Halophytes Increase Rhizosphere Microbial Diversity And Network Complexity In Inland Saline Ecosystem


 Background: Salinization is an important global environmental problem influencing sustainable development of terrestrial ecosystems. Salt-tolerant halophytes are often used as a promising approach to remedy the saline soils. Yet, how halophytes affect rhizosphere microbial diversity, and microbes’ association and functions in saline ecosystems remains unclear, restricting our ability to assess plant fitness to salt stress and to remediate saline ecosystems. Herein, we examined bacterial and fungal diversities, compositions, and co-occurrence networks in the rhizospheres of six halophytes and bulk soils in a semiarid inland saline ecosystem. We also established the relationship of microbial structure and network complexity to microbial functions.Results: The microbial communities in rhizospheres were more diverse and complex than those the bulk soils. The connections of taxa in the rhizosphere microbial communities increased with fungi-fungi and bacteria-fungi connections and fungal diversity, but decreased with bacteria-bacteria connections and bacterial diversity. The proportion of the fungi-related central connections were larger in the rhizospheres (13-73%) than the bulk soils (3%). Additionally, fungi accounted for 27-63% of the keystone taxa identified in the microbial co-occurrence networks present in the rhizospheres, whereas the keystone taxa identified in the bulk soils were all bacteria/archaea. Moreover, microbial activity and residues were significantly higher in the halophyte rhizospheres than the bulk soils, and were significantly correlated with microbial composition and co-occurrence network complexity.Conclusions: These results indicated that halophytes shaped rhizosphere microbiomes and increased microbial diversity and network complexity in inland saline ecosystem, while fungi enhanced rhizosphere microbiota associations. The increased microbial network complexity contributed to the higher microbial functions in rhizosphere soils.

It has been demonstrated that microbial diversity was smaller and the associations among microbes were less complex in rhizosphere than bulk soils in non-saline condition [42,43]. Such variations were mainly attributed to the host plant selection or root ltering [42,44,45] by secreting speci c chemical compounds that offer a selective advantage to some speci c microorganisms [46][47][48], or bioactive molecules that directly inhibit particular microbial taxa [46,49], or plant compounds that act as signals to trigger microbes' changes [46, 49,50]. Whereas, whether such response pattern exists in saline soils was not examined. On the other hand, while many studies have focused on soil-borne bacteria and archaea, soil fungi are also important for providing ecosystem services [15,27,51]. Some fungi (e.g., mycorrhizae) have the ability to help plants obtain resources under stressed environments and to tolerate salinity [15,51]. Fungi also interact with bacteria to maintain the stability of soil microbial network and support ecosystem functionality [27]. However, it is still unclear how fungi interact with bacteria in saline soils.
Such knowledge gaps highlight the need to investigate microbial associations in halophyte rhizosphere of saline soils.
Herein, we analyzed bacterial and fungal diversities, compositions, and co-occurrence networks in the rhizospheres of 6 halophytes (i.e., Phragmites australis, Calamagrostis epigeios, Kalidium foliatum, Suaeda salsa, Nitraria sibirica and Tamarix chinensis) and bulk soils in a semiarid inland saline ecosystem. We aimed to examine how halophytes alter rhizosphere microbial diversity and microbe's association in saline soils. We measured the nutrient contents, soil pH, electric conductivity and salinity, and extracellular enzyme activities to assess nutrient availability and salt stress in rhizosphere. We measured extracellular enzyme activities because they play critical roles in biogeochemical cycles and nutrient availability and are closely related with microbial communities in saline soils [52], and also could be used to indicate microbial activity [53]. We also measured total amino sugars to indicate total microbial residues [54]. We quanti ed the diversity and composition of soil bacteria/archaea by using 16S rRNA gene amplicon sequencing and those of fungi by using ITS rDNA gene amplicon sequencing, respectively. We conducted indicator species analyses to identi ed bacterial and fungal taxa that signi cantly associated with a given habitats. We constructed co-occurrence networks to explore how halophytes in uence the associations among microbes. We identi ed the central interactions and microbial keystone taxa, taxa that have an important in uence on community structure and functioning [24,25,55,56], to understood the associations among microbes. We also established the relationship of microbial composition and network complexity to microbial functions. We demonstrated that halophytes increased microbial diversities and co-occurring network complexity, and that fungi enhanced associations among rhizosphere microbiota in this inland saline ecosystem. Such facilitation in microbial community contributed to the higher microbial functions.

