Anatomical Boundaries in Plant Roots Form Important Barriers for Selection of the Soil Microbiome

The microbiome in plant-soil systems has a signicant inuence in promoting plant growth. The extent of selectivity that the plant exerts on the microbiome in the continuum of internal plant tissues is not well understood. This study analysed the root microbiome of a legume, Melilotus ocinalis (L.) Pall., sweet clover, and focused on dynamic shifts in the microbial community structure through the niches of bulk soil, rhizosphere, periderm, phloem and xylem, and further examined the effects of environmental factors, root exudates and root cell wall development on the microbiome assemblages in different root compartments.

to lead to a conserved microbiome among different plants and across niches. For example, 40% of the bacterial taxa in the Arabidopsis endosphere also colonized the dead wood of tree species such as Betula and Fagus [13]. Therefore, cell wall components might play important roles in determining the structure of root-inhabiting microbiota. Previous studies on root-associated microbiota have revealed a high degree of microbial dynamism from the seedling stage through to maturity in rice [22], sorghum [23] and Arabidopsis [24], while the mechanisms driving the change in microbiome structure during the plant's life cycle are poorly understood, especially in regard to root metabolism and development of root structure.
In contrast to the majority of plants, some legumes can develop a bene cial root endosymbiosis with symbiotic rhizobia and form nitrogen-xing nodules [25]. Rhizobial infection induces the rapid growth of epidermal, cortical, and pericycle cells, with a relative abundance of 60 to 90% symbiotic rhizobia observed in soybean nodules [15]. An exponential increase in populations of symbiotic rhizobia through bulk soil, rhizosphere soil, root tissue and nodule tissue has been reported in different legume species [26]. However, the distribution of symbiotic rhizobia among anatomical compartments within root tissue have not been studied in detail.
Melilotus o cinalis is a forage legume, characterised by salt tolerance and the development of a thick tap root with clear secondary anatomical compartments that are formed within a short growing period.
Young (primary growth) and mature (secondary growth) plants of M. o cinalis were sampled across 24 sites to evaluate microbiome assemblages from the exterior to the interior niches of the root. The objectives of the present study were (1) to understand the microbiome diversity, composition and the abundance of key components in bulk soil, rhizosphere soil, and endospheric compartments including the periderm, phloem, and xylem; and (2) to evaluate the selective forces applied by host plants that shape the microbiome in each niche and to compare the effects of geographic and environmental variability.

Methods
This study was situated in the Yellow River Delta (YRD) region of Shandong Province, China. As a sediment-laden river, the Yellow River carries the most sediment of any river in the world [27], and forms a large area of oodplain and wetland landscape with frequent channel migrations [28]. M. o cinalis is widely distributed in the YRD, possibly due to its high salt tolerance and biological N 2 xation capability [29]. In the present study, M. o cinalis were sampled from 24 sites across the YRD (Fig. S1). At each site, plants were excavated from a ~200 m 2 area, including young plants (one-year old, 8 to 10 cm height, and 1 mm tap root diameter) and mature plants (two-year old, 30 to 50 cm height, and ~10 mm tap root diameter). Both young and mature plants were at vegetative growth stage when sampling. Bare soils next to the sampled plants were collected as the bulk soil.

Microbial community analysis
The rhizosphere soil was sampled based on the protocol of Bulgarelli et al [13]. Firstly, root sections taken from 2 to 12 cm depth in the soil pro le were shaken until there was approximately 1 mm soil attached to the roots. Then, roots with attached soil were washed in Falcon tubes lled with PBS buffer (135 mM NaCl, 2.7 mM KCl, 1.5 mM KH 2 PO 4 , and 8 mM K 2 HPO 4 , pH 7.2) by shaking at 150 rpm for 30 min. The soil suspension was centrifuged at 8000 rpm (14310 g) for 5 min, and the pellet was collected as rhizosphere soil. The washed roots were transferred to a new Falcon tube, and cleaned further using a sonicator (180W, 20 min, SCIENTZ, SB-3200DT, China). As observed by Edwards et al [14], it is di cult to remove all the bacteria adhering to the root surface without damaging the epidermis, which was likely to have occurred in the present study. Therefore, bacteria that were tightly attached to the root after the sonication process were considered to be the microbial community from the periderm zone.
