Homogeneous selection shapes free‐living and particle‐associated bacterial communities in subtropical coastal waters

In microbial biogeography, it is crucial to link spatial patterns with underlying drivers in natural ecosystems. Bacterial communities driving key biogeochemical processes in coastal zones, which are important interfaces between terrestrial and marine ecosystems, are affected by perturbations due to both natural and anthropogenic factors. However, the assembly of bacterioplankton communities, either free‐living (FL) or particle‐associated, in coastal ecosystems is still poorly understood.


| INTRODUC TI ON
A central aspect of microbial biogeography is to link spatial patterns with underlying drivers in natural ecosystems (Hanson et al., 2012;Langenheder & Lindström, 2019;Martiny et al., 2006;Nemergut et al., 2013;Zhou & Ning, 2017). According to the metacommunity concept (Leibold et al., 2004), deterministic and stochastic processes are primarily responsible for the spatial turnover of natural communities (Vellend, 2010). The term "deterministic process" refers to two types of selective forces, namely, those that lead to either more (i.e. homogeneous selection) or less (i.e. heterogeneous selection) similar structures among communities due to homogeneous and heterogeneous environmental pressures, respectively (Zhou & Ning, 2017).
Meanwhile, the term "stochastic process" refers to homogenizing dispersal, dispersal limitation (combined with drift) and pure drift, which can obscure the turnover among microbial communities due to high dispersal; low dispersal; and random changes in birth, death and reproduction, respectively (Zhou & Ning, 2017). The relative importance of diversification, occurring on a large spatial and temporal scale, is disregarded because it is usually minor (Stegen et al., 2013. Marine bacterioplankton have traditionally been classified into two types of communities: free-living (FL) and particle-associated (DeLong et al., 1993), both of which play fundamental roles in marine ecosystems (Azam & Long, 2001). These two categories, however, exhibit significant differences, both in terms of composition and function (D'Ambrosio et al., 2014). For example, FL bacteria are generally smaller than their particle-associated counterparts (Dang & Lovell, 2016), whereas particle-associated bacteria exhibit higher production (Crump et al., 1998) and carbon demand (Lapoussière et al., 2011). Particle-associated bacteria are highly effective at converting high-molecular weight organic matter into smaller substrates, which are bioavailable to their FL cohorts (Arnosti, 2011;Azam & Long, 2001). The truly planktonic and sessile states of marine bacteria allow two entirely different lifestyles owing to differences in phylogenetically conserved traits (Salazar et al., 2015) in their respective genomes (Lauro et al., 2009;Smith et al., 2013). Importantly, these differences can largely mediate the balance between stochastic and deterministic processes (Xu, Zhang et al., 2020). Although many previous studies have described the community compositions of FL and particle-associated bacteria, how bacteria structure their ecological processes remains to be elucidated.
Coastal ecosystems harbour complex bacterial communities (Crump et al., 2004), which mirror the simultaneous influences of both natural and anthropogenic perturbations (Aguirre et al., 2017;Jeffries et al., 2016). One of the most salient characteristics of coastal ecosystems is the natural heterogeneity of physiochemical conditions, including light, temperature, salinity and nutrients, which are influenced by factors such as intermixing of freshwater and seawater, as well as transportation of dissolved and suspended matter. Prominent heterogeneities, such as differences in salinity, can acutely regulate bacterial communities (Lozupone & Knight, 2007). In addition, coastal ecosystems are intensely influenced by human activities, resulting in increased levels of nutrients (Cloern et al., 2016;Halpern et al., 2008) and hydrocarbons (Duran & Cravo-Laureau, 2016); these pollutants cause spatial heterogeneities in bacterial communities (Crump et al., 2004;del Giorgio & Bouvier, 2002;Kirchman et al., 2005). Therefore, it is important to improve our understanding of the bacterial communities in coastal ecosystems such as the coastal South China Sea influenced by the Pearl River (SCSPR), to reflect changes occurring in coastal waters globally (Rabalais et al., 2009). The Pearl River is the second largest river in China in terms of annual discharge (336 km 3 ), and the region around its delta is highly urbanized and industrialized (Zhang et al., 2007). As a consequence, the SCSPR is susceptible to intense perturbations caused by natural processes and anthropogenic activities (Zhang et al., 2007). This is especially obvious during the wet season, between April and September, when ~ 80% of the annual runoff occurs. The large input of freshwater is accompanied by heavy loading of nutrients, causing eutrophication, which is a major concern as it leads to the formation of a hypoxic bottom layer (Rabalais et al., 2009). The SCSPR is therefore a model system for studying threatened coastal ecosystems (Fennel & Testa, 2019), where changes in bacterial communities may precede detrimental effects on macroorganisms (Spitz et al., 2016).
In this study, we investigated FL, nanoparticle-associated (NA) and microparticle-associated (MA) bacterial communities at the surface and in the bottom layer of the SCSPR during the summer wet season. Bacterial communities were identified using environmental DNA metabarcoding based on the V4 region of 16S rRNA gene. We evaluated the underlying mechanisms of community assembly using null model analyses and identified a set of core bacteria, comprising potentially hydrocarbon-degrading members, which indicated acute perturbations due to anthropogenic activities. Altogether, the findings of this study provide a better understanding of the assembly of bacterial communities in subtropical coastal waters.

