The seawater samples were collected from 6 sites aboard R/V Dongfang Hong 2 in June 2018 (Figures 1A,C) and 7 sites aboard R/V Dongfang Hong 3 in December 2019 (Figures 1B,D), respectively. Summer seawater samples from the PN section were collected from 122.61–125.50°E, 29.12–31.33°N, while winter seawater samples were collected from 122.72–127.60°E, 28.15–30.96°N. The seawater samples were collected using a Sealogger CTD (SBE25, Electronic Inc., United States) rosette water sampler. Samples containing 1 L seawater were filtered serially through 3 μm and 0.22 μm polycarbonate membranes (Millipore Corporation, Billerica, MA, United States), respectively. The collections above 3 μm and between 0.22 μm and 3 μm were considered as the PA and FL samples, respectively. The detailed information of 52 summer samples (26 FL and 26 PA) and 68 winter samples (34 FL and 34 PA) in total are shown in Supplementary Tables S1, S2. In addition, 2 mL of water were taken from each sample into a sterile tube and were immediately fixed with paraformaldehyde (final concentration 4%, v/v) for 30 min in the dark at room temperature, and were used for subsequent determination of absolute abundance of eukaryotes and prokaryotes. Membranes and 2 mL water samples were frozen in liquid nitrogen immediately and stored at −20°C on board before transferred into −80°C in laboratory.

For environmental parameters, temperature, salinity and depth were obtained by CTD equipped on the water sampler in situ. The concentrations of Chlorophyll a (Chl a) were measured as described by Zhang Y. et al. (2014). Briefly, seawater samples were filtered by using GF/F filter with a pore size of 0.7 μm immediately after collection on board, and then soaked in 90% (v/v) acetone in the dark for 24 h to extract Chl a. The concentrations of Chl a in the extract were determined using a F4500 fluorescence spectrophotometer (Hitachi, Japan). The dissolved oxygen (DO) was measured by the Winkler method (Carpenter, 1965). Samples for dissolved inorganic nutrients (PO₄³⁻, NO₂⁻, NO₃⁻, SiO₃²⁻ and NH₄⁺) were filtered with 0.45 μm cellulose acetate membranes and were analyzed by an Auto-Analyzer (AA3, Seal Analytical Ltd., UK) (Liu et al., 2015). The abundances of Synechococcus (SYN), Prochlorococcus (PRO), picoeukaryotes (PEUK), and heterotrophic bacteria (HB) were measured by flow cytometer (BD FACSJazz, United States). SYN, PRO and PEUK were detected directly by their autofluorescence and size (SYN: orange and < 2 μm, PRO: red and < 2 μm, PEUK: red and > 2 μm). For HB, 50 μL seawater samples are diluted with 250 μL TE buffer solution (Tris EDTA, 100 mM Tris-Cl, 10 mM EDTA, pH = 8.0, Sigma, United States), and 4 μL nucleic acid dye SYBR Green I (Molecular Probes, United States) was added to stain for 20 min before detection (Zhao et al., 2012, 2017).

Total DNA were extracted from 52 summer and 68 winter samples using the Phenol-chloroform method (Yin et al., 2013; Sun et al., 2020). The extracted DNA was dissolved in 10 mM Tris–HCl (pH 8.0) and stored at −80°C for subsequent 16S rRNA gene amplicon sequencing and qPCR.

The abundances of total bacterial 16S rRNA genes in seawater samples were quantified by qPCR with the primer set 338F (5′-ACTCCTACGGGAGGCAGCAG-3′) and 518R (5′-ATTACCGC GGCTGCTGG-3′) (Yin et al., 2013) on StepOne^™ Real-time PCR System (Applied Biosystems). The PCR reactions, standard curves and melting curves were conducted as previous described by Sun et al. (2020). According to the standard curves, the detection range of qPCR was 5.02 × 10⁵ copies L⁻¹- 5.02 × 10¹¹ copies L⁻¹. The specific PCR reactions were as follows: an initial denaturation at 95°C for 3 min, then 35 cycles of 95°C for 30s, primer- specific annealing temperature 53°C for 30 s, 72°C for 30 s. The qPCR gene amplification efficiency ranged from 95 to 105%. Three techniques replicates were set for each sample. Double-distilled water was used as the template for negative control.