After sonication, nodules were separated from the root system, and tap roots were processed further. The whole tap root of the young plants was considered as the endosphere. For the mature plants, the 10 cm sampled tap root was cut into 5 × 2 cm sections, and each section was longitudinally split into three anatomical compartments including the periderm, phloem and xylem zones (Fig. 1) using a sterilised microtome (HistoCore AUTOCUT, Leica Biosystems, Wetzlar, Germany). All samples were stored at -80 ºC until DNA extraction.
DNA was extracted from the samples (0.4 g for soil samples and 0.5 g for plant tissues) using the PowerSoil DNA isolation kit (MoBio Labroratories, Carlsbad, CA, USA) according to the manufacturer's instructions. DNA from the same niche of the 10 young roots or 10 mature roots was combined as one sample for the site. In total, there were 216 samples for microbiome analysis: 9 rhizocompartments (bulk soil, rhizosphere -mature plant, nodules -mature plant, periderm -mature plant, phloem -mature plant, xylem -mature plant, rhizosphere -young plant, nodules -young plant, and endosphere -young plant) at each of the 24 sites.
The hypervariable V3-V4 region of the bacterial 16S rRNA gene was ampli ed using the primers 341F (5'-CCTACGGGNGGCWGCAG -3') and 805R (5'-GACTACHVGGGTATCTAATCC -3') [30]. A Qubit uorometer (version 2.0, ThermoFisher Scienti c, USA) was used to quantify the extracted DNA. PCR products were puri ed using magnetic DNA clean beads (ThermoFisher Scienti c, USA) according to the manufacturer's instructions, and then sequenced on the Illumina Miseq TM platform with 300 bp length of PE reads. All the sequence data generated in the present study have been uploaded on National Center for Biotechnology Information with the project ID: PRJNA603423 and accessions ID: SAMN13931281. Sequences of microbiome 16S rRNA genes were analysed on the Quantitative Insights Into Microbial Ecology 2 (QIIME2) platform [31]. Firstly, raw reads were demultiplexed ("q2-demux" plugin), and then trimmed and quality controlled ("q2-cutadapt" plugin). Representative read was picked using DADA2 denoising algorithm ("q2-dada2" plugin [32]) to generate an amplicon sequence variants (ASVs) table. ASVs were annotated against the SILVA reference database (vs. 132 [33]). ASVs assigned to chloroplasts or mitochondria were regarded as organellar ASVs, and were removed from the analysis. Low frequent ASVs (appearing in less than 9 samples, 5% of the total sample number) were also removed.

Plant and soil analysis
At each site, 30 mature plants and 30 young plants were collected for analysis of cell wall components of root tissues. Roots were surface cleaned, and horizontal fractions were separated, both as described in microbiome analysis. The total cell wall and lignocellulose content were determined by a dry-weightbased neutral detergent bre (NDF) method and acid detergent bre (ADF) method, respectively. Afterwards, the permanganate procedure was applied to estimate lignin and cellulose contents in ADF [34]. The bulk soils collected at each site were analysed for chemical properties (total N, P, K, organic carbon, pH and EC) and physical properties (texture) according to the protocol of Rayment and Lyons [35].

Root exudate analyses
Root exudates were collected using the methods of Aulakh et al [36] and Dietz et al [37], targeting plants grown in eld conditions. Five mature plants and 5 young plants were excavated at random at each site. Soils attached to the roots were removed by washing twice, rstly using tap water, and then sterilised deionized water. The intact roots of the plants were transferred into two tubes containing 200 ml sterilised water, one for 5 mature plants and the other for 5 young plants. After 2 hours, the collected exudate solution was ltered through a lter paper (Whatman No. 42), and a membrane lter (0.45 µm) to remove root detritus and microbes. The ltrates were frozen and stored at -80°C until further processing.
Carbohydrates and amino acids in the collected root exudates were analysed using the method of Conselvan et al [38]. Brie y, gas chromatography (GC) was performed to measure carbohydrates and sugars on an Agilent 6890 Gas Chromatograph with QQQ 7000 Mass selective detector with a HP5 column (0.25 mm i.d., 30 m, 0.25 mm lm thickness; Agilent Technologies, Santa Clara, United States).