Soil properties and microbial diversities
The contents of organic carbon (SOC), total nitrogen (TN), nitrate nitrogen (NO 3 − ), and soil moisture and the activities of extracellular enzymes were signi cantly higher, but electric conductivity (EC) and salinity were signi cantly lower in the rhizospheres of most halophytes tested than in bulk soils (Fig. S1), indicating that nutritional conditions were improved and the salinity stress were ameliorated in halophyte rhizospheres.
Halophyte rhizospheres had more diverse microbiota than did the bulk soils in this saline environment. Both bacterial and fungal alpha diversities were higher in the rhizospheres of the halophytes than in the bulk soils, except for S. salsa rhizosphere that had similar fungal Shannon index to that in the bulk soils ( Fig. 1). Additionally, rhizosphere microbial diversity varied signi cantly by halophyte species. Among the six halophytes, T. chinensis had the highest bacterial diversity, while C. epigeios and K. foliatum had the highest fungal diversities. Moreover, the observed OTUs, Faith's phylogeny diversity, Chao1 and ACE of bacterial communities were negatively correlated with those of fungal communities across the six halophytes rhizospheres (Fig. S2).
The bacterial and fungal communities signi cantly differed among the rhizosphere and bulk soils and among the six halophytes (Table 1). Constrained analysis of principal coordinate showed that microbial communities clustered by habitat (rhizosphere vs. bulk soils) and halophyte species, which explained ~ 46% and 49% of the total variations for bacteria and fungi, respectively (Fig. 2). Many OTUs (1563 and 505 for bacteria and fungi, respectively) were shared among the seven habitats (six rhizosphere and one bulk soils) and among the six halophytes rhizospheres (239 and 62 OTUs for bacteria and fungi, respectively), although less unique OTUs were identi ed in both bulk and rhizosphere soils (Fig. 2). This indicates that similar microbial species occur in both bulk and rhizosphere soils. The dissimilarity distance of each rhizosphere microbial community was calculated by comparing with bulk soils (Fig. 3). The S. salsa and K. foliatum rhizospheres had lower dissimilarity distance than other halophytes. Among the six plants, the dissimilarity distances for bacteria (0.57-0.62) were smaller than that for fungi (0.69-0.85). Furthermore, the rhizospheres had signi cantly lower variation (distance from centroid) for both bacteria (0.23) and fungi (0.37) than the bulk soils (0.41 and 0.54), whereas rhizosphere bacteria had a smaller variation (0.19-0.27) than fungi (0.15-0.52) among the six halophytes (Fig. 3). The variations in both bacterial and fungal communities were mainly explained by the halophyte species. The explanation of host species was greater, but that of environments (nutrients and EC) was smaller for fungi (0.176 vs 0.017) than bacteria (0.108 vs. 0.030) (Fig. 3). Hence, halophytes had greater effects on fungal than bacterial community assemblies.
The numbers of bacterial/archaeal indicator OTUs were larger in the bulk soils (216) than the rhizosphere soils (28-148). These OTUs were dominated by the members belonging to Halobacterota, Nanoarchaeota and Proteobacteria. Among the six halophytes, C. epigeios, P. australis and K. foliatum had more indicator OTUs than T. chinensis. Additionally, C. epigeios and P. australis rhizosphere soils shared many bacterial/archaeal indicator OTUs (46), mainly belonging to Proteobacteria and Gemmatimonadota (Fig. 4).
The numbers of fungal indicator OTUs were signi cantly larger in C. epigeios and P. australis rhizospheres (96 and 49) than the other plant rhizospheres and the bulk soils (17-36). The K. foliatum rhizosphere and the bulk soils shared many fungal indicator OTUs (93), most of which belonged to Ascomycota, while the rhizospheres of the six halophytes only shared a small number of fungal indicator OTUs among them (Fig. 4). Hence, indicator species composition varied by host plants in this saline ecosystem.