The 16S rRNA gene amplicon sequencing was conducted by Majorbio Bio-Pharm Technology Co. Ltd. (Shanghai, China). The hypervariable V4 region of the bacterial and archaeal 16S rRNA gene was amplified by using primers 515FmodF (5′-GTGYCAGCMG CCGCGGTAA-3′) and 806RmodR (5′-GGACTACNVGGGTW TCTAAT-3′) (Walters et al., 2016). The PCR amplification was conducted in 20 μL mixture comprising 4 μL of 5 × FastPfu buffer, 2 μL of 2.5 mM dNTPs, 0.8 μL of primers (5 μM), 1 U of TransStart Fastpfu DNA polymerase and 10 ng of template DNA. The PCR cycling was as follows: 95°C for 3 min; 29 cycles of denaturation at 95°C for 30 s, annealing at 55°C for 30 s, extension at 72°C for 45 s; final extension at 72°C for 10 min. The PCR amplification product was purified using the AxyPrep DNA Gel Extraction Kit (Axygen Biosciences, Union City, CA, United States), and the DNA was quantified using QuantiFluor^™-ST (Promega, United States). The purified amplicons were pooled in equimolar and sequenced paired-end (2 × 300 bp) on the Illumina MiSeq platform (Illumina, San Diego, United States) according to the standard protocols. After subsampling each sample to an equal sequencing depth according to the minimum number (27,570) of sample sequences, operational taxonomic units (OTUs) were clustered using Usearch7.0 method of the QIIME1.9.1 with 97% similarity cutoff. The taxonomic assignment of each OTU representative 16S rRNA gene sequence was determined by Silva 128 16S rRNA database¹ using confidence threshold of 70%.

The alpha diversity indices including Shannon, Chao1 and Good’s coverage were calculated to measure the species richness and diversity of the community using Mothur (version 1.30.2) (Sun et al., 2016; Bullock et al., 2017). For beta diversity, non-metric multidimensional scaling analysis (NMDS) were performed with ANOSIM based on Bray-Curtis distance matrices using the “vegan” package (version 3.9) in R software (version 4.1.1). The differences in bacterial diversity and richness between different groups were analyzed by Wilcoxon signed-rank test. The relationship between environmental factors and bacterial community structure was evaluated by distance-based redundancy analysis (db-RDA) with 999 Monte Carlo permutation tests using the Canoco software (version 5.0, Microcomputer Power). The correlations between environmental factors and bacterial 16S rRNA gene abundances were conducted using Spearman correlation test. Mann–Whitney U test was used to detect the difference of total bacterial abundance between summer and winter. The difference between the FL and PA bacterial abundance was examined by Independent Samples t-test. These statistical analyses were performed on SPSS version 25.0 (SPSS, Chicago, IL, United States). In addition, Mantel test based on Pearson’s correlations was carried out by the “ggcor” package (version 0.9.4) in R (v4.2.1). The test for differences in microbial composition between groups was performed by the software STAMP (v2.1.3) (Parks et al., 2014). The map of sampling sites was created using Ocean Data View (ODV, v5.1.7) and figures were drawn by Origin 2022b software² or GraphPad Prism 6.01.

A null model analysis was conducted following the statistical framework outlined by Stegen et al. (2013, 2015). The key metrics employed to discern these processes were the beta nearest taxon index (βNTI) and the Bray-Curtis-based Raup-Crick metric (RCbray), both calculated using the “picante” package (version 1.8.1) (Kembel et al., 2010). βNTI values exceeding 2 indicate that deterministic processes predominantly shape community structure. Specifically, βNTI values below −2 or above 2 correspond to homogeneous and heterogeneous selection, respectively. When |βNTI| values fall within the range of −2 to 2, it suggests that stochastic processes are the primary drivers of community assembly. These specific stochastic processes can be further deduced by considering the RCbray values. For |βNTI| < 2, RCbray < −0.95, RCbray > 0.95, and |RCbray| < 0.95 represent homogenizing dispersal, dispersal limitation, and ecological drift, respectively.