Split injection was made at 300°C with an initial oven temperature program of 60°C for 2 min, ramping to 300°C at a rate of 10°C min −1 and maintained for 10 min. Column ow rate was maintained at 1.5 mL min −1 . Amino acids in root exudates were analysed by LC-MS on a 1290 In nity LC system coupled to a 6520 QTOF Mass selective detector (Agilent Technologies, Santa Clara, United States). A 3.5 uL sample was injected into a Zorbax SB-C18 column (2.1 × 150 mm, 3.5 µ). The QTOF was set up as the low mass range (<1700 AMU), scan mode (60-1000 m/z), and positive ion mode. Peak integration and the relative quantities were calculated using Mass Hunter software (version B.07.01, Agilent Technologies). The measured abundances of root exudate compounds were adjusted for the trapping period (2 h) and dry weight of the root samples.

Statistical analysis
The microbiome present in nodules was dominated by one ASV (ASV0), which was annotated as Ensifer, accounting for over 97% relative abundance, so nodule samples were not included in the composition and diversity analysis, and ASV0 was regarded as being the symbiotic rhizobia of M. o cinalis. The statistical analysis was carried out using R software version 3.4.3 [39]. The trimmed mean of the Mvalues method [40] from the Vegan package (version 2.4-5 [41]) was applied to normalise the various library sizes between samples. Using the normalised data, principal coordinate analyses (PCoA) were conducted, based on Bray-Curtis distances. To test the effects of niche on ASV composition, permutational multivariate analyses of variance (PERMANOVA) was run with maximum 999 permutations using the adonis function. Similarity percentage analysis (SIMPER) on Bray-Curtis distance with 999 permutations was conducted to identify the contribution of each ASV to the dissimilarity of microbiome composition between two neighbouring niches (Vegan package). The Shannon diversity index of the ASVs in each sample was calculated using the iNEXT package in R [42], which provides asymptotic estimation by rarefying data to account for differences in sequencing depth.
Relative abundances of taxonomic groups at the phylum and ASV level were compared between different niches using the Statistical Analysis of Metagenomic Pro les (STAMP) package [43]. The Benjamini-Hochberg FDR option was selected as the multiple test correction method.
In each of the sampled niches, a distance-decay model was used to test the relationship between microbiome similarity and geographic distance between sites. The model was described as: log 10 (S) = (-2z)*log 10 (d) + b, where S is the quantitative Sørensen index for community similarity calculated as Bray-Curtis matrices here, d is the geographic distance between two samples (meters), and z is the turnover rate. Where the model tness was signi cant (P < 0.05), the z-value was compared between microbial communities based on the t-distribution method [44].
The effect of soil properties (Euclidean distance) or root exudate composition (Euclidean distance) on microbiome composition was tested using partial mantel analysis [41] with geographic distance as a controlled factor, and 9999 permutations for each of the sampled niches.
For variables including Shannon index and Bray-Curtis distances between two niches, the General Linear Model in Minitab (Minitab Inc. State College, PA, USA) was selected to examine the effects of niche. The regression analysis between cell wall components and microbiome Shannon index (relative to bulk soil) was performed for the root compartments including periderm, phloem and xylem, respectively. Each variable was tested and con rmed for normal distribution before running the General Linear Model.

Results
Both young and mature plants of M. o cinalis were sampled from 24 sites in the Yellow River Delta (YRD, Fig. S1). For the mature plants, niches through the soil-root continuum including bulk soil -rhizosphereperiderm -phloem -xylem ( Fig. 1) were separated and analysed for the pro le of bacterial community by 16S rRNA gene sequencing. Due to the lack of secondary growth in young plants, the whole root was considered as the endosphere, and the microbiomes across the bulk soil -rhizosphere -endosphere continuum were evaluated.
Microbial community composition and diversity through soil-plant niches Using amplicon sequence variants (ASVs) as operating units, microbial composition of soil-plant niches in young and mature plants was analysed by principal coordinate analysis (PCoA) based on Bray-Curtis distances. All the samples were clearly separated by niche group in the biplot ( Fig. 2a) with PCoA1 and PCoA2 together explaining 46% of the variation. The ASV composition of bulk soil and rhizosphere samples overlapped substantially, and differed from the root compartments. Among the endospheric microsites, the periderm of the mature roots and endosphere of the young roots harboured similar microbiome compositions, but these were differentiated from the more interior niches, such as the phloem and xylem of the mature root. Permutational multivariate analyses of variance (PERMANOVA) showed that the effects of niche on microbiome composition was signi cant (P < 0.01, R 2 = 0.92).