The top 50 most abundant OTUs from the rhizosphere and bulk soils were examined in more detail in order to better characterize the effects of halophytes on rhizosphere microbial composition. The relative abundances of the 50 most abundant OTUs together accounted for 29.1% and 60.0% of the bacterial and fungal sequences, respectively. Most of these OTUs (48 and 37 OTUs for bacteria and fungi) were also identi ed as indicator taxa and were enriched in the rhizosphere soils (Fig. 5). The abundant bacterial/archaeal OTUs were mainly a liated within Proteobacteria and Gemmatimonadota. The OTUs belonging to Halomonas and Candidatus_Nitrosop were identi ed as indicator OTUs in K. foliatum and N. sibirica rhizospheres. The relative abundance of OTUs related to the genera Limibaculum and Pelagibius were high in both rhizospheres and bulk soils, while those related to the genus Woeseia was signi cantly larger in the rhizospheres than the bulk soils. Most of the abundant fungal OTUs were related to the class Sordariomycetes and Leotiomycetes (Fig. 5). Speci cally, Trichoderma (Sordariomycetes) were enriched in the rhizospheres, particularly in those of S. salsa, N. sibirica and K. foliatum. The Trichoderma OTUs accounted for > 70% of the sequences identi ed in S. salsa rhizosphere, and for 2-52% in of the sequences in the other halophyte rhizospheres. In contrast, OTUs related to Blumeria was the major sequences enriched in the C. epigeios rhizosphere.
Microbial co-occurrence networks When combined bacterial/archaeal and fungal OTUs altogether, the rhizospheres of the six halophytes had signi cantly larger number of nodes (559 to 725 nodes, highest in the C. epigeios rhizosphere) and connections between microbes (1037 to 1715 edges, highest in the K. foliatum rhizosphere) than the bulk soils (322 nodes and 645 edges) ( Fig. 6; Table 2). All networks showed small, scale-free and non-random interaction patterns in the bulk and halophyte rhizosphere soils (Table S1). Although both co-occurrences (positive edge) and mutual exclusion (negative edge) were observed in the bulk and rhizosphere soils, the halophyte rhizospheres had higher proportion of mutual exclusion (3-16%) but lower proportion of cooccurrences (84-97%) in the microbial co-occurrence networks than those of the bulk soils (2% and 98%, respectively) ( Table 2). In addition, the numbers of modules, negative cohesion and negative/positive cohesion were greater in the networks of the rhizospheres from the six halophytes (64 to 92, -0.25 to -0.31, and 0.71 to 0.91) than in those from the bulk soils (29, -0.24 and 0.65). Moreover, the values of modularity were also greater in four out of six halophyte rhizospheres (except N. sibirica and K. foliatum) ( Table 2). Hence, halophytes increased complexity of microbial co-occurrence network in rhizosphere compared with bulk soils. The cohesion values are calculated as the sum of the signi cant abundance-weighted, with larger negative cohesion or Negative/Positive cohesion values indicating microbial networks tend to be more stable.
Although the numbers of bacteria-bacteria connections were higher in some halophyte rhizosphere networks, the proportions of fungal nodes out of total nodes and those of fungi-fungi and bacteria-fungi connections out of total connections were larger in the halophyte rhizosphere networks than the bulk soil networks (Fig. 6, Table 2). Moreover, the total number of connections increased with the numbers of fungi-fungi and bacteria-fungi connections (p < 0.05 by correlation analysis) but not with bacteriabacteria connections (Fig. S3). These results indicate that fungi might play more important roles than bacteria in maintaining rhizosphere microbial network complexity.
We generated the taxonomic pro le to identify the central interactions (edge betweenness centrality) [57]. Among the 30 central interactions, 29 were bacteria-bacteria co-occurrences in the bulk soils, mainly between Proteobacteria, Gemmatimonadetes and Acidobacteria. For the rhizospheres, the number of central interactions originated from bacteria-fungi or fungi-fungi co-occurrences ranged from four in T. chinensis to 22 in S. salsa (Fig. S4). Most of these interactions were occurred between Ascomycota, Basidiomycota and Rozellomycota.
