The research was conducted in the Xiashan constructed wetland near the Pingshan River in Shenzhen, Guangdong Province, China (114.34°E, 22.69°N). Xiashan is a subsurface constructed wetland with a treatment scale of 31000-40000 m³/d, and the influent of the wetland is the tail water of the Shangyang Wastewater Treatment Plant. The following plants had been planted and were growing well in the wetland: I. ensata, C. indica, H. littoralis, Thalia dealbata, Cyperus papyrus, and Cyperus alternifolius. From top to bottom, the substrate of the wetland consisted of A-type filler with a thickness of 900 mm (slow-release C source: activated C: oyster shell: Zeolite: sand = 1:3.9:10.95:33.15:114.91); crushed stone (diameter: 4-8 mm) with a thickness of 300 mm; and crushed stone (diameter: 16-32 mm) with a thickness of 300 mm. At the time of this research, the wetland system had been running stably and continuously for nearly 2 years and had provided substantial water purification (Gao et al., 2020). The area has a south subtropical monsoon climate and the annual temperature ranging from 5.3 - 36.6°C in 2021.

Three plant species (i.e., H. littoralis, I. ensata, and C. indica) were selected, because they have been shown to have good contaminant removal capabilities (Singh et al., 2025). In 2021, we sampled three wetland ponds with H. littoralis, I. ensata, and C. indica in January (i.e. dry season), and sampled five wetland ponds with the same three species in July (i.e. wet season). Each pond covered an area of approximately 1245 m². Plant roots were not contaminated by bulk soil in the constructed wetland, so we uprooted softly three plant samples at the five locations (the four corners and the center) at each pond, and the corresponding soil samples were collected using an ethanol sterilized shovel. After the mixed samples were placed in sterile Ziploc bags, they were transported to the laboratory. We then gently scrapped off a thin layer of soil (sand, 0-5 mm) from the roots and mixed the samples from the five locations to obtain one sample of rhizosphere soil per pond (Pei et al., 2018). Rhizoplane material was separated from plant roots after washing 3x in sodium-free phosphate buffer and filtered with a 0.22 µm polycarbonate filter prior to DNA extraction. The endosphere was represented by the remaining root material, i.e., roots minus rhizosphere soil and rhizoplane. Each sample of rhizosphere soil or rhizoplane and endosphere root material was represented by about 5 g of material.

As indicated, we collected three samples of the rhizosphere, rhizoplane, and endosphere from each of the three ponds for each of the three plant species. This yielded 27 samples: 3 plant species × 3 ponds/plant species × 3 root components (rhizosphere, rhizoplane, and endosphere)/pond/plant species. The samples were stored at -80°C and were then sent to Beijing Biomarker Biotechnology Co., Ltd. under low temperature conditions for microbial high-throughput sequencing.

A soil DNA extraction kit (MN NucleoSpin 96 Soi) was used to extract DNA from rhizosphere, rhizoplane, and endosphere samples of the three plant species. The V3+V4 region of the bacterial 16S rRNA gene was amplified using primers 338F/806R (5’-ACTCCTACGGGAGGCAGCA-3’/5’-GGAC-TACHVGGGTWTCTAAT-3’), and endophytic universal primers 335F/769R (5’-CADACTCCTACGGGAGGC-3’/5’-ATCCTGTTTGMTMCCCVCRC-3’). PCR reactions were carried out with 50 ng/uL ± 20% of genomic DNA, 5 μL of KOD FX Neo Buffer, 0.2 μL of KOD FX Neo, 2 μL of 2dNTP, and sufficient ddH₂O to increase the total volume to 10 μL. The PCR included an initial denaturation at 95°C for 5 min; followed by 25 cycles of 95°C for 30 s, 50°C for 30 s, and 72°C for 40 s; and a final extension at 72°C for 7 min. After 1.8% agarose gel electrophoresis (120 V for 40 min), the target fragment was cut and recovered. The products were purified, quantified, and homogenized to form a sequencing library. The library was first inspected for quality, and the qualified library was subjected to bidirectional sequencing using an Illumina HiSeq 2500 (Wright and Vetsigian, 2016).