In addition, the Bray-Curtis distances between adjacent niches of samples collected from the same site were calculated and compared (Fig. 2b). For mature plants, there was a greater change of ASV composition from periderm to phloem than for rhizosphere to periderm and phloem to xylem (P < 0.05), and the microbiome dissimilarity was the lowest between bulk soil and rhizosphere for both young and mature plants.
The Shannon index of ASVs was reduced signi cantly (P < 0.01, ANOVA test) and gradually when moving from the outer microhabitats (bulk soil and rhizosphere) to the inner tissues for both mature and young plants, and the decrease on average was from 642-162 and 1336-29 for mature and the younger plants, respectively (Fig. 3b). The diversity of the rhizosphere microbiome was greater than that of the bulk soil microbiome in young plants (P<0.05, Fig. 3b), while this difference was not signi cant in mature plants.
Geographic location, soil properties, and root exudation of host plants all in uenced the structure of root-associated microbiomes The distance-decay model was used to test how microbiomes changed with geographic distance between sites (Fig. 3b). We identi ed that the data for the outer niches (bulk soil, rhizosphere, and periderm) in mature plants tted the model signi cantly, and the turnover rate (z value, related to the slope of the model) was much lower for the periderm than for the bulk soil and the rhizosphere (P<0.01, Table 1). However, the endophyte compositions of phloem and xylem samples were stable with increasing geographic distance, as the trend cannot be explained by the distance-decay model (P>0.05, Table 1 and Fig. 4a). In the rhizosphere and endosphere of young plants, the trend between microbiome composition and geographic distance tted the distance-decay model (P<0.01), and the changes of rhizosphere-inhabiting microbiota were more sensitive, with a greater z value (P<0.01, Table 1 and Fig. 4b).
There was a large variation in measured soil physiochemical factors among the sampled geographic sites (Table S1). Partial mantel analysis was performed to test the effect of soil properties on microbiome assemblages at the different niches by controlling for the geographic in uence (Table 1). Only the microbial community structure for bulk soil and rhizosphere of the mature plant were signi cantly affected by soil properties (P < 0.05, partial mantel test, Table 1).
In addition, based on the 45 compounds detected in root exudates of young and mature plants (Tables  S4 and S5), there were more exudates (adjusted for root dry weight and trapping period) secreted by the young plants than by the mature plants (P<0.01, Fig S2a). The composition of root exudates was partially separated between young and mature plants across different sites, based on the PCoA analysis ( Fig  S2b). By restricting the geographic effect, partial mantel analysis revealed that root exudate compositions of the young and mature host plants were correlated with variation in microbial consortia in the rhizosphere (P < 0.01, partial mantel test, Table 1).

Core microbiomes of endorhizosphere inhabitants
For each of the studied niches, a core microbiome across the sampled sites was established, based on the most highly abundant and ubiquitous ASVs. In one niche, ASVs that appeared in at least 17 out of the 24 sampled sites (> 70%), and ranked in the top 10% for relative abundance in every site where it was present [45,46] were selected as the core ASVs. The number of core ASVs in bulk soil, rhizosphere and periderm was considerably larger than in the inner tissues (87 to 375 vs 6 to 18, Fig. 5). The lower number of core ASVs in phloem and xylem was possibly due to a lower total number of ASVs colonising these niches, as the proportion of unique core ASVs among total ASVs was highest in phloem and xylem, about 5.4-5.8% compared with 0.9-1.7% for the microbiomes in bulk soil and rhizosphere, and 3.3% in the periderm (Fig S3). Interestingly, about 86% (309 out of 375) periderm core ASVs did not appear as core ASVs in other microsites, while rhizosphere microbiota of young and mature plants shared a large proportion of core ASVs.
Among core ASVs of one niche, those with >5% relative abundance at every site where they were present were selected as the dominant ASVs. Only the niches in root tissues of mature and young plants had dominant ASVs, and totally three dominant ASVs were identi ed. All three dominant ASVs were present in the periderm (Fig. 6a), while the phloem and xylem harboured only one dominant ASV, ASV3 (Kineosporia) and ASV0 (Ensifer), respectively, both of which were also dominant in the endosphere of young plants. ASV0 (Ensifer) was the major taxon in clean nodule tissues of M. o cinalis, with over 97% relative abundance in samples taken from different sites and from plants of different ages. We therefore conclude that ASV0 detected in this study corresponded to the symbiotic rhizobia for M. o cinalis. Here, the symbiotic rhizobia appeared to be detected as ASV0 which was one of the three dominant ASVs (Fig. 6a), and the only core ASV that occurred in the rhizosphere, periderm and xylem of mature plants, as well as in the endosphere of young plants (Fig. 5).