| Sample collection
Between 10 July and 21 July 2017, water samples for molecular analyses were collected from the surface and bottom layers at 19 stations in the SCSPR (Figure 1) using Niskin bottles mounted on an SBE 32 carousel water sampler (Sea-Bird Electronics). Seawater (350-1,000 ml) was pre-filtered through a 200 μm mesh and then sequentially through 20-, 3-and 0.2-μm-pore size polycarbonate membranes (47 mm diameter, Millipore). The size continuum of 0.2-3, 3-20 and 20-200 μm represents the collections of FL, NA and MA bacterial communities, respectively (Andersen et al., 2016;Cui et al., 2019). The filters were stored at −20°C onboard and at −80°C for post-cruise analyses in the laboratory. Ten MA communities were derived from filters immersed in RNAlater solution (Ambion) or amplified using additional thermal cycles (see procedures below) due to difficulties in molecular analyses (Table S1).
The temperature, salinity and turbidity were measured using a conductivity-temperature-depth (CTD) profiler (Sea-Bird Electronics). Chlorophyll a concentration was measured using 90% acetone extractions of GF/F membrane-filtered water samples with a fluorometer (Turner Designs). The concentration of dissolved oxygen was determined onboard by the Winkler titration method, using a UV-1800 spectrophotometer (Shimadzu). The numbers of cyanobacterial, heterotrophic bacterial and pigmented eukaryotic cells were counted using a FACSCalibur flow cytometer (BD Biosciences) as described by Chen et al. (2009). Other parameters, including dissolved organic carbon (Dai et al., 2009), nutrients (NO 2 , NO 3 , PO 4

and
SiO 3 ) and total suspended matter, were also measured using standard protocols . Pearson's correlations and principal component analyses of standardized environmental variables (with a mean of 0 and a variance of 1 for each variable) were visualized using the corrplot (Wei & Simko, 2017) and vegan (Oksanen et al., 2018) packages, respectively, in R (R Core Team, 2018). Variables showing high Pearson's correlations (<−0.8 or >0.8) with other variables were excluded from the principal component analysis (i.e. temperature, dissolved organic carbon concentration, NO 3 , and SiO 3 ; Figure S1).

| DNA extraction and metabarcoding
Each filter was cut into pieces and then transferred into a column in the FastDNA Spin Kit (MP Biomedicals). Sodium phosphate buffer (978 μl) and MT buffer (122 μl) were added, and the mixture was homogenized using a Mini-Beadbeater-24 (Biospec Products) at a speed of 3,500 oscl min −1 . All subsequent steps for total DNA extraction were carried out according to the manufacturer's instructions.

| Diversity estimation
Faith's phylogenetic diversity (PD), Shannon diversity and Pielou's evenness were estimated (based on 100 bootstraps) using the picante (PD) and vegan (Shannon diversity and Pielou's evenness) packages in R. Differences in the means among the FL, NA and MA fractions were evaluated using one-way ANOVA, and the pairwise groups were further evaluated using Tukey's HSD test. In addition, abundant ASVs were defined as ASVs accounting for >1% of the relative abundance within any sample (Wu et al., 2017). Ternary plots were used to visualize the preference of abundant ASVs among the FL, NA and MA communities (Smith, 2017), which was indicated by orientation towards a corner using a threshold value of >60%.
Unweighted and weighted UniFrac distances were calculated with the rarefied ASV tables using the GUniFrac package in R (Chen et al., 2012). Principal coordinate analysis (PCoA) was performed on the weighted UniFrac dissimilarity, and ellipses were generated with 95% confidence intervals using the car package (Fox & Weisberg, 2019).
We analysed the distance-decay of community similarity relationship, which represents the decrease in similarities among communities (1 -community dissimilarity) with an increase in geographical distance (Green & Bohannan, 2006). Weighted UniFrac dissimilarity was used because relative abundances are informative for community structuring (Anderson et al., 2011). For samples taken from the surface and bottom layers at the same station, pairwise geographical distances were estimated based on the difference in water depth, which was ignored for estimations at different stations.
Constrained analysis of principal coordinates (CAP) using the vegan package allowed us to relate weighted UniFrac dissimilarities (standardized) with environmental variables. Prior to CAP analyses, forward selection (999 permutations) was run, followed by post hoc fittings using the envfit function (999 permutations) to select environmental variables that significantly accounted for dissimilarities among communities.

