Synergistic effects of plant genotype and soil microbiome on growth in Lotus japonicus

Abstract The biological interactions between plants and their root microbiomes are essential for plant growth, and even though plant genotype (G), soil microbiome (M), and growth conditions (environment; E) are the core factors shaping root microbiome, their relationships remain unclear. In this study, we investigated the effects of G, M, and E and their interactions on the Lotus root microbiome and plant growth using an in vitro cross-inoculation approach, which reconstructed the interactions between nine Lotus accessions and four soil microbiomes under two different environmental conditions. Results suggested that a large proportion of the root microbiome composition is determined by M and E, while G-related (G, G × M, and G × E) effects were significant but small. In contrast, the interaction between G and M had a more pronounced effect on plant shoot growth than M alone. Our findings also indicated that most microbiome variations controlled by M have little effect on plant phenotypes, whereas G × M interactions have more significant effects. Plant genotype-dependent interactions with soil microbes warrant more attention to optimize crop yield and resilience.


Introduction
The interaction between the microbiome and plant roots is an important factor affecting plant growth.These interactions are perv asiv e and can have extensive effects on host plants, including disease resistance (Santhanam et al. 2015, Busby et al. 2016, Carrion et al. 2019 ), stress tolerance (de Vries et al. 2019, Liu et al. 2020 ), nutrient supply (Zhang et al. 2019 ), and ov er all plant health (Berendsen et al. 2012 ).Consequently, there has been increasing interest in determining how plant-microbiome interactions are established and maintained, as well as which effects the plantr oot inter actions depend on in order to establish sustainable agricultural systems and improve our understanding of ecosystems (Mauchline and Malone 2017 ).
In the complex ecosystem of plant-microbiome interactions, the root microbiome is shaped by an assemblage of factors, namely plant genotype (G), soil microbiome composition (M), and the environmental conditions under which the plant grows (E).Genetic differences between plants are one of the most studied factors affecting the root microbiome structure (Weinert et al. 2011, Bulgarelli et al. 2012, Lundberg et al. 2012, Peiffer et al. 2013, Walters et al. 2018, Brown et al. 2021 ).Phylogenetic distance between plants and root microbiome dissimilarity appears to be correlated at the interspecies, or higher, taxonomic le v els (Bouffaud et al. 2014, Schlaeppi et al. 2014, Terrazas et al. 2020, Wang and Sugiyama 2020 ).In addition, the soil microbiome, being a v ast micr obial pool, substantiall y influences the r oot micr obiome (Edw ar ds et al. 2015 ).Soil factors, including soil type (Sc hr eiter et al. 2014 ), nutrient content (Yeoh et al. 2016, Agri et al. 2022 ), and pH (Qi et al. 2018 ), regulate soil microbe diversity and offer a wide range of microbes that roots can use to form their micr obiome.Macr oenvir onment, comprising of abiotic factors such as soil moisture (Bouskill et al. 2013, Naylor et al. 2017 ) and light conditions (Martin et al. 2018 ), also has significant effects on the r oot micr obiome .T hese en vironmental conditions can affect both plant physiology and the activity and survival of soil microbes, ther eby influencing r oot micr obiome composition.Noting the interconnected influence of these factors, isolating and understanding their individual impact in experimental systems becomes incr easingl y critical.
Man y studies hav e attempted to understand the individual effects of plant genotypes (G), soil microbiomes (M), and environmental conditions (E) on the root microbiome and their impact on plant growth (Lundberg et al. 2012, Yeoh et al. 2016, Naylor et al. 2017, Gallart et al. 2018, Azarbad et al. 2022 ).Ne v ertheless, two main obstacles interfere with our understanding of these complex relationships.First, isolating the M and E factors in field experiments is challenging because of the inherent complexities and interactions between the existing soil microbiome and environmental factors.Significant microbiome alterations have been reported in different soils and in soils exposed to differ ent tr eatments (Rousk et al. 2010 ).Secondly, the scope of plant genotypes used in pr e vious studies, as shown in tomato r esearc h (Oyserman et al. 2021 ), tends to be limited, which may restrict our understanding of the effects of plant genetic variation.Despite meaningful pr ogr ess in understanding the individual effects of these factors, there still is a need to understand their combined and inter activ e effects.Numer ous studies hav e documented inter actions between individual microbes and plant genetic variation affecting plant growth (as reviewed by Bamba et al. 2018 ), suggesting a probable interaction between plant genetic variation and the r oot micr obiome on plant gr owth.Yet, these inter action effects ar e not full y elucidated.Giv en these ga ps, it is essential to use experimental systems that car efull y r econstruct the effects of G, M, and E to better understand their influence on plant phenotypes.
Consequently, this study aimed to e v aluate the impacts of G, M, and E, and their interactions on the root microbiome and plant growth.To disentangle these effects, we established an in vitro experimental system by combining nine geneticall y differ entiated Lotus accessions, three soil microbiomes (and one microbiomeabsent control), and two different environmental conditions .T he soil microbiomes used in this study were extracted from two adjacent fields in northern Japan: one that had been subjected to salt treatment for 3 years and the other that had not.Furthermore, we included a combined community of both microbiomes and a noninoculated control for four inoculation groups .En vironmental conditions were set to reflect those in which the soil microbiomes were collected, with or without salt treatment (adding 100 mM NaCl to the medium).In the present study, plants wer e gr own under 72 combinations of these conditions, and the root microbiome and plant phenotypes were compared.The effects of G, M, and E, and their interactions on the root microbiome and plant growth wer e quantitativ el y assessed by examining the collected data.