The relative abundance of the three dominant ASVs as root-associated inhabitants was compared in mature and young plants, and the P value was adjusted using Benjamini-Hochberg FDR (Fig. 6b). In mature plants, ASV0 (the symbiotic rhizobia) was highly enriched in the periderm (15%), less abundant in xylem and rhizosphere (2.5-6.2%), and minor in bulk soil and phloem (0.1-0.8%, Fig. 6b). The relative abundance of ASV3 (Kineosporia) was greater in periderm and phloem tissues than in other niches (12-14% vs 0.02-2.4%, adjusted p < 0.01, Fig. 6b). The proportion of ASV10 (Azohydromonas) was 10% in the periderm, but was quite minor in other niches (up to 1.2%, Fig. 6b). In young plants, all three dominant ASVs were remarkably abundant in the endosphere (2.7-30%, adjusted p<0.01), but were present in only trace amounts in the bulk soil and rhizosphere (<0.1%).
Similarity percentage analysis (SIMPER) revealed that the three dominant ASVs were also the top contributors to the dissimilarity of microbiome composition between two neighbouring niches ( Fig. 6c and Table S2-3). For example, the increased relative abundance of ASV0, ASV3 and ASV10, accounted for 15% of microbial community shift from rhizosphere to periderm for the mature plant, and the decrease of their relative abundance contributed to about 40% of microbiome dissimilarity from periderm to phloem. ASV0, the symbiotic rhizobia, was one of the top contributors to the microbiota changes between bulk soil and rhizosphere (1.1% contribution, mature plant), between phloem and xylem (2.1% contribution, mature plant), and between rhizosphere and endosphere (3.9% contribution, young plant). Besides the three dominant ASVs, another microbe (ASV6 annotated as Allorhizobium-Neorhizobium-Pararhizobium-Rhizobium spp.) also affected the microbial community changes along rhizosphere-periderm-phloem, to a great extent (1.5%-5.0%, Table S2).

Cell wall components of root compartments
The total cell wall, cellulose, and lignin content increased from the outer to the inner root compartments (Table S4). In each of the root niches, we tested the correlation between the Shannon diversity index of the microbiome (divided by the value for the respective bulk soil in the same site for normalisation) and the content of cell wall components across sampled sites (n = 24). Microbial diversity was positively correlated with total cell wall and lignin content in xylem ( Fig. 7b and c), but not in other root compartments.

Discussion
Melilotus o cinalis exhibited a clear selection of its root microbiome and there was a preference for colonisation by certain microbial taxa through the continuum of soil, rhizosphere, periderm, phloem, and xylem zones. The periderm zone exerted a stronger selection on the microbial community than other boundaries, such as the transition from bulk soil to rhizosphere, and from phloem to xylem. The geographic distance, soil environment, root exudation activity, and root cell wall components differentially in uenced microbiome composition at the various microsites.
Most previous studies of this type have used the entire sampled root as a single community [13,14,26].
Here, we physically dissected individual roots into separate tissues (periderm, phloem and xylem), and identi ed that the microbiome assembly in the innermost tissues (phloem and xylem) was conserved, with less variation among different geographic sites, compared with differences among the rhizosphere, rhizosphere plane and endosphere components. Both roots with primary growth (young plants) and secondary growth (mature plants) were evaluated for the development of endophyte microbial assemblies through transverse layers of root from the root surface to the central tissues. Our result provides a novel, detailed understanding of host plant -endophytic community interactions.