We identi ed ve and 6-15 keystone taxa in the bulk and rhizosphere soils, respectively. The ve keystone taxa identi ed in the bulk soils belonged to the bacterial phyla Acidobacteria, Actinobacteria, and Gemmatimonadetes, and archaeal class Nanosalinia (Nanosalinaceae family). For the rhizospheres, 27-63% of the keystone taxa were fungi, mainly those belonging to Ascomycota (mainly Hypocreales) (Table S2). These results suggest that fungi enhanced associations among rhizosphere microbiota in this inland saline ecosystem.
Linkages of soil microbial diversity and co-occurrence network complexity to microbial function We calculated the average of the z-score of each extracellular enzyme activity to assess soil microbial activity [53,58]. We measured total amino sugar to indicate microbial residue in soils [54]. The microbial activity and residues were signi cantly higher in rhizospheres soils than the bulk soils, with highest values in S. salsa (Fig. 7). The results from FAPROTAX and FUNGuild prediction showed that rhizospheres soils had signi cantly higher relative abundance of nutrients cycling functional groups and arbuscular mycorrhizal fungi than the bulk soils (Fig. S5). Moreover, both microbial activity and residues were signi cantly correlated with soil nutrients, microbial composition and co-occurring network complexity (Fig. S6). The results from structure equation model showed that microbial network complexity was more important than microbial composition in affecting microbial activity and residues. Soil microbial network complexity has signi cantly direct effects, while microbial composition has signi cant indirect effects through changing microbial network complexity (Fig. 7). Additionally, soil nutrients not only directly impact microbial activity and residues, but also indirectly exert their in uences through changing microbial composition, while soil EC exerted its effects only through changing microbial composition (Fig. 7). Therefore, the increased microbial functions in the halophyte rhizospheres were mainly due to the increased co-occurrence network complexity.

Discussion
Halophyte increases microbial diversity in saline soils The higher microbial diversity in the halophyte rhizospheres than the bulk soils was primarily ascribed to the higher available resources (nutrients) and less environmental stress. In this study, the contents of organic carbon, total nitrogen, nitrate nitrogen and soil moisture, and activities of various enzymes were signi cantly higher in rhizospheres than the bulk soils, whereas the EC and salinity were lower in four out of six rhizospheres than the bulk soils. The larger organic carbon, nutrient and moisture contents and lower slat stress in the rhizosphere might have promoted the microbial metabolisms and thereby increasing the overall diversity. This explanation was supported by our observation that bacterial diversity increased with organic carbon, nutrient and moisture contents but decreased with EC and salinity (Fig.  S7).
Our result was not in agree with previous understanding that rhizosphere usually has lower microbial diversity than bulk soils because of host plants select [42,44,45]. The host plant selection often decreases rhizosphere microbial diversity compared with bulk soils in managed ecosystem (e.g., agricultural ecosystem) or unstressed conditions because of root ltering [42,44]. Such decreased microbial diversity in rhizosphere was observed in wheat [42] and wheat-maize/barley rotation systems [59]. However, Schmidt et al. [45] found that the bacterial diversity in maize rhizosphere was larger than bulk soils in long-term conventional agroecosystem, but the opposite was found in organically managed agroecosystem. Therefore, the effect of host selection on rhizosphere microbiome largely depends on plant type and environments [60, 61]. It has been suggested that, under salt or desert conditions, host plant selection increases rhizosphere microbial diversity because of the alleviated environmental stresses in rhizosphere [40,41,57]. For example, Marasco et al [57] reported higher fungal diversity (but similar bacterial diversity) in the desert speargrass rhizosphere than the bulk sand. Some recent studies also showed increased microbial diversity in the rhizosphere in coastal saline soils [40,41].
