. Archaea showed the strongest and most deterministic coupling with the environment, bacteria exhibited an intermediate response influenced by both environmental and stochastic factors, and eukaryotes were largely governed by stochastic or biotic interactions. Across all domains, PA communities were more spatially dispersed than FL counterparts, consistent with PA dispersal limitation and niche heterogeneity.

For iCAMP analyses, low-abundance ASVs were filtered out to reduce stochastic noise, resulting in 183 archaeal, 931 bacterial, and 608 eukaryotic ASVs retained for downstream assembly inference. Null model partitioning revealed the assembly processes of three-domain microorganisms in different regions and lifestyles. In the archaeal community, the PRE group strongly favored drift, with heterogeneous selection providing a secondary influence (Figure 3a). In the NSCS, FL archaea appeared to be primarily influenced by dispersal limitation, while PA assemblages showed a mixed signature of drift and dispersal limitation, with only minor traces of selection or homogenizing dispersal. Overall, the archaeal evidence showed greater links to geographical factors than to environmental conditions.

In bacteria, patterns diverged according to lifestyle (Figure 3b). Whether it is PRE or NSCS, the PA community exhibited a combination of dispersal limitation and a relatively large proportion of drift. In PRE, the FL community primarily manifested as dispersal limitation, whereas in NSCS, homogeneous selection was the dominant process which showed greater environmental influence in offshore areas.

For eukaryotes, assembly exhibited a pronounced dependence on habitat. PRE-FL communities displayed a mix of drift and homogeneous selection, with drift being the dominant process. In contrast, PRE-PA communities were predominantly shaped by homogeneous selection. In the NSCS, FL eukaryotes predominantly experienced homogeneous selection with greater dispersal limitations compared to the estuary, whereas PA eukaryotes were primarily influenced by dispersal limitations (Figure 3c).

Overall, the assembly of communities were jointly structured by region and lifestyle, but their influence depends on specific ecological domains. Archaeal communities were more strongly influenced by drift in estuarine environments, whereas dispersal limitation dominated in offshore free-living assemblages. Bacteria showed that PA had dispersal limitation but NSCS-FL had homogeneous selection, suggesting the distribution patterns based on lifestyle. Eukaryotes showed the strongest habitat dependence because selection dominates in PA and offshore FL while dispersal limitation continues to affect PRE-FL and NSCS-PA.

Networks were constructed at the genus level using a stringent relative abundance threshold (≥1%), resulting in a limited number of dominant archaeal and eukaryotic nodes but a larger set of bacterial genera, thereby emphasizing robust and ecologically interpretable cross-domain interactions. Network topology changed systematically with both geographic region and lifestyle (Figure 4; Supplementary Table 2). The NSCS-FL network formed the densest web and contained the most cross-domain links, whereas PRE-FL network was comparatively sparse and weakly connected. Both PA networks were more modular, consistent with particles acting as discrete microhabitats that partition niches. Because the number of available samples differed between PRE and NSCS (6 vs. 20), network metrics related to connectivity should be interpreted in a relative rather than absolute sense. Nevertheless, the qualitative patterns reported here, including higher modularity in PA networks and greater cross-domain connectivity in NSCS networks, were consistent across regions and therefore unlikely to be driven solely by differences in sampling effort.

The PRE-PA network was dominated by archaeal bridge taxa, particularly AOA such as Nitrosopelagicus and Nitrosopumilus, which linked archaeal and eukaryotic modules (Figure 4a). Choanozoa and Syndiniales served as key eukaryotic mediators, showing positive associations with AOA but negative correlations with co-occurring bacteria. Prochlorococcus has a negative correlation with other bacteria, but a positive correlation with the eukaryotic group Mamiellales. Syndiniales, MGII, and Pirellula or OM190 formed positive three-domain co-occurrence modules, highlighting archaeal-bacterial-eukaryotic coupling on particles.

In contrast, the PRE-FL network displayed low connectivity and strong modular isolation (Figure 4b). MGIII developed unfavorable bonds with Mamiellales and Prochlorococcus but MGII built favorable connections with Syndiniales and Marinoscillum. The absence of dominant bridge taxa indicated weak cross-domain coupling in the highly dynamic estuarine water column.