Co-occurrence networks were constructed using the “igraph” (version 0.10.7), “Hmisc” (version5.1–1) and “qvalue” (version 2.34.0) libraries in R (v4.2.1). To simplify the analysis, we retained only those Operational Taxonomic Units (OTUs) that had a relative abundance exceeding 0.01% across all samples and were present in more than 20% of the samples. We calculated pairwise Spearman’s correlations between these selected OTUs. Correlation coefficients exceeding |0.7|, with a p-value below 0.01 (adjusted using the Benjamini and Hochberg method), were considered as meaningful relationships. We assessed both network-level and node-level topological features of the co-occurrence network. Network-level characteristics included average degree, average network distance, average clustering coefficient, modularity index and diameter. Node-level characteristics encompassed nodes, edges, and proportion of positive correlation. Finally, the network visualization was done using Gephi (version 0.10.1). In each cooccurrence network, keystone species with high degree and low betweenness centrality were identified as those ranking within the top 20% for degree centrality among all nodes, and simultaneously ranking within the bottom 20% for betweenness centrality among nodes with high degree as described by Berry and Widder (2014).

Raw reads from the 16S rRNA gene amplicon sequencing were deposited in the NCBI BioProject database under the accession number PRJNA648032 (summer samples from June 2018) and PRJNA985391 (winter samples from December 2019).

In summer, surface seawater from P1 exhibited low temperature (23.28°C) and salinity (26.51 PSU) compared to other sites, indicating a strong influence by the river plume of Changjiang Diluted Water (CDW). Correspondingly, the dissolved inorganic nitrogen (NO₃⁻, NO₂⁻, NH₄⁺) and silicate (SiO₃²⁻) showed highest concentrations, while phosphate (PO₄³⁻) was depleted in P1 surface water (Figure 2; Supplementary Table S1). The temperature (25.7–27.7°C) and salinity (31.7–32.6) of P2 and P3 surface seawater suggested that these samples were also located in the water mass of continental coastal water (CCW) (Zhou et al., 2019). In comparison, the surface seawater of P4, P5 and P6 seemed to be influenced by the Taiwan Warm Current (TWC), a major current from the open sea with its surface water originating from the Taiwan Strait (Shi et al., 2014), resulting in a salinity of ~34. Most importantly, clear features of the Kuroshio subsurface water (KSSW) were observed in the subsurface water along the PN section, representing by distinctive temperature of as low as 18°C and high salinity of over 34 PSU (Yang et al., 2018). The invasive KSSW water that was rich in nutrients resulted in increased SiO₃²⁻ and PO₄³⁻ concentrations in the subsurface seawater of P2-P6 below 30 m (Figure 2; Supplementary Table S1).

The hydrological conditions along PN section during winter were markedly different from those in summer (Figure 1). The reduction in Changjiang River discharge and the prevailing northeasterly wind in winter confined the influence of the Changjiang River water to a narrow band along the coast. Consistent with Yang et al. (2018), the seawater along PN section was well mixed from bottom to top due to the strong northeast monsoon (Figure 3; Supplementary Table S2). Although clear vertical variations of environmental parameters such as the temperature and salinity along water depths were observed in summer, the temperature, salinity, pH, DO and nutrients were more homogeneous above 100 m in P2-P7 during winter with no evident vertical variations. However, the typical KSSW was largely absent in winter (Yang et al., 2018), while Kuroshio surface water (KSW) and also TWC intruded along the PN section, resulting in the high salinity in P3-P7. Generally, the concentrations of NO₃⁻, SiO₃²⁻, and PO₄³⁻ were higher in the surface seawater samples of winter compared to those of summer (p < 0.05, Wilcoxon rank sum test), as indicated by previous studies (Wang et al., 2003; Zhang et al., 2007). The temperature, salinity and nutrient gradients along PN section indicated that these samples reflected typical hydrological conditions and environmental gradients in both summer and winter (Figure 2, 3; Supplementary Tables S1, S2). Distinct variations of other environmental factors between summer and winter seawater samples were also observed. In summer, the Chl a content was highest in P1 surface seawater under the influence of Changjiang River plume (7.7 μg/L), representing a location with high primary productivity, and the invasion of TCW and Kuroshio water led to lower Chl a concentration in P2-P7. Seawater from P1 also exhibited the lowest DO concentrations in summer. On the contrary, in winter, the Chl a content was lowest in P1 (0.14–0.20 μg/L), and was consistently higher in P2-P7 (0.17–0.54 μg/L) above 100 m. The DO content of P1 was the highest and reduced from P2-P7 in winter (Figure 3).