Cross-inoculation experiments
A cross-inoculation experiment was performed to quantify the effects of G, M, and E, and their interactions on plant phenotypes and r oot micr obiomes.Nine Lotus accessions with four soil microbial inoculations (two microbiomes from the soil, one mixture, and a noninoculated control) were cultivated under two environmental conditions, resulting in 72 combinations.
Soil micr obiomes wer e obtained fr om soils collected in May 2020 from the Kashimadai fields of Tohoku University (38.46 • N, 141.09 • E) in northern Japan.Soil samples were obtained from two adjacent plots (Field No. 5 Control and Field No. 5 Salt Treated;F5C and F5S,r espectiv el y), wher e L. japonicus has been cultiv ated for the last 3 years .T he salt treatment in F5S was conducted from 2017 to 2019 by irrigation with under gr ound water containing salt at a concentration approximately equivalent to one-fourth of seawater concentr ation, whic h is about 0.8% (8 g/l) (Shah et al. 2020 ).F5C was sim ultaneousl y irrigated with r egular water.A total of 5 kg of soil were collected from five sampling points far enough a wa y fr om eac h other to avoid sampling bias.Prior to sample collection, at least 15 cm of soil were first r emov ed to avoid sampling surface soil.The soils collected from each respective location were crushed as finely as possible using a hammer.The soil samples wer e separ ated into 250 g batches of each soil and crushed using a mixer and 250 ml of cold phosphate-buffered saline (PBS).The crushed soils were precipitated by centrifugation at 1000 × g for 10 min at 10 • C, and the supernatants were collected.The precipitates were returned to the mixer and the process was repeated three times to extract the microbes in the clumps .T he collected supernatant was filtered using Advantec 5A filter paper (particle size > 7 μm; ash content < 0.01%).The filtered solutions were centrifuged at 8000 × g for 20 min at 10 • C, and the pelleted microbes (fungi and protists should be removed in the filtration and precipitation pr ocesses) wer e collected and diluted with 250 ml of PBS to obtain a 1 ml per g soil microbial community extract (250 ml of extr act fr om 250 g of soil).Micr obial comm unity extr acts fr om F5C, F5S, and a 1:1 mixtur e (MIX) wer e used for the cross-inoculation experiments.As the difference in OD600 values between the F5C and F5S extracts was less than 2%, their concentrations were not adjusted.
To establish differences in the growth environment corresponding to the soils used for the extraction of microbiomes, two types of media (salt-treated and control) were used for plant cultiv ation.Initiall y, two 20 l tanks of B&D medium (Broughton and Dilw orth 1971 ) w ere prepared, each enhanced with 1 mM KNO3 to r estrict symbiotic inter actions with nitr ogen-fixing nodule bacteria, thereby ensuring plant growth independence from these bacteria.Subsequently, NaCl was added to one tank to ac hie v e a concentration of 100 mM.The medium from each tank was then distributed into four 5 l tanks, totaling eight tanks, which were autocla ved to sterilize .Six of these tanks were supplemented with one of three microbial communities-F5C, MIX, and F5S-at a 1% (v/v) concentration, with two tanks designated for each SALT and non-SALT condition.The medium from these eight tanks was then allocated into 19 pots filled with vermiculite, culminating in 152 pots .T hese pots were utilized for transplanting the plants in the ensuing experiments.
P artl y scrubbed Lotus seeds were sterilized by immersion in 2% sodium hypochlorite for three min and rinsed three times with sterile MilliQ water.After overnight immersion, the swollen seeds w ere so wn on 1% agar plates, incubated in the dark for 3 days at 25 • C, and then grown at the same temper atur e under 16/8 light/dark conditions for 24 h.Rooted plants were transplanted into pots with lids, filled with 300 ml of sterilized vermiculite and 250 ml of media, and grown at 25 • C under the same light conditions for 4 w eeks.Gro wth pots were closed with lids to prev ent cr oss-contamination.Tw o pots w er e used for eac h plantmicr obiome-envir onment combination.A total of 144 pots (nine plant accessions × four inoculants × two conditions) were simultaneousl y gr own in a gr owth c hamber.The 144 pots wer e r andomly assigned to 10 groups (14 or 15 pots per group) and placed on tra ys .To pr e v ent une v en placement of the pots, pot positions in the tray and the position of the trays in the growth chamber wer e r otated e v ery week.Furthermor e, once e v ery 2 weeks, the groups of pots were changed randomly.To avoid heterogeneous lighting conditions, a growth chamber with lights on the top was used and photon flux densities ranging from ∼140-150 μmol/m 2 /s were maintained.Six plants were cultivated per pot.The detailed growth system was shown in Fig. S1 ( Supporting Information ).
The whole plant bodies wer e harv ested, all individuals wer e imaged using a high-resolution scanner, and their roots and root nodules wer e separ ated.Individuals whose first leaves could not be observ ed wer e consider ed as dead plants were not included in observational data, since they did not possess sufficient roots to detect their r oot micr obiomes .T he shoot length (SL), number of leaves (NOL), number of br anc hes (NOB), and r oot length (RL) wer e measur ed fr om the scanned data as plant phenotypes .T he r oots wer e washed with sterilized distilled water, frozen in liquid nitr ogen, and pr eserv ed at −80 • C until DNA extr action.