Halophyte increases complexity of rhizosphere microbial co-occurrence network in saline soils In this study, we observed that microbial co-occurrence networks were more complex in the halophyte rhizospheres than the bulk soils, probably because halophyte rhizosphere has greater potential for nichesharing and interactions between microbes [62]. Similar to this study, more complex bacterial networks were identi ed in the rhizosphere than bulk soils from greenhouse microcosms [62] and eld sand dune [63]. According to the Stress Gradient Theory, competitive (negative) interactions increase but facilitative (positive) interactions decrease with decreased stress [64,65]. In this study, salt stress (the EC and salinity) were signi cantly lower (except for S. salsa), but the availability of resources (organic carbon and nutrients) were signi cantly higher in the rhizosphere than the bulk soils, while the competitive taxa, fast-growing species (e.g., Rhodobacteraceae, Nitrosomonadaceae, Cyclobacteriaceae, Desulfuromonadaceae, Nitrincolaceae, TRA3-20 and BIrii41 families, S0134 and AKAU4049 classes), was signi cantly higher, and replaced slow-growing, halophilic species (e.g., Balneolaceae, Haloferacaceae, Halomicrobiaceae and Salinisphaeraceae families and Woesearchaeales order) (Fig. S8). Additionally, EC and salinity were negatively correlated with competitive taxa, but positively correlated with facilitative taxa (Fig. S8). The EC was also negatively correlated with the ratio of negative to positive connections and the ratio of negative to positive cohesions (Fig. S9). Hence, such decreased environmental stresses contributed to the increased complexity of microbial co-occurring networks in halophyte rhizospheres than the bulk soils.
The complexed microbial co-occurring network in rhizospheres was also due to the increased interaction of fungi with bacteria. The host plant usually selected some fungi in harsh environment because of the high ability of these fungi to tolerate salt and desiccation [66]. Generally, coexisting fungal and bacterial communities are more stable than fungi-or bacteria-dominated communities [67], particularly under stressed conditions [25,41,68]. In this study, the proportion of bacteria-fungi and bacteria-bacteria interactions out of total interactions was larger and smaller, respectively, in the microbial co-occurrence networks of the halophyte rhizospheres, and the negative/positive cohesions increased with the proportion of bacteria-fungi connections. Therefore, increases in the functional and taxonomic diversities increase the complexity of halophyte rhizosphere microbial communities. While we quanti ed the cooccurrence network of microbial communities by using statistical tools, empirical studies should be conducted to further support our ndings.

Fungi enhanced associations among microbiota in halophytes rhizosphere
Our results demonstrated that fungi enhanced associations among rhizosphere microbiota in this inland saline ecosystem. The keystone taxa identi ed in the bulk soils were all bacteria and archaea, while fungi composed 27-63% of the keystone taxa in halophyte rhizospheres. We identi ed Ascomycota as the keystone taxa in the rhizospheres, and most central interactions were originated from the co-occurrence with them, indicating that Ascomycota played a key role in the interaction with other taxa in the rhizosphere. This is probably because Ascomycota has a strong adaptability to stresses and ability to utilize resources and thus niche preferences [69]. Among Ascomycota, ve Trichoderma were identi ed as keystone taxa in four (C. epigeios, K. foliatum, N. sibirica and P. australis) out of six rhizospheres. In addition, 15 out of 50 top most abundant OTUs identi ed in the rhizosphere soils were related to Trichoderma, and 14 of which were identi ed as indicator OTUs. Trichoderma often inhibit the growth of plant pathogens and root-knot nematodes [70][71][72], and thus to promote plant growth. Additionally, we also identi ed Basidiomycota as the keystone taxa in the rhizosphere communities. Most members of Basidiomycota tend to live in relatively harsh environments, and their relative abundance in soils is related to their ability to decompose lignocellulose, and to the availability of resources [73,74]. The increased nutrients in the rhizosphere might have resulted in the higher relative abundance of Basidiomycota.
We also observed that the total connection of rhizosphere microbial network was positively correlated with fungal diversity but negatively correlated with bacterial diversity (Fig. S10), further indicating that the complexity of halophyte rhizosphere microbial network was enhanced by fungi in this saline ecosystem. These results were consistent with recent evidences that fungi are more important than bacteria for distinguishing among root-zone microbiomes of halophytes [41]. However, whether fungi are more important than bacteria for sustaining rhizosphere microbial network complexity under managed conditions or unstressed conditions merits further examination.