As shown by the flow cytometry results, the average counting of heterotrophic bacteria in summer was 7.92 ± 7.56 × 10⁸ cells L⁻¹ and 3.12 ± 2.37 × 10⁸ cells L⁻¹ in winter. The total microbial abundance along the PN section was also higher in summer (1.11 × 10⁸ copies L⁻¹ – 7.45 × 10⁸ copies L⁻¹) than those in winter (1.83 × 10⁶ copies L⁻¹ – 1.59 × 10⁸ copies L⁻¹) (p < 0.01, Mann–Whitney U test) (Figure 4). The reduction in total microbial abundance was more evident in winter samples from P3-P7 (1.47 ± 0.23 × 10⁷ copies L⁻¹) when compared with the summer samples from P3-P6 (3.28 ± 0.35 × 10⁸ copies L⁻¹) (p < 0.01, Mann–Whitney U test), indicating that less microbes were brought in to the ESC shelf along PN section by Kuroshio current in winter. Additionally, in winter, the microbial abundance of P3-P7 (1.83 × 10⁶ copies L⁻¹ – 7.61 × 10⁷ copies L⁻¹) was significantly lower compared to those of P1-P2 (3.62 × 10⁷ copies L⁻¹ – 2.04 × 10⁸ copies L⁻¹) (p < 0.01, Mann–Whitney U test). Vertically, the microbial abundance of P1 exhibited an increasing trend with water depth in both summer and winter; no apparent variation pattern with the water depth was observed in other PN sites. When considering the different microbial lifestyles, the abundance of FL bacteria (3.00 ± 0.30 × 10⁸ copies L⁻¹) was generally higher than that of PA (6.15 ± 0.73 × 10⁷ copies L⁻¹) in summer (p < 0.05, Independent Samples t-test), with exceptions in some specific samples (P1_30m, P4_0m). While in winter, the abundance of PA bacteria (2.01 ± 0.31 × 10⁷ copies L⁻¹) was slightly higher than that of FL (1.32 ± 0.22 × 10⁷ copies L⁻¹) (p < 0.05, Independent Samples t-test), with exceptions in pelagic regions (P6-P7).

A total of 59,594±15,349 reads in each sample were obtained and were clustered into 12,015 OTUs at a 97% similarity level. The diversity and richness of microbial communities along the PN section indicating by the Shannon and Chao 1 indices, respectively, presented distinct variation patterns between summer and winter (Supplementary Figure S1; Supplementary Tables S3, S4). In summer, the Shannon and Chao 1 indices remained at relatively stable levels along the PN section (p > 0.05, Wilcoxon signed-rank test); both Shannon and Chao 1 were slightly higher in PA communities when compared to the FL counterparts. While in winter, the Chao 1 index showed a clear decreasing trend from P1 to P7 in both PA and FL communities (p < 0.05, Wilcoxon signed-rank test, Supplementary Figures S1C,D); the Shannon index also decreased significantly from P4-P7, especially in the PA communities of these offshore samples (p < 0.05, Wilcoxon signed-rank test). The community distribution pattern revealed by Nonmetric multidimensional scaling (NMDS) analysis showed a shift of microbial communities in different seasons (Figure 5A), with separations of FL and PA samples in both summer (p < 0.001, Mann–Whitney U test) and winter (p < 0.05, Mann–Whitney U test). Additionally, the winter communities presented a more evident trend of spatial shift along the PN section (from P1-P7, Figure 5C) compared with the summer communities (Figure 5B).