DN A extr action for MiSeq sequencing
A total of six to se v en living individuals were selected from each combination of G, M, and E (72 combinations, three individuals of noninoculant samples) to identify root microbiomes; when sufficient individuals were collected from two pots, the selection of individuals fr om eac h pot was k e pt equal as possible.Prior to DNA extr action, eac h r oot sample was cut into ∼2 cm pieces and randomly collected into sterilized tubes.Genomic DNA of each root sample was extracted using the Qiagen MagAttract 96 DNA Plant Core Kit (QIAGEN Inc., Valencia, CA, USA) according to the manufacturer's instructions.In addition, three DNA samples were extr acted fr om eac h of the soil extr acts (F5C and F5S; Day 0) used in the experiments.Six DNA samples for each combination were extr acted fr om the soil collected fr om nonplant pots (F5C, MIX, and F5S in both saline and nonsaline envir onments), whic h had been k e pt stationary while the plants wer e gr owing.Eac h soil microbiome genomic DNA sample was extracted using the FastDNA Spin Kit for Soil (MP Biomedicals, Illkirch, France).
P air-end libr ary pr epar ation for MiSeq sequencing was conducted using the two-step tailed PCR method described by Illumina (San Diego, CA, USA).The following primer pairs were used to amplify partial sequences of the 16S rRNA gene: V5F_MAUI_799 (forw ar d): 5 -TCGTCGGC AGCGTC AGA TGTGT A TAA GA GA CA GNNNHNNNWNNNHAA CMGGATTA GATA CCCKG-3 V7R_1192 (r e v erse): 5 -GTCTCGTGGGCTCGGAGA TGTGT A T AA GA GA CA GA CGTCATCCCCA CCTTCC-3 The 3 end to 18 bases and 19 bases of each primer (forw ar d and r e v erse, r espectiv el y) wer e 16S rRNA universal sequences (Chelius and Triplett 2001 ).The 19-30 base from the 3 end of the forw ar d primer is a unique molecular identifier (Fields et al. 2021 ).The other regions were the Illumina Overhang adapter sequences.Second-round PCR was performed using primer pairs with 16 unique indices: D501-D508 and A501-A508 (forw ar d) and D701-D712 and A701-A712 (r e v erse) (Illumina).
The DNA concentrations of the purified PCR pr oducts wer e adjusted and pooled into two different tubes, as the MiSeq run was performed in two separate runs .T he 16 × 24 indexing system used could not accommodate six samples per condition, necessitating the division to analyze our sample set.The samples used in the experiments are listed in Table S1 (Supporting Information ).The pair ed-end libr aries wer e mixed with 3% PhiX DNA spike-in control and used for sequencing on the MiSeq platform using Illumina MiSeq v.3 Reagent kit for 2 × 300 bp PE.All the sequence files obtained in this study have been deposited in the DDBJ database (DRR511003-DRR511426).

Da ta anal ysis for microbiome
Quality control was performed for the sequenced reads and pair ed-end r ead assembl y using PEAR v0.9.6 (Zhang et al. 2014 ).The low-quality tails in each read were trimmed using a Phred score of 20 as the threshold, and trimmed reads with lengths less than 200 bp were discarded.P air ed-end r eads with an overlap of more than 10 bp and a total length of more than 300 bp were combined.The UMI, primer, and target sequence regions of eac h r ead wer e identified based on sequence length.UMIs were counted without duplication, and the abundance of each sequence was determined based on the number of UMIs .T his process can inhibit multiple counts of amplified products derived from the same molecule, thus reducing the PCR amplification bias.Sequences with an av er a ge number of UMIs greater than or equal to 0.05 for all samples were chosen and used in the following analyses.UMI identification and counting were performed using Python 3 software ( https:// github.com/mbamba2093/ Mauiseq _ for _ microbiome ).
A BLAST searc h (Camac ho et al. 2009 ) was conducted for each sequence using a database containing the RDP11 bacterial 16S rRNA sequences (Cole et al. 2014 ) and L. japonicus Gifu genome v1.2 (Kamal et al. 2020 ) to assign the sequences to microbial taxa.Part of the classification rank of RDP11 that was out of alignment was manually corrected.All taxa, except for species-le v el assignment (taxonomic le v els wer e as follo ws; phylum, or der, class, family, and genus), were compared with the List of Prokaryotic names with Standing in Nomenclature (Parte et al. 2020 ) to prevent misclassification, and those that did not match were marked as "No-tAssigned."For the BLAST r esults, m ultiple sequences had the highest matc h r ate, and the sequence with the exact genus name or earliest RDP ID was selected.The sequences that showed the highest match rate in the BLAST search of the L. japonicus Gifu genome but not in the RDP11 database were plant-derived and were excluded.All remaining sequences were analyzed as amplicon sequence variants (ASVs).Based on the BLAST top hit results, each ASV sequence was categorized into taxonomic groups (domain, phylum, class, order, family, and genus), and sequences with the same top hit were classified as OTU.