The original data were spliced (FLASH, version 1.2.11) (Magoc and Salzberg, 2011), and the spliced sequences were filtered by quality (Trimmomatic, version 0.33) (Bolger et al., 2014). The chimera (UCHIME, version 8.1) were then removed to obtain high-quality tag sequences (Edgar et al., 2011). Sequences were clustered at a 97% similarity level (USEARCH, version 10.0) (Edgar, 2013) and then used 0.005% of all sequences sequenced as a threshold to filter OTU (Bokulich et al., 2013). Based on Silva SSU and LSU databases 138 (Quast et al., 2013), we annotated OTU using the RDP Classifier software (version 2.2, confidence threshold 0.8) (Wang et al., 2007) to derive the species classification for each OTU. We then counted OTUs to determine the community composition of each sample.

The water streams are already filtered and passed through secondary treatment (e.g., aeration and clarification) before being discharged into the constructed wetland. Water samples were collected from the water inlet and outlet of the three ponds, three replicates were collected from each inlet and outlet. All of the water samples were transported to the laboratory for chemical analyses. Chemical oxygen demand (COD) was measured using a spectrophotometer (DR/2010, Hach Co., Loveland, CO, USA). Total nitrogen (TN), total organic carbon (TOC), NO₃ ⁻-N, NO₂ ⁻-N, NH₄ ⁺-N, and total phosphorus (TP) were analyzed according to standard methods (APHA, 1998). For these pollutants, the removal efficiencies for each pond were calculated from the difference in concentration between the water inlet and outlet.

The OTU data were clustered at a 97.0% similarity level after standardization. The 10 most abundant genera in the root components of the three plant species were analyzed using R 3.2.5. The statistical power analysis has been conducted to ensure that the sample size is large enough in data analysis. The α diversity indices (ACE, Chao1, Shannon-Wiener) were calculated using Mothur version v.1.30 (http://www.mothur.org/) and SPSS21.0 software, and were visualized with Origin software (Shao et al., 2016). ACE and Chao1 indices were used to assess bacterial community richness. The Shannon-Wiener index was used to assess bacterial community diversity. To investigate the patterns of bacterial community structure, β diversity of the bacteria was assessed. We also performed nonmetric multidimensional scaling (NMDS) with the weighted_UniFrac distance calculated from the OTU community matrix (Mitter et al., 2017). Biomarker analysis with linear discriminant analysis effect size (LEfSe) (http://huttenhower.sph.harvard.edu/lefse/) was used to identify bacteria abundance differed among plant species and root components (Chen et al., 2020). After that, PICRUSt (phylogenetic investigation of communities by reconstruction of unobserved states) was used to predict the potential functions of the OTUs (relative abundance > 1%) in the rhizosphere vs. the rhizoplane and in the rhizoplane vs. the endosphere based on the KEGG (Kyoto Encyclopedia of Genes and Genomes) database (Langille et al., 2013). Redundancy analysis (RDA) was then performed using CANOCO 5.0 (Microcomputer Power, Ithaca, NY, USA) to assess the relationships between water physiochemical properties and species composition of biomarkers (Morris and Blackwood, 2015).

The pollutant removal efficiencies of three plant species varied significantly between two seasons ( Figure 1 ). During the wet season, plants exhibited heightened pollutant removal capacities compared to the dry season. Surprisingly, the pollutant removal efficiencies of I. ensata to purify NO₃ ⁻-N weakened in the wet season. In the dry season, I. ensata exhibited significantly higher pollutant removal efficiencies for NO₂ ⁻-N, NO₃ ⁻-N, and NH₄ ⁺- N compared to the other two plant species. In the wet season, I. ensata had the highest pollutant removal efficiency for COD and TN, while C. indica had the highest pollutant removal efficiency for NO₃ ⁻-N. H. littoralis had a significantly higher pollutant removal efficiency for TOC and TP than the other two plant species in both seasons.

At the phylum level, significant differences were observed in the bacterial composition among different root components and between two seasons ( Figure 2 ). Proteobacteria had the highest relative abundance in the root among three plant species (over 40% except in the rhizoplane in the wet season). Alphaproteobacteria and Gammaproteobacteria were the predominant bacteria within the Proteobacteria phylum among the three plant species, collectively constituting an average relative abundance as high as 50% ( Supplementary Figure S1 ). Caynobacteria, Bacteroidetes and Acidobacteria were also had large relative abundance. Acidobacteria had the second largest relative abundance in the Rhizosphere, Caynobacteria had the second largest relative abundance in the Rhizoplane, while Bacteroidetes had the second largest relative abundance in the Endosphere in the dry season. Bacteroidetes had the second largest relative abundance in the Rhizosphere, and Acidobacteria had the second largest relative abundance in the Endosphere in the wet season. Caynobacteria of the Rhizoplane increased significantly in the wet season than that in the dry season.