| Null model analysis
To determine whether phylogenetic turnover can be used to infer ecological processes (Stegen et al., 2013), we tested phylogenetic signals using Mantel correlograms as previously described (Dini-Andreote et al., 2015). Briefly, we calculated the relative abundance-weighted means of ASVs for each environmental variable.
The resulting mean for each ASV represented its environmental optima (i.e. ecological niches). Differences in environmental optima between ASVs were represented as Euclidean distances using normalized axes. Phylogenetic distances between ASVs were calculated using FastTree with the picante package in R (Kembel et al., 2010). Pearson's correlation coefficients of differences in environmental optima and phylogenetic distances were quantified based on Mantel correlograms using the vegan package in R (mantel.correlog function, 50 phylogenetic distance classes and 999 permutations associated with progressive Bonferroni correction).
We analysed the phylogenetic structure of local communities to obtain insights into the drivers of community assembly. Because of the possibility that conservatism in bacterial traits occurs mostly at terminal levels in the phylogeny, we calculated the nearest taxon index (NTI) that quantified the terminal structure of phylogenetic clustering and overdispersion (Webb et al., 2002) using the picante package in R. NTI measures the deviation of observed mean nearest taxon distance (MNTD) from the null expectation (taxa.labels model, 999 randomizations). An NTI of >+2 indicates that the ASVs in a local community are more closely related than expected by chance, suggesting the role of selective pressures (e.g. environmental conditions) in phylogenetic clustering. An NTI of <−2 represents phylogenetic overdispersion, indicating two possible biotic interactions: competition and facilitation. In contrast, a mean NTI across multiple communities that is significantly greater or less than zero indicates phylogenetic clustering or overdispersion, respectively (Zhou & Ning, 2017).
We further assessed the relative importance of heterogeneous selection, dispersal limitation (combined with drift), drift (acting alone), homogenizing dispersal and homogeneous selection using a two-step framework (Stegen et al., 2013). In the first step, we calculated intercommunity phylogenetic turnover using the mean nearest taxon distance (βMNTD). Then, we estimated the β-nearest taxon index (βNTI), representing the difference between the observed βMNTD and the mean null expectation in units of standard deviation (999 randomizations). βNTI>+2 or <−2 signified heterogeneous selection or homogeneous selection, respectively. In the second step,

| Core ASV analysis
Core microbiomes, which are crucial for community functions (Shade & Handelsman, 2012), are commonly defined based on cut-off values of shared taxa occurrence across communities (Hernandez-Agreda et al., 2017). We defined core ASVs separately in the FL, NA and MA fractions using a 100% threshold (i.e. present in all 38 samples in each group). For phylogenetic analyses, sequences of core ASVs were aligned using MAFFT, and a maximum likelihood tree was constructed using a model recommended by MEGA X (Kumar et al., 2018). Moreover, Levin's habitat niche breadths of core ASVs were calculated using the spaa package in R with the following formula (Zhang, 2016): where B j represents the habitat niche breadth of ASV j and P ij is the proportion of ASV j in community i among a total of N communities. proxies of freshwater and marine habitats) and sampling depth/ temperature (serving as proxies of surface and bottom habitats), respectively ( Figure 4). All the three types of communities displayed significant (p < .001) distance-decay relationships ( Figure S3). Based on phylogenetic analyses, a total of 13 core ASVs were closely related to hydrocarbonoclastic bacteria (Figure 5d).