Microbial community analysis
Comm unity anal yses wer e performed to e v aluate the effects of combinations of G, M and E, and their interactions with the plant r oot micr obiome .T he follo wing analyses w ere performed using data a ggr egated fr om sequences with the same O TU, genus , family, and order.
Prior to the analysis, to reduce biases due to differences in sampling depth, the community data was subsampled based on the r ar efaction curv e using the rarefy function implemented in vegan R (Oksanen et al. 2020 ).The sample with the lo w est community cov er a ge was identified based on the slope at the endpoint of the r ar efaction curv e and the number of reads was adjusted to match this slope for all samples (Chao and Jost 2012 ).The r ar efied community data was converted into frequency data.
The effects of G, M, and E on div ersity wer e calculated using a generalized linear model (GLM), excluding noninoculant and soil samples.In the GLM, α-diversity was the response variable and the effects of G, M, and E and their interactions were explanatory variables.Gamma distribution was chosen as the error distribution and the log-link function for the model.Statistical significance was e v aluated using the F -test.A separ ate GLM was constructed for the soil samples .T he first model was used to e v aluate differences between F5C and F5S at the time of inoculation.The second model was used to e v aluate the effects of M and E and their interactions 28 days after inoculation.For significant variables in the F -test ( P < .05), the Tuk e y-Kramer test was conducted to compare α-diversity among all groups .T he vegan package (Oksanen et al. 2020 ) were used to calculate diversities and the glm , Anova , and glht functions in R 3.6.1 (R core team, 2019 ) were used to estimate the effects of G, M, E, and interactions.
To distinguish root microbiome structures, β-diversities among samples were calculated using the Morisita-Horn index (Horn 1966 ) since the index is likely to be resistant to undersampling bias (Wolda 1981 ) and the data fr om short-r ead sequences could be ambiguous with other phylogeny-based β-diversities.To visualize the similarity of microbiomes, a nonmetric multidimensional scaling (nMDS) analysis was conducted using the metaMDS function in vegan R (Oksanen et al. 2020 ) with 100 random parameters .In the nMDS analysis , four differ ent assessments wer e conducted: one incor por ating all samples, one excluding noninoculated samples, one excluding both noninoculated and soil samples, and another utilizing only soil samples.PERMANOVA with the adonis function in vegan, R (Oksanen et al. 2020 ) was used with 99 999 permutations to evaluate which factors shape the micr obiome structur e. PERMANOVA was conducted exclusiv el y with r oot micr obiome samples , excluding noninoculated samples .To estimate the effects of variation within L. japonicus species on the r oot micr obiome, β-div ersity anal yses wer e performed using these data, excluding L. burttii.These anal yses wer e also conducted for eac h micr obiome-envir onment combination to clarify the effects of the host genotype in different combinations.
In addition, the following permutation analyses were performed to deal with potential confounding factors caused by each pot, since the individual plants in the same pot shared a unique environment.One of the pots from each combination was selected to exclude pot bias and G, M, and E and their interaction effects on αand β-diversity were evaluated using the GLM model and PERMANOVA for 1000 permutations.
Furthermor e, the corr elation between host genomes and root micr obiome differ ences in eac h micr obiome-envir onment combination was also investigated using the Mantel test implemented in the a pe pac ka ge in R (P ar adis and Sc hliep 2019 ).Genetic distances of L. japonicus genomes were calculated using identical-by-state kinships based on the population genome information reported by Shah et al. ( 2020 ).The pairwise similarity distances of the micr obiomes wer e calculated using the Morisita-Horn index, whic h was calculated by av er a ging the microbiomes of each host.
The effects of G, M, E, and their interactions on the frequency of individual bacteria (OTU-le v el classification) were estimated using a nonparametric regression model called the generalized smoothing model (GSM).Bacterial OTUs observed in more than six individual plants were selected for analysis to exclude excessive r esults fr om bacteria with minor distributions.In the model, eac h bacterial frequency was the response variable and the effects of G, M, and E and their interactions were explanatory variables.Nominal spline smoothing was chosen as the smoothing function for all explanatory variables, with default settings for the other parameters.Statistical significance was evaluated using the F -test.Model fitting and F -tests were implemented in the npreg package in R3.6.1 (Heiwig 2022 ).Fisher's exact test was used to e v aluate whether the significantly affected strains were distributed disproportionately in specific families using R3.6.1.

Da ta anal ysis for plant phenotypes
Heat maps were first generated using host-standardized phenotypic values, whose mean values for each host genotype were set to zero to visualize the variation in phenotypes.Correlations among phenotypes were estimated using Pearson's productmoment correlation.To detect the effects of G, M, and E on the correlations among phenotypes, separate correlation tests were performed for each G, M, and E group.Heatmaps were illustrated using the heatmap.2pr ogr am implemented in gplots in R3.6.1 (R Core Team 2019 ).Correlation analyses were performed using the function implemented in ggpairs of R3.6.1 (R Core Team 2019 ).
To analyze the effects of G, M, and E and their interactions on plant phenotypes (we focused on only SL and RL because all phenotypes wer e corr elated with eac h other), a GLM was used instead of analysis of variance (ANOVA), because the distribution of phe-notypic values deviated significantly from a normal distribution (Sha pir o-Wilk test, P -v alue < .05,for all phenotypes).In the GLM, each phenotype was the response variable and the effects of G, M, and E and their interactions were explanatory variables.Gamma distribution was chosen as the error distribution and log link function for all phenotypes because the distribution did not deviate from the expected distribution.We calculated the type II sums of squares for each variable, evaluated their statistical significance using F -tests, and estimated each variable's effect size ( η 2 ).In addition, the Tuk e y-Kramer test was performed to compare plant phenotypes among the G, M, and E groups .T hese analyses of variance were performed using the Anova function implemented in the car library (Fox and Weisberg 2019 ) and the etaSquared function implemented in the lsr library in R.3.6.1 (R Core Team, 2019 ).
The Tuk e y-Kramer test was performed using the glht function implemented in the multcomp library in R.3.6.1 (Hothorn et al. 2008 ).The same analysis was performed using a dataset that excluded noninoculated individuals to e v aluate the effects of differences in the inoculation community.
In addition, the following statistical analyses were performed to deal with potential confounding factors caused by each pot, since the individual plants in the same pot shared a unique envir onment.The interclass corr elation coefficients (ICC: v ariance between pots/all variance) of pots for each combination of inoculation tests (72 G × M × E combinations) wer e e v aluated using the glmer function in R3.6.1.The ICCs were calculated for two plant phenotypes, SL and RL, owing to the low variance in the other phenotypes.Even if there was bias due to the combination of pot effects, a m ultile v el anal ysis containing pot information as a random effect was unsuitable since the pot variables completely masked the combination information.One of the pots from each combination was r andoml y selected to exclude pot bias and G, M, and E and their interaction effect were then evaluated using the GLM model for 1000 permutations.
As both plant phenotypes and microbiomes depend on the effects of G, M, and E and their interactions, the extent to which root micr obiome structur es explain the variance in plant SL was calculated.The variance component was calculated using the following equation: where Y is a SL vector standardized for each host accession, ε is an error term and μ is the similarity matrix of the root microbiome based on 1 -the Morisita-Horn similarity index matrix and the identical matrix used in the comm unity anal ysis .T he emma function in the R pipeline was used to calculate the variance component u (Kang et al. 2008 ).