For each of the three plant species, the α-diversity of bacterial communities decreased from the rhizosphere to the rhizoplane and further to the endosphere ( Figure 3 ). In the rhizosphere, taxon richness, as indicated by ACE and Chao1 indices, was significantly higher for C. indica and H. littoralis than for I. ensata (p<0.05), although the Shannon-Wiener diversity index did not significantly differ among the plant species. On the rhizoplane, the Shannon-Wiener diversity index was significantly higher for C. indica compared to H. littoralis and I. ensata (p<0.05). In the endosphere, the ACE and Chao1 indices were significantly higher for C. indica and H. littoralis than for I. ensata (p<0.05) ( Figure 3 ). Additionally, the diversity and abundance of bacterial community was higher in the wet season than in the dry season.

The β diversities of bacterial communities, assessed by Bray-Jaccard dissimilarity, showed distinct variations among the three root components across different species and seasons ( Figure 4 ). Similarity in bacterial communities was observed within the same root compartment across different plant species, while differences in the composition of bacterial community were evident among the three components within the same plant species. The rhizosphere had the largest overlapping region, followed by the endosphere, and the smallest overlap was observed in the rhizoplane. This pattern indicates that the similarity of bacteria was highest for the rhizosphere, intermediate in the endosphere, and lowest in the rhizoplane.

Different bacterial biomarkers were identified between various plant species and different root components ( Supplementary Figure S1 ). The taxonomic hierarchy, ranging from phylum to genus, was represented by concentric rings ( Supplementary Figures S1A–C ). Using an LDA score threshold of 3.0, the number of discriminative biomarkers among the three plants tended to increase from the rhizosphere (13 clades) to the rhizoplane (24 clades) and further to the endosphere (29 clades). In the case of I. ensata, the number of bacterial biomarkers in the rhizosphere, rhizoplane, and endosphere was 1, 0, and 5, respectively. These numbers were 10, 21, and 12 for C. indica, and 2, 3, and 12 for H. littoralis. Notably, the number of biomarkers in bacterial communities was highest on the rhizoplane of C. indica and lowest for of I. ensata. Specific bacterial taxa with the highest relative abundances in different root components included Oxyphotobacteria (LDA = 4.06) in the rhizosphere of C. indica, Deltaproteobacteria (LDA = 4.26) on the rhizoplane of C. indica, Caulobacterales (LDA = 4.44) in the endosphere of I. ensata, and Devosiaceae (LDA = 4.18) in the endosphere of C. indica.

The PICRUSt analysis identified a total of eight gene families that relate to metabolic functions of the bacterial communities in both dry and wet seasons ( Supplementary Figures S2, S3 ). Principal functional pathways, including metabolism, environmental information processing, and organic systems, were consistently detected as the main functional genes across all plant species in both seasons. Among these, metabolism-related genes were the most abundant. The metabolic functions reached their peak for I. ensata roots in the dry season (41.5 and 41.9, Supplementary Figures S2A, D ) and for H. littoralis roots in the wet season (42.6 and 42.6) ( Supplementary Figures S3C, F ).

The plant species had a significant effect on the predicted functions of bacterial communities among root components in different seasons ( Supplementary Figures S2, S3 ). Specifically, in the dry season, the abundance of genes related to metabolism was lower in the rhizosphere and rhizoplane across three plant species compared to the endosphere. Conversely, in the wet season, the relative abundance of genes related to metabolism and environmental information processing was higher in the rhizoplane and rhizosphere than in the endosphere ( Supplementary Figures S2D–F ).

The RDA results revealed significant changes in the correlations between bacteria at the genus level within three plant species and environmental variables across different root components and seasons ( Figure 5 ). In the dry season, certain bacteria (e.g., Bryobacter, SWB02, Sphingomonas and Terrimonas) were positively related to TN, COD, and NH₄ ⁺-N in the rhizosphere, rhizoplane and Endosphere, respectively. Similarly, other bacteria (e.g., Allorhizobium, Sphingomonas, Nitrospira, Flavobacterium and Novosphingobium) and additional bacteria (e.g., Allorhizobium, Acidovorax, Hydrogenophaga, Flavobacterium and Haliangium) exhibited positive correlations with TN, COD, and NH₄ ⁺-N in the rhizosphere, rhizoplane and Endosphere, respectively. While in the wet season, some bacteria communities (e.g., Ellin6067, Lactovacillus and RB41) were positively related to TP, TOC, and NO₃ ^–N in the rhizosphere while other bacteria (e.g., Nitrospira, Hyphomicrobium, and Piscinibacter) were positively related to TP, TOC and NO₃ ^–N in the Endosphere. These findings underscore the dynamic nature of bacterial correlations with environmental variables across different root components and seasonal conditions.