Based on the distinct seasonal patterns of microbial abundance and diversity shown above, we further investigated the dominant microbial communities along the PN section in summer and winter. In summer, the dominant microbial classes were similar in FL and PA fractions, including Alphaproteobacteria (30.94% in FL and 30.89% in PA), Gammaproteobacteria (16.34% in FL and 16.36% in PA) and Flavobacteria (13.12% in FL and 10.65% in PA), Cyanobacteria (9.54% in FL and 11.81% in PA) and Actinobacteria (10.84% in FL and 5.97% in PA) (Figures 6A,B). However, more apparent differences were shown at the genus level between summer FL and PA communities (Figure 7; Supplementary Figure S3C). In the FL samples, genera belonging to SAR11 clade, Actinobacteria and Marine Group I were most abundant, and Candidatus Nitrosopelagicus in Thaumarchaeota showed high abundance in several subsurface water samples (30-80 m) (Figure 7A; Supplementary Figure S3C). While in the PA samples (Figure 7B), Synechococcus (6.53%), Alteromonas (4.04%), Sulfitobacter (3.03%) and unidentified genera in Cyanobacteria (4.92%) and Rhodobacteraceae (4.54%) were highly abundant. Consistent with the NMDS results (Figure 5B), the dominant microbial communities showed no evident differences from nearshore and offshore along the PN section in summer.

In winter, the microbial community composition along the PN section was featured by a dominance of bacteria belonging to Ralstonia (Betaproteobacteria), which accounted for 12.49% in FL and 56.69% in PA samples (Figures 7C,D; Supplementary Figure S3D). In the FL fractions, these Ralstonia mainly exist in P1 (11.75%), P2 (23.18%), and P7 (41.82%) with the abundance increased with water depth. While in the PA fractions, the abundance of Ralstonia slightly increased from P3 (62.14%) to P7 (81.67%), reaching as high as 90.28% in P7_750m. Except Betaproteobacteria (mostly Ralstonia), Alphaproteobacteria (26.54% in FL and 7.80% in PA) and Gammaproteobacteria (15.54% in FL and 9.44% in PA) were the abundant groups in winter samples as in summer (Figures 6C,D, 7C,D). The abundance of Flavobacteria, Actinobacteria and Cyanobacteria were lower in winter compared with those in summer (Supplementary Figure S2A).

db-RDA based on Bray-Curtis distances was used to investigate the influence of environmental parameters on microbial community compositions along the PN section (Figure 8). In general, depth, salinity, temperature, DOC, pH, DO, pH, SiO₃²⁻, PO₄³⁻ NO₃⁻, and SYN were correlated with the distribution of microbial communities along the PN section in summer. However, except for the above environmental parameters, latitude and longitude also significantly affected the winter microbial communities, while SYN showed no effect.

The Mantel test results indicated that stronger correlations between various environmental factors and winter communities than summer communities along the PN section (Figure 8; Supplementary Table S5). The changes of summer FL communities were significantly correlated with temperature, depth and, SiO₃²⁻ (r > 0.4, p < 0.001), and weakly correlated with salinity, pH, DO, and PO₄³⁻ (r > 0.2, p < 0.001). However, only temperature and pH showed weak correlations with the summer PA communities (r < 0.2, p < 0.001). In winter, the FL communities were significantly correlated with depth, temperature, DO, PO₄³⁻, SiO₃²⁻, NO₃⁻, pH, longitude and latitude (r > 0.2, p < 0.001), whereas salinity, longitude, latitude and DOC were significantly correlated with the PA communities (r > 0.4, p < 0.001).