Results
A cross-inoculation experiment was performed using nine Lotus accessions (G) and three microbiome inoculants with one noninoculant control (M) under two environmental conditions (E), resulting in 72 combinations.749 individuals (6-12 individuals per combination) were collected ( Table S1 , Supporting Information ).Although 12 plants were cultivated for each combination, ∼9% of these plants did not survive.On the roots of collected individuals, a few nodules observed (average 0.12 nodules on the root of one Lotus indi vidual, exce pt for noninoculant).Comparison of α-diversity of a root microbiome between combinations of two groups; combinations of host genotypes (G) and environmental conditions (E), that of (G) and microbiome inoculants (M), and that of (E) and (M), r espectiv el y. (D) Comparisons of α-diversity of a soil microbiome .T he gray boxplots indicated the diversity at the time of extraction from soils and colored plots indicated the soil microbiome after 28 days from inoculation.

Plant root microbiomes and effects of G, M, and E
The r oot micr obiomes of 382 plants wer e inv estigated.Using MiSeq sequencing, 45 031 087 r eads wer e obtained, pr epr ocessed, and allocated to each individual (ranging from 13 878 to 198 897 per individual).All quality-filter ed r eads wer e used to count the UMI.In addition, 68 023 unique sequences with an av er a ge number of UMIs greater than or equal to 0.05, were used for the BLAST searc h.As a r esult of the BLAST searc h, 62 836 sequences consisting of 16 785 973 r eads wer e deriv ed fr om the bacterial 16S rRNA genes.Bacterial sequences were assigned to 5891 different bacterial OTUs (sequences assigned the same BLAST top hit were a ggr egated as an OTU-le v el taxon), 277 gener a, 85 families, 41 orders , 22 classes , and 11 phyla ( Supplementary material s).The lar gest pr oportion of the micr obiome was Pr oteobacteria (r oot micr obiome, 85%; soil micr obiome at extraction, 61%; and soil microbiome at 28 d, 79%), follo w ed b y Bacteroidetes and Firmicutes, with these three phyla accounting for ∼90% of the community ( Fig. S2 , Supporting Information ).Meanwhile, Actinobacteria were observed in the soil microbiome (8.8% at extraction and 4.1% at 28 days), but r ar el y in the r oot micr obiome (0.03%).Furthermor e, our analysis detected few Mesorhizobium bacteria (occupying ∼0.1% of the micr obiome), whic h ar e common rhizobial symbionts of L. japonicus (Bamba et al. 2019 ).
Prior to the div ersity anal ysis , co v er a ge-based r ar efaction was performed and the lo w est slopes at the end of the r ar efaction curv es wer e 0.1665, 0.0275, 0.0016, and 0.0005 for the ASVs, O TU, genus , and family levels, respectively ( Fig. S3 , Supporting Information ).The α-diversities of the Lotus root microbiome were calculated using the Shannon index (Shannon and Weaver 1949 ) based on r ar efied composition data without noninoculant root samples .T he α-diversity at the micr obiome OTU le v el r anged fr om 1.643 to 7.376 (Fig. 1 ).The α-diversity of the soil microbiomes w as lo w er than that of the r oot micr obiomes, and the div ersity of the F5S inoculants was lower than that of F5C at the time of soil extraction (Fig. 1 ).At 28 da ys , the α-diversity of the soil microbiome did not differ between F5S and F5C in nonsalt environments (Tuk e y test; P = .9658),and the diversity was higher in the MIX-inoculated soils (MIX-F5C: P = .0031and MIX-F5S: Ho w e v er, the div ersities of the F5C-and MIX-inoculated soils were much lo w er in the salt environment than in the nonsalt environment (Tuk e y test; both P < .0001),and the diversity of the F5S-inoculated soils was slightly higher in the salt environment than in the nonsalt environment ( P = .3791).In the F5C inoculated soils, 73% of OTUs were less frequent in salt environments than in nonsalt en vironments .T his result is in line with the expectations of the experimental design, that microbiomes from salt-stressed environments do not decrease in diversity when exposed to nonsalt en vironments , whereas microbiomes from nonsalt environments are stressed in salt en vironments .
The same pattern for the α diversity of the root microbiome as for the soil microbiome was observed, but the effect was limited.A significant effect of G and C on the α-diversity was detected, whereas their interactions and the environment had no effect ( P < .05;Table S2 , Supporting Information ).The Tuk e y-Kramer test indicated that the root microbiomes of MG20 were more diverse than those of Gifu and MG68 and that the microbiomes with MIX wer e mor e div erse than those with F5C inoculants ( P < .05;Table S3 , Supporting Information ).The α-diversities at other taxonomic le v els ar e listed in Table S1 ( Supporting Information ).On the other hand, only 20% of the microbes that were less frequent in saline environments in F5C-inoculated soils were also less abundant in the plant root microbiome grown in F5C-inoculated soils.
The community structures of the root microbiomes were c har acterized based on the βdiversity (Morisita-Horn index).The nMDS analysis sho w ed no ov erla p between the noninoculated samples and soil or r oot micr obiomes ( Fig. S4 , Supporting Information ).This a ppar ent differ ence suggests that the micr obiome detected in the noninoculated samples was either not removed by seed coat sterilization or was derived from contaminating bacteria during the experiment.The same data analysis without the noninoculated samples confirmed the root and soil microbiome community structures (Fig. 2 ).At the time of soil extraction (day 0 samples), a clear difference was observed between F5C and F5S micr obiomes, indicating differ ences in the micr obiomes betw een the tw o soils used in the experiment (Fig. 2 D).A total of 4 weeks after inoculation, the nonsalt environment sho w ed no appar ent differ ence fr om the 0 days micr obiomes, ho w e v er, in the salt environment, the microbiomes deviated from the 0 days community, and the degree of deviation was greater for MIX and F5C than for F5S (Fig. 2 D).
Ta ble 1. P ermanov a r esults for v ariation in r oot micr obiomes.In the root microbiomes, the nMDS analysis sho w ed an appar ent differ ence between envir onments (E) and among inoculants (C), whereas the differences between hosts (G) were unclear (Fig. 2 A-C).This pattern was observed across all taxonomic le v els, although differ ences in the C effect were attenuated for famil y le v el classifications ( Fig. S5 , Supporting Informatio n).PER-MANOVA analysis indicated that G, M, and E, and their interactions significantly affected the root microbiome structure ( P < .05;T able 1 ; T able S4 , Supporting Information ).At the OTU le v el microbiome, the effects of C and E wer e the lar gest, explaining about 22% of the variance .T he others (G, G × M, G × E, M × E, and G × M × E) were around 4%, and 35% of the variance was residual.This result was comparable to that of the community structure of the root microbiomes of L. japonicus ( Table S5 , Supporting Information ), indicating that the root microbiome of L. burttii accessions did not de viate fr om that of L. japonicus .Ev aluating the G effect in conditions where the combination of M and E was fixed showed that the differences in G could explain 25%-40% of the variation in microbiome composition ( Table S6 , Supporting Information ).In addition, differences in the root microbiome among host plants did not correlate with the genetic distances between host accessions ( P > .05;Fig. S6 and Table S7 , Supporting Information ).