Conclusions And Implications For Sustainable Development Of Saline Ecosystem
Given the continuous rising of saline soils as a result of intensi ed climate change and human activities, and the increasing dependence of human on saline ecosystems, it is timely and of the highest importance to comprehensively analyze the interactions between halophytes and soil microbiota for the remediation of saline soils. Here we demonstrated that halophytes shaped rhizosphere microbiomes and increased microbial diversities in inland saline ecosystem due to decreased salt stress. Combined with previous understandings, our results highlighted that host plant selection might have divergent effect on rhizosphere microbial diversity, with decreased diversity under management or natural conditions but increased diversity under some stressed conditions (i.e., drought and salt). Therefore, the improvement of microbial communities and functionality might be an important mechanism for the remediation of saline soils with halophytes, and also for any other practices that has ability to decrease salt stress. We further showed halophytes have greater effects on fungal than bacterial communities and that fungi enhanced microbiota associations, which provides new insight that halophytes sustain rhizosphere microbial network complexity by enhancing association of fungi with bacteria. These ndings emphasize the need to develop technologies to increase rhizosphere microbial diversity and include important fungal taxa (e.g., Trichoderma) to facilitate sustainable development of saline ecosystem.

Study site and soil sampling
This study was conducted in an inland semiarid saline ecosystem around the largest salt lake, Huamachi lake, in Shaanxi, China. The Huamachi lake is located in Dingbian of Shaanxi province (37º68′N, 107º53′E). The study site has a temperate continental semi-arid monsoon climate, with a mean annual temperature of 7.9°C, mean annual precipitation of 312 mm, and mean evaporation demand of 2523 mm. Precipitation primarily occurs from July to September. The soil in the dry lakebed and the surrounding region is a saline soil, with a texture of sandy loam, an EC of 6.72-12.08 mS cm − 1 and a salinity of 2.79-5.79%. The covering vegetation is dominated by Phragmites australis, Calamagrostis epigeios, Kalidium foliatum, Suaeda salsa, Nitraria sibirica and Tamarix chinensis, with a canopy cover of > 60%.
In early July 2020, we established nine adjacent plots (10 m × 10 m) for the sampling of rhizosphere and bulk soils in the surrounding region of Huamachi lake. In each plot, we randomly selected six plants for each of the six halophytes (P. australis, C. epigeios, K. foliatum, S. salsa, N. sibirica and T. chinensis) for the eld sampling of rhizosphere samples. The roots of each plant from 0-15 cm depth were carefully taken out, lightly shaken to remove loosely bound soils, then the soils that tightly attached to the roots (rhizosphere soils) were brushed off to compose the rhizosphere sample of each halophyte for the plot.
We also collected 6 soil samples at 0-15 cm depth from bare land without any plant using a sterilized soil auger to combine as a composite bulk soil sample for the plot. Totally, we have 63 samples (six rhizosphere samples and one bulk soil sample for each of the nine plots). All the moisture rhizosphere and bulk soil samples were stored in ice box, transported to local laboratory and then divided into three sub-samples: one air-dried for the measurement of soil physiochemical properties, one stored at -20°C for the measurement of extracellular enzyme activities, and one stored at -80°C for high-throughput sequencing.

Laboratory analysis for soil properties and extracellular enzyme activities
A small fraction of moisture soils was dried at 105°C to measure soil moisture content. Another fraction of moisture soil was used for the measurement of available nitrogen (NH 4 + and NO 3 − ). Air-dried soils were ground to pass through a 2-mm sieve for the measurement of available phosphorous (OP), soil pH, EC and salinity. A small fraction of < 2-mm samples were ground to pass through a 0.25-mm sieve for the measurement of SOC and TN. Soil metrics were measured using standard methods as described by Qiu et al. [75]. The SOC and TN were measured using the Walkley-Black and Kjeldahl method. Soil NH 4 + and The uorescence signals for the enzymes were obtained at 365 nm excitation and 450 nm emission (BioTek, Synergy TM LX, USA). The average of the z-score of each extracellular enzyme activity was calculated to indicate soil microbial activity [53,58].