| Community assembly
Significant phylogenetic signals (p < .01) were observed across relatively short phylogenetic distances (<20% of the maximum phylogenetic distance) ( Figure S4

| Dominance of homogeneous selection
Our results suggested that homogeneous selection could be the most important ecological process in shaping bacterial communities across the FL, NA and MA fractions (Figure 6b). This scenario is somewhat unexpected in the perturbed SCSPR, characterized by marked environmental heterogeneities (Figure 1), since highly variable factors such as salinity can heterogeneously select coastal bacterial communities (Campbell & Kirchman, 2013). In contrast, ho-  (Figures 1b and S1).
Moreover, eutrophication levels of coastal ecosystems can adversely affect the phylogenetic structure of local communities (Dai et al., 2017). Correspondingly, most phylogenetic structure of the local communities in this study were closely clustered (Figure 6a), indicating that the bacterial communities have undergone heavy selection (Zhou & Ning, 2017). This phylogenetic clustering is, therefore, consistent with the dominance of homogeneous selection at the regional scale (Figure 6b). In line with our finding, recent studies have reported temporal dominance of homogeneous selection during phytoplankton blooms in a eutrophic coast (Zhang et al., 2018) and a eutrophic urban lake (Xu, Zhao et al., 2020).
We further propose that eutrophic conditions can impose homogeneous selection on potentially native bacterial communities in the SCSPR. First, we suggest the existence of native bacterial communities in the SCSPR rather than a simple mixture of freshwater and seawater communities (Crump et al., 2004). The productive SCSPR harbours bacterial communities with high growth rates; however, these bacterial communities have longer residence periods in the SCSPR compared to riverine systems due to the largely reduced level of directional transportation. These two conditions facilitate the formation of native coastal bacterial communities (Crump et al., 2004). Notably, native bacterial communities can adapt to copiotrophic environments, even those with long-term disturbances (Xiong et al., 2015), through a combination of genomic architecture (Lauro et al., 2009) and metabolic strategies such as nutrient costs (Grzymski & Dussaq, 2012), leading to a consistent change in the compositions of communities across large gradients (Fodelianakis et al., 2014). Thus, homogeneous selection can be enhanced in potentially native bacterial communities because wide adaptions to eutrophic conditions strengthen community structures rather than generate variations.

| Community assembly across multiple fractions
Free-living, NA and MA bacterial communities share remarkable similarities in their community assembly patterns, as demonstrated by the dominance of homogeneous selection (Figure 6b). Suspended particles in coastal ecosystems are usually old (i.e. mineral-rich) and mainly comprise inorganic matter of terrestrial and sediment origins (Dang & Lovell, 2016). The nature of these particles may Acinetobacter sp. ONE4, KU353547

Moraxella osloensis NCTC 10465, KT989843
Pseudoalteromonas shioyasakiensis SO240BG28, MK254673 Alteromonas macleodii PRG12 94, MN723607 Halomonas aquamarina 2B, MN726518 Halieaceae bacterium LSUCC0673, MK603669 Halieaceae bacterium LSUCC0552, MK603636 Halieaceae bacterium LSUCC0757, MK603709 Uncultured gammaproteobacterium clone S1-8-02, KF786735 Uncultured SAR86 clone HF10 32C08, EU361700 Uncultured gammaproteobacterium clone S2-1-137, KF786856 Uncultured bacterium clone S72 002, KC874251 Uncultured proteobacterium clone NA1 1128782, KM277184 Uncultured Rhodospirillaceae bacterium clone S1-1-72, KF786382 Uncultured alphaproteobacterium clone S1-1-15, KF786357  largely minimize the effect of particle colonization, which results in different assemblies among FL, NA and MA communities (Dang & Lovell, 2016), unlike organic content-rich particles in pelagic waters with largely differentiated FL and particle-associated bacterial communities (Ortega-Retuerta et al., 2013). Moreover, it has been reported that FL and particle-associated bacterial communities likely interact with each other through their attachment to and detachment from diverse particles (Kiørboe et al., 2003). However, the pattern of community assembly was also considerably different across the different fractions ( Figure 6b). First, dispersal limitation among the FL, NA and MA communities is increased, likely because water movement-driven passive dispersal is less limited in FL bacteria compared to that in particle-associated bacteria (Villarino et al., 2018). Second, the chemical properties of particles are diverse , and thus, particles have higher substrate availabilities (i.e. niches) in their microenvironments than in the FL habitat (Mestre et al., 2017), resulting in significantly higher PD estimates in NA and MA than in FL communities ( Figure 2c) (Xu, Zhao et al., 2020). At the same time, locally enriched substrates can largely increase the relative abundance of favoured groups, leading to uneven community structures. Therefore, the three fractions showed non-significant differences in their proportional Shannon diversities (Figure 2d), despite the significantly higher PD estimates of NA and MA communities. In particular, MA communities exhibited significantly lower evenness than both FL and NA (Figure 2e).
The large differences in PD and evenness among MA communities indicate the occasional presence/absence of many taxa in some microparticles due to distinct substrate availabilities, which tends to trigger drift (Figure 6b) (Xun et al., 2019). Taken together, deterministic (mainly represented by homogeneous selection) and stochastic (mainly represented by dispersal limitation and drift) processes decrease and increase, respectively, across the FL, NA and MA fractions.