DF
To identify which bacterial OTUs were affected by G, M, and E; that would be masked by the β div ersity anal ysis, these effects wer e e v aluated using a nonpar ametric r egr ession model, the GSM, in which the response variable was bacterial frequency.Of the 5510 O TUs (O TU-le v el classification that was observed in more than six plant individuals), 4712 were significantly affected by the G, M, and E v ariables and their inter actions ( Table S8 , Supporting Information ).The G variable had a significant effect on 218 of these strains; ho w ever, 2741 and 3318 strains were affected by M and E, r espectiv el y.Details of the ov erla p of each effect are shown in Fig. S7 ( Supporting Information ) as a Venn dia gr am.The str ains affected by G, M, and E were shared by G vs. M (99 OTUs), G vs. E (122 OTUs), and M vs. E (1558 OTUs) ( Fig. S7A , Supporting Information ).Variables containing G (G, G × M, G × E, and G × M × E) significantly affected 269 OTUs ( Fig. S7B , Supporting Information ).Variables containing M and E (M, E, M × E, and G × M × E) significantly affected 4643 OTUs ( Fig. S7C , Supporting Information ).Three bacterial families , Bacillaceae , Fla vobacteriaceae , and Methylophilaceae, wer e observ ed to be significantl y sensitiv e by v ariables containing G (Bacillaceae: G, Flavobacteriaceae: G × M × E, and Methylophilaceae: G × M and G × M × E. Fisher's exact test FDR-P < .05;Table S9 , Supporting Information ).These families accounted for 0.07%, 2.16%, and 1.57% of the microbiomes, r espectiv el y, r anking 22nd, 8th, and 10th out of 85 families, r espectiv el y.

Plant phenotype and effects of G, C, and E
Four phenotypes were obtained (SL, RL, NOL, and NOB) from 749 individuals (Fig. 3 ).All phenotypic traits were positively correlated (Pearson's product-moment correlation: P < .001;Fig. S8 , Supporting Information ).All combinations of phenotypic traits, except those between RL and NOB, wer e significantl y corr elated in groups G, M, and E (Pearson's product-moment correlation: P < .05;Figs S8 -S10 , Supporting Information ).Because all four phenotypes wer e corr elated, SL and RL w ere used in the follo wing analyses as r epr esentativ es of shoot and root phenotypes, respectively.The phenotypic distributions in all combinations w ere sho wn in Fig. S11 ( Supporting Information ).
In the cross-inoculation experiment, significant effects of G, M, and E and their interactions on SL and RL were detected using the GLM, except for M × E on SL ( F -test P < .05;Table S10 and Fig. S12 , Supporting Information ).The most prominent effects on SL and RL wer e observ ed for G.The second largest effect on SL was on E, whereas that on RL was on M ( Table S10 , Supporting Information ).The coefficients of salt addition as an E factor wer e positiv e for SL and RL, and the Tuk e y-Kramer test indicated significant differences between the salt and nonsalt environments ( P < .001;Table S11 , Supporting Information ).It should be noted that the Lotus genus has different sensitivities to salt in soils between closely related species and within species, with MG20 and L. burttii being less inhibited by salt and Gifu being more inhibited.These results were consistent with those reported by Melchiorre et al. ( 2009 ).The C coefficients for F5C, MIX, and F5S in the GLM were mostly negati ve, exce pt for F5C in RL.The Tuk e y-Kramer test sho w ed significant differences between noninoculated and inoculated conditions ( P < .001;Table S11 , Supporting Information ).This result indicated that inoculation of the microbiomes obtained from the Tohoku fields had adverse effects on plant growth.
The GLM without noninoculant data sho w ed that all G, M, and E cases and their inter actions significantl y affected SL and RL, except for M × E in SL (Table 2 ; Fig. S13 , Supporting Information ).The largest effect size among these models was G.The second largest effects on SL and RL were E and G × M, r espectiv el y.The M v ariables sho w ed that the differences in the inoculant microbiomes in this model had a less significant effect on plant phenotypes (Fig. 4 ).The η 2 of variables M for SL was 0.008, less than "small" by Cohen 1998s guideline (small: 0.01 < , moderate: 0.01 < 0.06, and Large: 0.06 < ); meanwhile, that for RL was 0.047, which is moderate .T he η 2 of variable G × M on SL and RL were 0.041 and 0.062, whic h ar e assigned to "moder ate" and "lar ge," r espectiv el y.These   In this study, the potential confounding factors derived from each pot could have caused an overestimation of all effects because pot differences masked all G × M × E combinations.First, the extent to which the variation in plant phenotypes could be explained by the differences in pots for each G × M × E combination was calculated.On av er a ge, 15% and 8% of the variance in SL and RL, r espectiv el y, was deriv ed between pot replicates, indicating that variation existed between them.To assess if variations among pot replicates could lead to an ov er estimation of the effects, we conducted an analysis where one pot from each treatment combination was randomly selected and evaluate the impacts of G, M, E, and their interactions on plant phenotypes.Although the G × M × E effects could not be distinguished from the pot effects during the permutations, other effects could be estimated by considering the v ariability deriv ed fr om the pot effects.From the permutation analysis results, variance was observed for all effects, whereas the distribution of each effect was comparable to that of the full dataset ( Fig. S14 , Supporting Information ).For SL and RL, the effects of G were the largest.The second largest v ariables wer e E and G × M for SL and RL, r espectiv el y.A comparison of the effects of G × M and M on permutations sho w ed that the number of trials with G × M > M was 95.1% in SL and 100% in RL, suggesting that the plant phenotypes were more sensitive to the interaction between G and M than M alone.
As both plant phenotypes and microbiomes depend on the effects of G, M, and E and their interactions, we attempted to integrate the variation in SL and root microbiome structure with v ariance component anal ysis.SL v alues standar dized b y the G factor were used to calculate SL variation because this factor explained a large amount of SL and little of the root microbiome structure.Variation in root microbiome structure was calculated based on the Morisita-Horn similarity index matrix, which is identical to the matrix used in community analysis.In this analysis, 35% of the variance in SL could be explained by the similarity in root microbiome structures .T his result indicates that identifying whic h micr obes could affect plant gr o wth w as difficult, e v en though many types of microbes in the soil microbiome would have favorable or adverse effects on plant phenotypes.