The content of total amino sugar (TAS) was measured to indicate microbial residue in soils [54]. Brie y, the soil samples were hydrolyzed with 10 mL of 6 M HCl at 105℃ for 8 h containing myo-inositol. The hydrolysate was ltered after adding myo-inositol, adjusted to pH 6.6-6.8 and centrifuged. The supernatant was freeze-dried and the amino sugars in residues were extracted with methanol. The recovered amino sugars were rst cyanated with hydroxylamine hydrochloride and 4dimethylaminopyridine, and then acetylated with acetic anhydride to form aldononitrile derivatives. The AS derivatives were analyzed via a Beifen 3420A gas chromatograph (Beijing Beifen-Ruili Analytical Instrument. Co. Ltd., Beijing, China) equipped with a DB-5 column (50 m × 0.25 mm × 0.25 µm) and a ame ionization detector. The detailed temperature program was set by Zhang and Amelung [54].
Methylglucamine was added as the recovery standard before derivation. Each individual amino sugar was quanti ed, and then summed to calculate total amino sugars content.

DNA extraction, PCR and sequencing
Soil total genomic DNA of each sample was extracted using the MP FastDNA spin kit for soil according to the manufacturer instructions (MP Biomedicals, Solon, OH, USA). The 16S rRNA gene was ampli ed by polymerase chain reaction (PCR) using the targeting primer pairs 341F (5′-CCTAYGGGRBGCASCAG-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3). The ITS1 rDNA gene was ampli ed by PCR using the targeting primer pairs ITS5-1737F (GGAAGTAAAAGTCGTAACAAGG) and ITS2-2043R

Bioinformatics analysis
Raw paired-end reads were assigned to samples based on the unique barcode and truncated by cutting off the barcode and primer sequence, and then merged using FLASH (V1.2.7) with default parameters [77]. Then, sequences with more than three consecutive low-quality base calls (Phred quality score smaller than 20) were truncated, and only reads with > 75% consecutive high-quality base calls and no ambiguous characters were retained for further analyses based on QIIME v1.9.1 [78]. Chimeric sequences were removed using a combination of de novo and reference-based chimera checking based on UCHIME algorithm [79]. Sequencing depths were rare ed to 15,751 and 36,914 sequences for bacteria/archaea and fungi, respectively. We also veri ed that the removal of rare OTUs did not affect our results based on the correlation analysis for alpha diversity (R 2 : 0.76-0.99, p < 0.0001; Fig. S11 S12) [84]. In addition, we constructed OTUs rarefaction curves to evaluate richness saturation using ggrare function in ranacapa package (Fig. S13) [85]. We used the Functional Annotation of Prokaryotic Taxa (FAPROTAX) [86], and FUNGuild [87] to extrapolate functional groups of bacteria and fungi in the soil, by using FAPROTAX_v1.2.1 and FUNGuild_v1.1 python script with default settings. We used only taxa that were rated as either "probable" or "highly probable" according to the con dence ranking for the guild assignment in order to avoid over-interpreting the fungal functional groups. The sequencing data were deposited into the NCBI (https://www.ncbi.nlm.nih.gov/) with bioprojects accession number PRJNA738372.

Statistical analysis
All statistical analyses were conducted in the R environment (v3.6.3, http://www.r-project.org/). To assess the microbial alpha diversity, community richness (Observed OTUs, Chao1 and ACE) and diversity The separations of mean values (distance to centroid) among habitats were evaluated by one-way ANOVA with Tukey's HSD tests. All ordination analyses were performed using the phyloseq package [92].
We also calculated the pairwise dissimilarity distances between each halophyte rhizosphere and bulk soil bacterial or fungal community based on Bray-Curtis distance metric to explore the differences in bacterial or fungal community structures across halophyte rhizospheres and bulk soils.
Indicator species analyses were performed to identify bacterial and fungal taxa signi cantly associated with a given habitats on OTUs level using the multipatt function in the indicspecies package [93]. The analyses were permutated 999 times, and the signi cance of association between the taxa and the habitat was evaluated at a false discovery rate (FDR) corrected p value of < 0.05 [94]. Bipartite networks were implemented to visualize the identi ed indicator taxa using Gephi software (https://gephi.org). To better characterize the host effects of halophytes on rhizosphere microbial composition, we retrieved the dominant taxa (top 50 abundant for bacterial and fungal OTUs, respectively) and identi ed whether they belong to the indicator taxa. Phylogenetic distributions and relative abundance of the dominant taxa were constructed using ggtree function in ggtree package [95] and using ggplot function in ggplot2 package [96].