The NSCS-PA networks showed increased structural complexity because they used eukaryote-based bridging connections (Figure 4c). The Syndiniales and Pelagomonadales groups served as vital connections between different domains because they formed beneficial relationships with MGIII and heterotrophic bacteria but showed negative interactions with various archaea and bacterial species. The Choanozoa clade presented wide-ranging negative relationships among its different groups, whereas MGII maintained relatively few connection points. The observed patterns demonstrated that offshore particles need eukaryotic mediators and heterotrophic bacteria to establish connections between different modules.

The NSCS-FL network formed a highly coupled structure, anchored by AOA (Nitrosopelagicus, Nitrosopumilus) and core heterotrophic clusters such as SAR86, SAR116, and AEGEAN-169 (Figure 4d). Pelagomonadales served as cross-domain connectors but Mamiellales and Choanozoa maintained limited negative connections between their groups. The Prochlorococcus MIT9313 strain established itself in the middle through its negative connections to heterotrophic clusters yet AOA maintained high betweenness centrality, indicating their bridging role between functional modules.

Overall, the research results showed that geographical position together with individual behaviors determine how different domains connect through their networks. The PRE-PA network needed AOA bridges while NSCS-PA network needed eukaryotic mediators to operate and NSCS-FL network depended on heterotrophs and AOA for their co-anchoring. The PRE-FL community showed the lowest connection strength because domain coupling remains weak when stochastic conditions become dominant. This progressive shift from archaeal to eukaryote-dominated bridging reflects an ecological transition from environmental filtering to functional integration along the PRE-NSCS continuum.

The top 20 microbial taxa at the ASV level from the three domains were identified based on their cumulative relative abundance across all samples and subjected to phylogenetic and relative abundance analyses to determine the dominant lineages shaping community structure (Supplementary Table 3). The research showed that core taxa exist throughout all regions but domain groups showed distinct abundance patterns which indicated their success in different ecological environments from PRE to NSCS.

Among the top 20 most abundant archaeal lineages, MGII and AOA (mainly Ca. Nitrosopelagicus) were most prominent (Figure 5a). Most subgroups of MGII, MGII-bO1, MGII-bN1, MGII-bN2, MGII-aK1, MGII-aM, and MGII-aL1, maintained low and stable levels across different sites, but their relative abundance remained higher in NSCS than in PRE because they thrive in offshore oligotrophic waters. Conversely, the PRE samples showed high Ca. Nitrosopelagicus levels which indicated AOA prefer to live in areas with elevated nutrient levels and particular redox gradient conditions.

The bacterial community consisted mainly of oligotrophs and phototrophs which included SAR11, Prochlorococcus, Synechococcus, Alteromonas (Supplementary Figures S3a–c). SAR11 was the most ubiquitous group, appearing in almost all samples, with Clade Ia showing a stable high proportion (Figure 5b). Prochlorococcus and Synechococcus were significantly enriched in NSCS-FL because they thrive best in environments with plenty of light and limited nutrient availability. The opportunistic copiotroph Alteromonas showed occasional peak occurrences because it responds to short periods of organic matter availability.

The eukaryotic assemblages exhibited significant spatial and temporal differences (Figure 5c). The Bacillariophyta, Cryptomonadales, and Phaeocystaceae occasionally bloomed at certain sites but Mamiellaceae established a permanent presence in the PRE which linked primary producers to microbial heterotrophs. Despite their low abundance, Eustigmatophyceae were consistently detected at multiple sites which showed they can thrive in various environmental conditions. The observed patterns indicated that eukaryotic microalgae generate local primary production and form particles which affect the habitats of bacteria and archaea.

Across the PRE-NSCS continuum, the three domains maintained broadly similar dominant taxonomic groups, but their relative abundances and spatial distributions varied markedly between regions. This pattern suggests that environmental constraints and lifestyle strategies shape geographic differentiation without completely restructuring the core taxonomic composition. The oligotrophic NSCS contains MGII and SAR11 as its dominant species but AOA and Mamiellaceae predominate in the nutrient-rich PRE. The three taxonomic groups function as operational bases which facilitate carbon-nitrogen coupling and cross-domain interactions between estuarine and offshore environments.