| Core ASVs
The core ASVs identified in this study indicate that anthropogenic activities may markedly influence the assembly of bacterial communities in the SCSPR ( Figure 5). In particular, the most abundant core ASV01 shares 100% similarity with the strain Acinetobacter venetianus BU71, isolated from the surface water of petroleum-contaminated coastal sites ( Figure 5d) (Mahjoubi et al., 2013). There were nine core ASVs belonging to hydrocarbonoclastic bacteria or taxa with cooperative interactions, which are closely related to taxa detected in an oil sheen in the Gulf of Mexico (McGenity et al., 2012).
Obligate hydrocarbonoclastic bacteria have long been recognized for their roles in the removal of petroleum hydrocarbons from polluted coastal waters (Yakimov et al., 2007). The influx of hydrocarbons has been a major concern in the SCSPR, both in the waters (Wang et al., 2007) and sediments (Chen et al., 2006), especially during the wet season (Wang et al., 2007). Crude oil represents the most complex mixture of organic compounds on Earth (Head et al., 2006), and thus, hydrocarbonoclastic bacteria have evolved a versatile range of metabolic functions (Mason et al., 2012). Hydrocarbonoclastic bacteria are widely found in the SCSPR (Wu et al., 2014), suggesting that the SCSPR is environmentally favourable for these bacteria.
Importantly, the core ASVs accounted for large proportions of the whole communities (Figure 5c,d). For example, the most abundant core ASV01 had remarkable relative abundances (FL, 6%; NA, 6.3%; MA, 9.1%) (Figure 5d), indicating that core ASVs can influence responses to coastal environments. Notably, high relative abundances and widespread occurrence of core ASVs reduce dissimilarities among communities and contribute to the relative importance of homogeneous selection as a primary mediator of community assembly (Östman et al., 2010). According to the threshold value (i.e. >3) of habitat niche breadths (Logares et al., 2013) (Figure 5b) just a few days (Bunse & Pinhassi, 2017;Osterholz et al., 2018 (Yakimov et al., 2007). Since our samples were collected immediately before the tropical storm Roke (July 21-23, 2017), we speculate that a combination of pre-storm conditions (e.g. low wind speed and intense sunlight) might have promoted the occurrence of oil-degrading bacteria (Bacosa et al., 2015) and contributed to the formation of temporally unique bacterial communities in the SCSPR (Bryant et al., 2016). Therefore, we suggest that more efforts are urgently needed to adequately understand the mechanism underlying the assembly of bacterial communities in complex coasts.

| CON CLUS IONS
In this study, the observed ASVs were most abundant in the FL, followed by the NA and MA communities, in the SCSPR. Approximately half of the abundant ASVs showed no preference among the three fractions. The detection of a few core ASVs, which are closely related to oil-degrading bacteria, indicates the effect of anthropogenic activities on bacterial communities. The FL, NA and MA communities differed significantly but weakly. Importantly, homogeneous selection might have been the primary factor shaping the bacterial communities during the sampling time, although more investigations are needed to achieve a better understanding of bacterial community assembly in the SCSPR. Moreover, subdividing the whole community into FL, NA and MA fractions is helpful for revealing new information about their underlying assembly mechanisms; we observed that deterministic processes (mainly represented by homogeneous selection) were most predominant in FL, followed by NA and MA communities, whereas stochastic processes (mainly represented by dispersal limitation and drift) exhibited the reverse order.
In summary, this study provides an unexpected snapshot of bacterial communities in the SCSPR, filling knowledge gaps regarding bacterial community assembly. The results of this study may inspire future studies focused on anthropogenic influences on coastal ecosystems. [Correction added on 18 November 2020, after first online publication: the Acknowledgements have been updated.]

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ddi.13193.

DATA AVA I L A B I L I T Y S TAT E M E N T
Raw sequences have been deposited in the Sequence Read Archive Homogeneous selection shapes free-living and particleassociated bacterial communities in subtropical coastal waters.