The microbial co-occurrence networks were constructed based on 16S and ITS gene sequencing data to explore the bacterial-fungal co-occurrence relationships. Microbial co-occurrence networks were constructed using Molecular Ecological Network Analyses Pipeline (MENAP, http://ieg4.rccc.ou.edu/MENA/), using random matrix theory (RMT) to determine the correlation cut-off threshold automatically [97,98]. The RMT theory was applied to de ne the transition point of nearestneighbor spacing distribution of eigenvalues from Gaussian orthogonal ensemble (random) to Poisson (non-random) distribution, and this transition point was then used as the correlation cut-off for network construction [97,98]. Brie y, only OTUs detected in at least eight replicates (out of nine replicates for each habitat) were used for each network construction. OTUs with missing values remained blank, and the pairwise correlation of the two pairs of OTUs were calculated based on relative abundance data and Pearson correlation coe cients. Then, the optimal threshold (0.92-0.94) recommended by RMT theory was used to lter the adjacency matrix (only correlations above the optimal threshold were used for de ning the adjacency matrix) and construct co-occurrence networks.
Network topological indices were calculated in the MENAP interface, including node, edge, R 2 of powerlaw distribution, average path distance (geodesic distance), harmonic geodesic distance, modularity and numbers of module. The smaller average path distance and harmonic geodesic distance indicate that the co-occurring network is close [97]. In addition, we calculated the edge betweenness centrality by edge.betweenness function and counted numbers of bacterial node, fungal node, positive edge, negative edge, bacteria-bacteria edge, fungi-fungi edge, and bacteria-fungi edge in each network in igraph package [99]. To validate the nonrandom co-occurrence patterns, 100 random networks were constructed for each empirical network in an equal size (constraining numbers of node and edge), following the Maslov-Sneppen procedure [100]. The mean values and standard deviations of topology properties from 100 random network were calculated and compared with the corresponding empirical network. Topological roles of nodes in networks were determined by within-module connectivity (Zi) and among-module connectivity (Pi) and classi ed into four categories: peripheral nodes (Zi < 2.5, Pi < 0.62), connectors (Zi < 2.5, Pi > 0.62), module hubs (Zi > 2.5, Pi < 0.62), and network hubs (Zi > 2.5, Pi > 0.62). Module hubs are highly connected to many nodes in their own modules, while network hubs act as both module hubs and connectors. Network hubs, module hubs and connectors are regarded as keystone nodes [97,98,101]. All networks were visualized in R package igraph [99].
Cohesion, as a measure of the connectivity of microbial communities, was the sum of abundanceweighted and null model corrected index based on pairwise correlations across taxa [102,103]. Cohesion can be used to identify associations among taxa caused by the interaction of positive and negative species and/or by the similarities and differences in the niches of microbial taxa [65,102]. In order to evaluate the differences in cohesion values among different networks and to be consistent with the network analysis, we focused on those taxa that were present in the network [65,104], and then calculating two cohesion values (positive and negative) for each sample: where n is the total number of taxa in a given community, and the construction process is mainly divided into two steps, including the calculation of the positive and negative connectedness matrix of all taxa in each sample and the nal calculation of cohesion. For example, we rst subtracted the pairwise associations matrix of all taxa within a given community (observed correlations) from the pairwise associations matrix of the null model iterations (expected correlations) to obtain pairwise correlations matrix corrected by null model of each taxa. Then, the positive and negative connectedness matrix of each sample was generated by averaging the positive and negative null model-corrected correlations separately. Finally, the positive and negative cohesion value of each sample was obtained by multiplying relative abundance table by the connectedness metrics according to the above formula [102]. The ranges of negative and positive cohesion were − 1 to 0 and 0 to 1, respectively. The more negative cohesion value indicates more complex microbial networks [65,102]. Additionally, the absolute ratio of negative to positive cohesion were calculated to predict the co-occurrence network complexity, higher ratio indicates       Supplementary Files