The study highlighted that the characteristics of bacterial communities associated with roots were significantly affected by different plant species (I. ensata, C. indica, and H. littoralis), root components (rhizosphere, rhizoplane, and endosphere), and environmental factors ( Figures 2 , 3 ). Across the rhizosphere, rhizoplane, and endosphere of three plant species, Proteobacteria, Cyanobacteria, Bacteroides, and Acidobacteria were predominantly components ( Figure 2 ). The rhizosphere consistently exhibited the highest relative abundance of bacteria, aligning with the findings of Beckers et al. (2017) and Coleman et al. (2016). The widely acknowledged notion that the rhizosphere constitutes an environment that fosters more abundant bacterial populations compared to the non-rhizosphere soil further supports this observation (Qiao et al., 2017). Bacterial communities in the rhizosphere have the potential to translocate into plant tissues and establish residence in root endosphere, as demonstrated by previous studies (Hacquard et al., 2015; Vandenkoornhuyse et al., 2015). Within the same plant species, the alpha diversity of bacteria communities decreased from the rhizosphere to the endosphere. This decline may be caused by the selective effect exerted by the endosphere environment (Philippot et al., 2013). The influent sewage contains a high concentration of non-native bacterial species, and the plant’s immune mechanisms effectively eliminate numerous pathogenic bacteria (Mucyn et al., 2006). This process leads to a variation in microbial diversity between the rhizosphere and the endosphere. The selection process may be related to the expression of genes involved in various functions (Bulgarelli et al., 2012). Generally, the colonization of bacteria in different root components is primarily driven by two factors: (i) root rhizodeposits, including root exudates and mucilage of root caps, and (ii) the relatively simple or inelaborate chemo-attraction of the bacteria to the root exudates (Walker et al., 2003). These two factors may explain the α diversity were lower in I. ensata roots than in C. indica and H. littoralis roots ( Figure 3 ). Specifically, C. indica and H. littoralis exhibited more extensive root systems compared to I. ensata, and they released greater amounts of oxygen and organic compounds to facilitate bacterial proliferation (Lee and Scholz, 2007). These emissions contributed to the enrichment of bacterial communities associated with their roots, increasing the bacterial diversity for C. indica and H. littoralis.

Unlike the α diversities of bacterial communities, the β diversities exhibited low variability during the dry season but demonstrated a high variability during the wet season within the rhizosphere ( Figure 4 ). Bacteria, characterized by small size and a short life cycle, are inherently susceptible to disturbances from the external environment (Zhang et al., 2020). The polluted water releases a complex array of pollutant and microorganisms into the environment to affect the bacterial communities in the rhizosphere directly (McLellan et al., 2010), making bacterial composition became similar. However, this phenomenon was changed in wet season ( Figure 4 -Wet). The bacterial abundance and diversity increased significantly under the relatively high temperature and humidity in constructed wetland. Besides, environmental variables, specifically TOC, TP, and NH₄ ⁺-N, provided nutrients and energy sources for different bacterial communities ( Figure 1 ). More bacterial communities involved in the absorption and transformation of organic carbon, phosphorus, and nitrogen increased to meet the rapid growth of plants, which resulted in the similarity of bacterial communities in the endosphere.

The numbers of bacterial biomarkers increased from the inside of the root system to the outside (i.e. from the endosphere to the rhizosphere) in the three plants ( Supplementary Figure S1 ). The abundance and composition of biomarkers were used to characterize the level of pollutant degradation and to determine which microbial taxa are more sensitive to the external environment (Wu et al., 2023). Within the bacterial communities of the root components of C. indica, the community on the rhizoplane had the highest number of biomarkers, indicating that it was most sensitive to the external environment. This observation was consistent with the findings of Chen et al. (2020). The results of LEfSe further revealed differences in the dominated bacterial communities in root components, i.e. the number of primarily dominant bacterial communities in the rhizosphere was smaller for all three plants compared to the endosphere, and the bacterial composition was largely distinct, with each having an LDA score of > 4.0. Our results showed soil/water environment might be the main factor in changing the structure of the bacterial community, while it turned the equilibrium state in the root systems.