Discussion
In the present study, we found that the synergistic effects of plant genotype and soil microbiome have larger impact on plant growth but different impacts on root microbiomes via 72 G × M × E combinations of in vitro inoculation experiments .T he initial differences observed between the inoculant microbiomes at the point of collection from the soil allo w ed us to conduct an experiment that disentangled the effects of G, M, and E, and we could e v aluate their highly independent impacts on root microbiomes and plant phenotypes.Ho w e v er, it should be noted that this in vitro experiment deviated from the interactions observed in the field experiment, in which Lotus was grown directly at the site where the inoculated community used in this study was collected (Bamba et al. 2020 ).The dominance of Proteobacteria observed in this study was also observed in the field, but was e v en mor e pr onounced during the growing period.In contrast, the Actinobacteria observed in the field decreased during the gr owing period, especiall y in the root microbiome.Our experimental methods explained these unique patterns .T he gro wth pots w er e filled with v ermiculite and media and k e pt anaer obic; these conditions ar e unfavor able for most Actinobacteria that are aerobic (Trujillo 2016 ).Despite these disparities, our compr ehensiv e in vitro a ppr oac h, and impr ecise reconstruction of in natura interactions, this comprehensive appr oac h not only helps to clarify the individual effects of these factors , but also pro vides valuable insight into the interactions that under pin the structur e of the r oot micr obiome and its impact on plant growth.
The majority of the Lotus root microbiome was determined by the sole effects of M and E, with a single effect of G, whereas the effects of their interactions remained minor.The small G effects and minor interaction effects on root microbiomes shown in our in vitro pot experiments were corresponding to the field experiments of Azarbad et al. ( 2022 ).In the present study, the response to E (salt tr eatment) differ ed between the soil and r oot micr obiomes, with a larger interaction effect between M and E in the soils .T he α-diversity of the soil microbiomes sho w ed that the soil derived fr om the salt-tr eated field, F5S, contained mor e micr obes ca pable of overcoming salt stress than the soil derived from F5C.This suggests that the expected interaction between M and E was observ ed, e v en with the experimental material used in this study.Ho w e v er, its effect on the plant root microbiome was changed, and some microbes that declined in the F5C soil microbiome under salt treatment did not decrease in the roots even under salt treatment.This suggests that the microbial sensitivity to the environment changes because of their recruitment to the soil by plants.Modification of the root microbiome in response to environmental stress by the plant has been demonstrated in a previous study (Naylor and Coleman-Derr 2017 ), indicating that our study potentiall y observ ed the influence of envir onmental factors on the r oot micr obiome, both dir ectl y and indir ectl y, via the plant.
The r elativ el y small effect of G on the r oot micr obiome is also consistent with pr e viousl y r eported findings in m ultiple studies (Weinert et al. 2011, Bulgarelli et al. 2012, Lundberg et al. 2012, Peiffer et al. 2013, Walters et al. 2018, Brown et al. 2021 ).Our inv estigation e v aluated the effect of G in the presence of M and E effects; ho w e v er, e v en when including their inter activ e effects, the magnitude of G did not r eac h that of M and E. T his ma y be attributed to the low abundance of microbes sensitive to G effects (such as Enterobacteriaceae (0.7%) identified in this study) within the comm unity.Furthermor e, the lac k of a significant correlation between genetic distance within Lotus species and the similarity of root microbiomes, coupled with no clear differences in root microbiomes between Lotus and L. burttii , suggests that the genetic differ ences gov erning G effects may not be fixed among these closely related species.Typically, larger genetic dissimilarities among plants coincide with larger differences in root microbiomes (Bouffaud et al. 2014, Schlaeppi et al. 2014, Terrazas et al. 2020, Wang and Sugiyama 2020 ), and species with low intraspecific div ersity, suc h as Boec hera stricta , exhibit less noticeable G effects (Wagner et al. 2016 ).Considering that Lotus and L. burttii are not highl y differ entiated, it is suggested that onl y genetic v ariations underlying G effects on certain micr obes, suc h as Enter obacteriaceae , ha v e accum ulated, indicating that the contribution of the Lotus genotype to the community structure remains relatively small.
Lotus growth may be more sensitive to plant-microbe interactions depending on the G than the differences among the encountered M. The smaller effect of M on plant phenotypes, particularly plant shoot phenotypes, than that of G × M indicated that the majority of the altered microbes had little effect on plant phenotypes; ho w e v er, their inter action with G had mor e significant outcomes.These G × M effects differed between plant shoot and root phenotypes; the root phenotype was more sensitive to differences in the encounter ed micr obiome than shoot phenotypes .T his suggests that plant roots interact directly with soil microbes and respond to them in a genotype-dependent manner.In contrast, the interaction effects can be buffered/facilitated by each host genotype and spread to the shoots.
How can the G × M effects be gener ated?Ther e ar e two possible pathways thr ough whic h a particular microbe can exert its Gdependent effects .T he first is when the effect of microbes on the plant is constant, but their frequency of localization to the roots depends on the plant genotype.In the present study, a few significant genotype-dependent microbes were detected, and if they ar e effectiv e on plants , this pathwa y is likel y to follow.Secondl y, the effect of microbes on plants depends on the plant genotype, regardless of the frequency of the microbes.For instance, nodule bacteria ar e likel y to follow this latter pathway because their effects are genotype-dependent (Bamba et al. 2020 ) (e v en though the effects of nodule bacteria were excluded in this study).The G × M effect is exhibited if the effects produced by these pathways are M-dependent.There are two possible scenarios for this pr ocess: either micr obes showing genotype-dependent effects ar e distributed differ entl y in eac h soil micr obiome or, ther e was no difference in the distribution of microbes among different soil microbiomes, but genotype-dependent effects were observed in a particular micr obiome, suc h as micr obe-micr obe inter actions in the comm unity.Ev en though ar ound 68% of bacterial OTUs were distributed in different inoculant communities and could support the former scenario, it is still challenging to determine which scenario each G × M-related strain would follow.
It was also challenging to e v aluate the effect of each bacterium on plant phenotypes because G, M, and E and their interactions affected both plant growth and the root microbiome and did not allow us to separate them.Ho w e v er, the most notable difference in plant growth in this study was observed as a negative effect in the inoculated groups compared to the noninoculated groups.This was not consistent across all plant accessions; a positive effect on plant growth was also detected (specifically in MG11, MG63, and MG67 under salt str ess).Consequentl y, by extr acting micr obial comm unities fr om the soil and e v aluating the effects of M, it is possible to identify adv anta geous inter actions for plant growth.The ability to evaluate such effects in noninoculated groups is also one of the adv anta ges of conducting experiments in an in vitro system such as the one used in this study .Accordingly , more detailed experiments and analyses, such as inoculation studies using synthetic communities (Finkel et al. 2017 ), would be more efficient in clarifying whic h micr obes can affect plant phenotypes .Furthermore , by focusing on the natural diversity of L. japonicus (Shah et al. 2020 ), we can elucidate the genetic loci underlying the effects of G, G × M, and G × M × E on plant phenotypes and r oot micr obiomes .T his a ppr oac h would be v aluable for disentangling the shape and maintenance of plant-microbiome interactions in nature.