Selection of the soil and root microbiome from the outer to inner niches The structure of the whole microbial community was considerably modulated from the soil niches (bulk soil and rhizosphere soil) through to the endospheric niches, indicated by the changed composition and reduced diversity. More importantly, the spatial compartments inside the root tissue (periderm, phloem and xylem) differed in the pattern and diversity of microbial colonisation to a greater extent than the differences observed between the two soil microsites (bulk soil and rhizosphere soil, Fig. 2a). Our ndings demonstrated that the selective forces exerted by the root for the soil microbiome occurred not only in the rhizosphere, but also across anatomical boundaries within the root system to form a relatively distinct assemblages of microbial communities throughout the periderm, phloem and xylem. Previous studies on rice [14], Arabidopsis [13] and speargrass species [47] also found that there was a larger change of microbiome from rhizosphere to endosphere than from bulk soil to rhizosphere. Endophytic colonization of plant roots occurs in the intercellular spaces for most bacteria, possibly due to the greater quantity of carbohydrates, amino acids, and inorganic nutrients originating from apoplastic transport [48,49]. The penetration of microbes into internal root tissues involves both passive processes, e.g., through the cracks induced by lateral root emergence, and through active mechanisms, e.g., secreting cell-wall degrading enzymes [6]. In the present study, the increased presence of cell walls from the periderm towards the xylem (Table S4), and the appearance of Casparian strip (Fig. 1) probably contribute to the selective pressure, indicated by the selection of a less diverse microbiome (Fig. 3b). In addition, the microbial diversity of the endophyte communities in young plants were lower than those in the periderm, phloem, or xylem tissues of mature plants, possibly due to fewer lateral roots being initiated in young plants and/or shorter soil residence period of the young plant root.
The periderm must play a crucial role in controlling entry of microbes into the root, because (1) the magnitude of the shift in microbiome structure from periderm to phloem exceeded the changes across other boundaries (Fig. 2b), and (2) there was a larger number of core ASVs only present in the periderm than occurred in other niches (Fig. 5a). The importance of the periderm is that it acts as the rst gate for controlling entry of the microbes from the soil niche to root tissue. Other endospheric boundaries separate two compartments within the root. The effect of the rhizoplane (part of the periderm in our study) on ltering the microbiome has been de ned as a 'critical gating role' in previous studies [14,19].
Bio lm formation and adhesion ability during the plant-microbe interaction occurs on the root surface such as the periderm, which could lead to a more pronounced selection of microbial communities [11].
The effects of geographic location, soil properties, root exudation, and cell wall content on shaping the microbiome of root niches For the niches in close proximity to soils including bulk soil, rhizosphere, periderm (mature plants), and endosphere (young plants), their microbiome assembly was driven by stochastic processes (geographic factors) and deterministic processes (edaphic factors and root activity). However, the colonization and persistence of endophytes in the interior tissues (phloem and xylem) of mature plants were more conserved and stable across environments, and their diversity was in uenced by the amount of cell-wall components formed in different plants.
A distance-decay model was used to describe microbial community similarity in relation to increased geographic distance [50]. Microbiomes inhabiting bulk soil and rhizosphere were more sensitive to geographic change than those of the periderm and endosphere (young plants), indicated by the greater turnover rate in the distance-decay model (Table 1 and Fig. 4), possibly because the microbiome composition in the periderm (mature plants) and the endosphere (young plants) is subject to not only the microbial reservoir of the background soil, which varies across geographic locations, but also to recruitment by the host plant, involving features such as the cell wall (about 37-45% of the total tissue). Moreover, soil properties were only associated with microbiome structure in the two soil-related niches (bulk soil and rhizosphere soil) of the mature plants, but not for the young rhizosphere, possibly because the soil residence period, which has been regarded as the key factor forming the rhizobiome structure for perennial plants [51], was too short for the young plants to modulate their rhizosphere microbiome, compared with the mature plants.
The composition of root exudates in uenced the rhizosphere-inhabiting microbiota in both mature and young plants, while the variation in root exudation activity did not show any exceptional trends with geographic and soil factors. The rhizosphere is rich with diverse metabolites secreted from the root, and this process is controlled by plant genetics, i.e., genes related to synthesis of speci c root exudates [52,53], as well as the heterogeneous growing environments, e.g., soil, climate, and neighbouring plant species [54]. Our ndings here showed that variation in root exudation activity did not affect the structure of the endophyte community, which was more stable across environments than the rhizosphere microbiome.