The content of TN, TP, TOC, COD, NO₄ ⁺-N and NO₃ ⁻-N in constructed wetland significantly decreased after being treated with plants, and the pollutant removal efficiency was higher in wet season ( Figure 1 ). The consequence of pollutant removal efficiency was supported by previous research that both nitrification and denitrification would be inhibited strongly at temperatures below 20°C (Yao et al., 2013). Cyanobacteria, mainly Oxyphotobacteria, were significantly enriched in the rhizoplane of all three plant species ( Figure 2 ; Supplementary Figure S3 ). Oxyphotobacteria, known for their strong ability to decompose and transform organic matter, can photosynthesize under anaerobic conditions and release oxygen that degrades nitrite in water, thereby enhancing water purification (Ren et al., 2023). The characteristic of Oxyphotobacteria of three plant species suggested that the catabolic degradation of contaminants occurred mainly in the rhizoplane. Proteobacteria were proven to reduce N and P levels in constructed wetlands by facilitating nutrient exchange and metabolite (Fu et al., 2006). For example, both Gammaproteobacteria and Deltaproteobacteria can removing nitrate and nitrite because they can fix N and perform denitrification under anaerobic conditions (Valaskova et al., 2009; Zhang et al., 2021). The abundance of Oxyphotobacteria, Gammaproteobacteria and Deltaproteobacteria were mainly in the root rhizosplane and endosphere, indicating that the process of water purification occurred primarily in the rhizosplane and endosphere. Dominant members of the root endosphere communities, such as Ignavibactera, Dependentiae, and Babeliales, have also been empirically proven to yield advantageous outcomes for plant growth metabolism and health (Innerebner et al., 2011). The distinct endosphere environment determined by the root selective absorption different plant species (Anisimov and Suslov, 2023), which also affected the structure of bacterial communities. Thus, we speculate that the water purification by bacterial communities depends on the different elements absorbed by the host plant’s immune system and is regulated by seasonality. This process mainly occurs in endosphere.

The potential functions of bacterial communities were different in different plant species and root components. The predominant bacterial genes were associated with metabolism, environmental information processing, and membrane transport in the rhizosphere and the rhizoplane ( Supplementary Figure S2 ). Conversely, the predominant bacterial genes were related to metabolism in the endosphere. Metabolic functional genes in bacteria play a vital role in facilitating the uptake and utilization of amino acids, energy, carbohydrates, etc (Ruiz-Gonzalez et al., 2015). Bacteria reduce nitrate to nitrogen by denitrification ( Figure 5 ). The bacterial communities with metabolic functions were predominantly preserved inside the roots, indicating that the main role of the bacteria in the roots was nutrient cycling (such as denitrification) and providing plant roots with the essential nutrients required for their growth. Compared to C. indica and H. littoralis, I. ensata preserved more metabolism pathways in all root components, indicative of high metabolically active (Chen et al., 2019). Genes associated with membrane transport were also important in the rhizosphere and rhizoplane of three plant species. With the increase in bacterial diversity, their genetic contributions facilitated the dissolution of iron and small molecules, as well as the transfer of dissolved oxygen (Wu et al., 2023; Ren et al., 2024). This process ensured and enhanced the absorption of minerals and amino acids from the rhizosphere and rhizoplane, promoted nutrient cycling within the endosphere, and ultimately influenced plant physiological processes.

The bacterial community demonstrated a positive correlation with TN, COD, and NO₄ ⁺-N in the dry season ( Figure 5 ). COD, containing heavy metals (HMs) and polycyclic aromatic hydrocarbons (PAHs), was significantly degraded by Bryobacter, Nitrospira, and Sphingobium (Liang et al., 2021; Yang et al., 2022). RB41 and MND1, functioning as module hubs, have important ecological roles in triggering resistance or tolerance to toxicity (Wang et al., 2019). RB41 and MND1 were positively related to C. indica in the rhizosphere ( Figure 5 ), indicating bacterial communities enhanced stress resistance for C. indica. Thus, bacterial communities in dry season is to protect plants from toxic pollutants. In the wet season, the bacterial species composition was all positively related to TP, TOC, and NO₃ ^–N in three root components. The biomass of plants and abundance microbes increased in wet or hot seasons (Pang et al., 2016; Schofield et al., 2022). When the content of water pollutants is stable, the increase of microbial abundance can significantly enhance the transport of rhizosphere and the metabolic function of endosphere to promote plant growth (Makgato and Chirwa, 2020; Singh et al., 2025). Our predictions of gene functions indicate that microbial community ensure survival by removing pollutants in the dry season, while the microbial community expands in the wet season, achieving better pollutants removal efficiency.