Ac kno wledgements
Wild accessions of L. japonicus used in this r esearc h wer e pr ovided by the National BioResource Pr oject ("Lotus/Gl ycine") of the Ministry of Education, Culture, Sports, Science, and Technology, Japan.

Figure 1 .
Figure1.α-diversities of root and soil microbiomes.α-diversity based on the Shannon index with an OTU-level taxonomic assignment.(A), (B), and (C) Comparison of α-diversity of a root microbiome between combinations of two groups; combinations of host genotypes (G) and environmental conditions (E), that of (G) and microbiome inoculants (M), and that of (E) and (M), r espectiv el y. (D) Comparisons of α-diversity of a soil microbiome .T he gray boxplots indicated the diversity at the time of extraction from soils and colored plots indicated the soil microbiome after 28 days from inoculation.

Figure 2 .
Figure 2. Micr obiome structur es based on β-div ersity.nMDS for Lotus r oot and soil micr obiome dissimilarity (Morisita-Horn index) is shown.(A), (B), and (C) Colors r epr esent differ ent plant genotypes, micr obiome inoculants , and en vironmental conditions , and sha pes r epr esent the sample types.(D) Soil microbiome structures.Shapes represent environmental conditions in the experiments, and triangle shapes indicated the microbiomes at the time of extraction.Colors represent the microbiome inoculants, or original soils for the samples at the extraction.

Figure 3 .
Figure 3. Plant phenotypic variation in the cr oss-inoculation experiments.Heatma ps of the four plant phenotypes: (A) SL, (B) RL, (C) NOL, and (D) NOB.Each cell color indicates standardized phenotypic values for each plant genotype.

Figure 4 .
Figure 4. Effect sizes of G, M, E, and their interactions on plant phenotypes .T he portion of each color and number on the bar chart represent the effect size η 2 of each variable in the GLM without noninoculant data.

Table 2 .
GLM for plant phenotypes in the cross-inoculation experiment without noninoculant data.
a Degree of freedom.b Sums of squares.c Shoot length.d Root length.