The Spearman correlations between the bacterial abundance and the environmental parameters were calculated (Supplementary Table S6). In summer, the abundance of both total (r = 0.491, p < 0.05) and FL bacteria (r = 0.426, p < 0.05) showed significant positive correlation with Chl a. The abundance of summer PA bacteria was positively correlated with PO₄³⁻ (r = 0.404, p < 0.05) and NO₃⁻ (r = 0.708, p < 0.01) and negatively correlated with pH (r = −0.460, p < 0.05) and DOC (r = −0.636, p < 0.01). In winter, the abundance of total bacteria was positively correlated with NO₃⁻ (r = 0.401, p < 0.05). In particular, the abundance of PA bacteria showed significant positive correlation with latitude (r = 0.491, p < 0.01) and DO (r = 0.457, p < 0.01), and negatively correlated with longitude (r = −0.491, p < 0.01) and salinity (r = −0.457, p < 0.01). In addition, we found the dominant Ralstonia in winter PA samples was positively correlated with longitude, depth, salinity and temperature, while chemical parameters such as DO, SiO₃²⁻, PO₄³⁻, NO₃⁻, and DOC were found to be negatively correlated with Ralstonia (Supplementary Table S7).

The βNTI analyses based on the null model revealed a clear difference of dominant processes in structuring the microbial communities along the PN section between winter and summer (Figure 9). In summer, the stochastic process represented by dispersal limitation was the most important process in either FL and PA communities (56.3 and 57.8%, respectively), while in winter, the major process turned out to be the deterministic process, homogeneous selection (51.5% in FL and 64.5% in PA). Additionally, heterogeneous selection accounted for higher proportions in determining PA communities (5.8%) compared to the FL communities (1.9%) in summer. Although selection was the most important structuring process in winter, the contribution of dispersal limitation was higher for the FL (13.7%), whereas homogenizing dispersal was slightly higher for the PA (3.9%) among the stochastic processes.

To study the microbial co-occurrence patterns along the PN section, we built co-occurrence networks for samples from different seasons and different lifestyles (Figure 10). Generally, the summer network exhibited lower average degree and average clustering coefficient compared with those of the winter network, whereas the average network distance and modularity was higher in summer than in winter. These results suggested that the microbial communities tended to be more connected in winter than in summer along the PN section. Specifically, the network of winter PA communities presented distinct topological properties compared with either summer communities or winter FL counterparts, showing lower node numbers and extremely high edge numbers (Supplementary Table S9). Correspondingly, higher average degree and clustering coefficient was shown by the winter PA network, while its modularity was lower (0.099). The OTUs from Alphaproteobacteria, Gammaproteobacteria, Flavobacteria, Actinobacteria, and Cyanobacteria displayed more correlations with others in these networks (Figure 10; Supplementary Table S9). Keystone OTUs in each network with high degree (>25) and low betweenness centrality (<262.32) were identified. Interestingly, the keystone OTUs in winter were dominated by Alpha- and Gammaproteobacteria (100% for FL and 89% for PA), while high proportions of archaeal keystone OTUs (Thaumarchaeota and Euryarchaeota) were identified in summer microbial networks (43% for FL and 28% for PA) (Supplementary Table S8).

We further built separate co-occurrence networks for nearshore and offshore PN section samples, respectively, which were influenced by the distinct hydrographic conditions (Supplementary Figure S4). In both seasons, the networks of offshore FL samples possessed more edges and higher average degree than the nearshore FL samples, while the networks of offshore PA samples were with less edges and lower average degree than the nearshore samples (Supplementary Table S10). This indicated that the FL microbial networks were more complex in offshore sites, whereas the nearshore PA microbial networks were more complex in nearshore sites along the PN section. Additionally, the complexity of the winter PA microbial network at P5-P7 was significantly reduced, with less nodes, edges and lower average degree, possibly due to the high abundance of Ralstonia species. In the networks of winter samples, we found these Ralstonia exhibited mainly positive correlations with some Alphaproteobacteria and Gammaproteobacteria in P1-P2, while more negative correlations between Ralstonia and specific Alphaproteobacteria, Gammaproteobacteria and Cyanobacteria were found in P5-P7 (Supplementary Table S11), indicating that the dominance of Ralstonia may inhibit these bacteria communities.