The apoplast mainly comprises plant cell walls, and bacteria commonly grow in this environment, attached to plant cell walls [55]. Our analysis of chemical cell wall constituents also showed a large variation between the periderm, phloem and xylem, with total cell wall and lignin content increasing from the outer to inner tissues (Table S4). Additionally, the microbiome diversity decreased from phloem to xylem (Fig. 3b) while the total cell wall and lignin content increased (Table S4), so cell walls were likely the structural barriers leading to the selection of certain endophyte species. Lignin is not metabolically active, and its deposition has been regarded as an important mechanism contributing to structural root defences to control microbes [56]. The woody appearance of mature root systems and the increased ligni cation of cell walls from secondary growth leads to a homogeneous root endophytic community across different plant species [8]. For example, around 40% of bacterial endophytes colonising Arabidopsis also inhabited wooden splinters of two tree species incubated in the same soils [13]. Here we demonstrated that increased cell wall and lignin content in the xylem was associated with higher diversity of xylem-inhabiting endophytes (Fig. 7), possibly because cell wall components can act as growth substrates for certain bacteria that can produce cell-wall degrading enzymes.
Selection for speci c microbes from the outer to inner niches The composition of phylum groups differed between soil and root niches, while the variation within the root i.e., among periderm, phloem and xylem tissues, was minor (Fig. 3a). Compared with soil niches, root tissues harboured a larger proportion of Proteobacteria and Actinobacteria, but less Acidobacteria, which is in line with observations for Arabidopsis [13] and rice [14]. In addition, three microbes that were ubiquitous across sampled sites and highly abundant in speci c root niches, were identi ed as Proteobacteria or Actinobacteria, including ASV0 (Ensifer spp., Proteobacteria), ASV3 (Kineosporia spp., Actinobacteria), and ASV10, (Azohydromonas spp., Proteobacteria). These three microbes also act as the top contributors to the shift of microbiome composition between the neighbouring niches, e.g., rhizosphere-periderm, periderm-phloem, phloem-xylem, and rhizosphere-endosphere (young plants).
As an important microbe for legumes, ASV0 (Ensifer spp., Proteobacteria) is the symbiotic nitrogen xing partner for M. o cinalis, and was highly abundant in the periderm (mature plants) and endosphere (young plants Fig. 6a). This was likely related to the nodulation process: formation of the nodule primordium is usually facilitated through activation of cell division in the epidermal, cortical, and pericycle zones [57], that were sampled as the periderm zone (mature plant) and endosphere (young plant) in our study. Additionally, besides the periderm and endosphere, the mean relative abundance of M. o cinalis symbiotic rhizobia was second highest in the xylem, but quite minor in the phloem (located between periderm and xylem). Interestingly, our results are in line with a previous study [58] that during nodule formation there was a nodule-inducing factor present in the protoxylem, and an inhibitor in the protophloem, although the conclusion was based on a model legume plant, Medicago truncatula, with a short lifespan and only primary growth.
Another Kineosporia (ASV3) was enriched in the periderm and phloem of mature plants, and the young plant's endosphere (Fig. 6b). As a non-streptomycete actinobacteria, Kineosporia has been found to inhabit the tissues of plant roots and leaves [59,60], which is possibly associated with its ability to utilise complex substrates [61], such as cell wall components in the present study. Network analysis on Arabidopsis root endophytes showed that Kineosporia was one of the ve keystone taxa determining negative interkingdom connections between bacterial and fungal microbiota [59].
Furthermore, the soil-attached root tissues (periderm of the mature plant and endosphere of the young plant) were colonised by Azohydromonas bacteria (ASV10) with a larger proportion than other niches (Fig. 6c). Azohydromonas are free-living nitrogen xers (containing nifH genes), and have been widely observed in the rhizospheres and endospheres of leguminous and non-leguminous plants [62,63]  HL and RT analysed the soils, and prepared the plant and soil samples for sequencing. PY and ST undertook the root exudation analysis and interpreted the results. MR and HY performed the eld trials and analysed the results. All authors contributed to revising the manuscript, and approved the nal submission. Table   Table 1. The effects of geographic distance, soil properties, and root exudation on the microbiome composition at root-associated niches of Melilotus o cinalis mature and young plants.
A distance-decay model was used to test the relationship between microbiome similarity and geographic distance. The z value is the turnover rate in the model, indicating the sensitivity of microbiome changes with increased geographic distance. A partial mantel test was performed to examine the effect of soil or root exudation by controlling for geographic distance, and the r value is presented. ns: not signi cant, *: signi cant at P < 0.05, and **: signi cant at P < 0.01     The effect of geographic location on microbial composition in soil and root niches of Melilotus o cinalis The distance-decay relationship between microbiome similarity and geographic distance between Page 27/29 sampled sites was tested in (a) mature plants and (b) young plants, respectively. The regression line is presented with 95% con dence intervals.