2.1. Study design and logistics

Our main objective was to develop and test methods to reproducibly study plant microbiomes within the sterile EcoFAB 2.0 device (Fig 1a). We hypothesized that the inclusion of Paraburkholderia sp. OAS925, a dominant B. distachyon root colonizer into SynCom [11], would reproducibly influence the microbiome assembly, metabolite production, and plant growth across multiple laboratories using the EcoFAB 2.0 device. To test the hypothesis, we deployed the grass B. distachyon with a SynCom consisting of 16 or 17 members (either with or without Paraburkholderia sp. OAS925) that was originally developed to span the diversity of bacteria isolated from a grass rhizosphere, including representatives from the Actinomycetota, Bacillota, Pseudomonadota, and Bacteroidota phyla (Fig 1b) [9]. Our study was conducted across five laboratories (designated A–E) and consisted of four treatments with seven biological replicates each (Fig 1a): an axenic (mock-inoculated) sterile plant control, SynCom16-inoculated plants, SynCom17-inoculated plants, and plant-free medium control. Each laboratory followed written protocols and annotated videos, gathered root and unfiltered media samples for 16S rRNA amplicon sequencing, filtered media for metabolomics, measured plant biomass, and performed root scans. At the end of the study, the collected data and samples were sent to the organizing laboratory for sequencing, metabolomics, and data analysis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Overview of the interlaboratory comparison study.

(a) Experimental design where five laboratories across three continents conducted the same experiment using shipped materials. These included a detailed protocol (https://dx.doi.org/10.17504/protocols.io.kxygxyydkl8j/v1) [15], SynComs and Mock solution stocks, light and temperature loggers, Brachypodium distachyon seeds, EcoFAB 2.0 device parts, and various lab supplies (growth medium, filters, sampling tubes). We inoculated B. distachyon plants with either a 16- or 17-member SynCom, with controls being axenic plants and medium-only technical control (Mock-inoculated), n = 7. We tested sterility and imaged roots at multiple time points. Finally, we quantified plant biomass, analyzed exudate metabolite composition, and measured root and medium microbiomes. (b) The phylogenomic tree is based on 120 marker genes, where SynCom members are highlighted in bold, with phylum-level classification shown by colored strips and SynCom membership by circles (SynCom16 in blue, SynCom17 in orange). The 2 closest taxonomic genomes are included, with GenBank accession numbers in parentheses. Nodes with over 50 bootstrap support values from 100 replicates are labeled. Chloroflexus aggregans (bold gray) served as an outgroup. The 17 genomes can be accessed via Hugging Face (https://doi.org/10.57967/hf/5885) [16] or links in S2 Table.

https://doi.org/10.1371/journal.pbio.3003358.g001

The detailed protocol with embedded annotated videos used by all five laboratories is available via protocols.io (https://dx.doi.org/10.17504/protocols.io.kxygxyydkl8j/v1) [15]. The general procedure follows these steps: (i) EcoFAB 2.0 device assembly; (ii) B. distachyon seed dehusking, surface sterilization, and stratification at 4 °C for 3 days; (iii) Germination on agar plates for 3 days; (iv) Transfer of seedlings to the EcoFAB 2.0 device for an additional 4 days of growth; (v) Sterility test and SynCom inoculation into the EcoFAB 2.0 device; (vi) Water refill and root imaging at three timepoints; (vii) Sampling and plant harvest at 22 days after inoculation (DAI). Since differences in labware and material can cause experimental variation, the protocol specifies the part numbers used in this study. Organizers provided critical components, including growth chamber data loggers, in the initial package of nonperishable supplies, while the SynComs and freshly collected seeds were shipped just before the study. Given the time zone differences, it was difficult to synchronize all activities, so each laboratory performed the experiment independently within 1.5 months of each other (S1 Table). All participants followed data collection templates and image examples.

2.2. Protocols resulted in reproducibly sterile conditions

During the study, all participating laboratories tested the sterility of the EcoFABs 2.0 devices by incubating spent medium on Luria–Bertani (LB) agar plates at two different time points. Less than 1% (2 out of 210) of all tests showed colony formation (Fig 2a). Namely, a single colony was observed in one treatment of laboratory D in SynCom17, and multiple colonies for laboratory B in medium-only control (plate had cracked lid).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Plant phenomics and EcoFAB 2.0 device sterility.

(a) Sterility of uninoculated EcoFAB 2.0 devices tested across laboratories A–E and treatments (Trt.: Ax-Axenic, 16-SynCom16, 17-SynCom17, Tx-Medium technical control) at day 0 and day 22 after inoculation (DAI). The medium from these devices was incubated on LB agar plates for 22 days to observe bacterial colony formation. Each square represents one test with gray fill indicating sterility and black contamination. The photos of the agar sterility test can be found at https://doi.org/10.6084/m9.figshare.26409220. (b) Plant biomass weight combined across all laboratories (Lab A–E in different colors), measured as shoot dry weight, shoot and root fresh weight. One-way ANOVA with Tukey test, n = 7, ns p > 0.5, *p < 0.05, **p < 0.01, ***p < 0.001. The shoot photos can be found at https://doi.org/10.6084/m9.figshare.26409310. (c) Root system development was analyzed using RhizoNet (Lab B–E) and ImageJ (Lab A). The raw root pixel counts were normalized to the maximum value in each lab. Two-way ANOVA with Dunnet’s test vs. Axenic control, n = 7, ns p > 0.5, *p < 0.05, **p < 0.01, ***p < 0.001. The root scans and Rhizonet reports can be found at https://doi.org/10.6084/m9.figshare.26131291. The data underlying this figure can be found at https://doi.org/10.6084/m9.figshare.26401315.

https://doi.org/10.1371/journal.pbio.3003358.g002

2.3. Reproducible plant growth

When plant biomass data were combined across laboratories, we observed a significant decrease in shoot fresh weight and dry weight of plants inoculated with SynCom17, and to a lesser extent SynCom16, relative to the axenic treatment (Fig 2b). This said, we did observe some variability between laboratories (S1 Fig), which is presumably due to growth chamber differences including light quality (fluorescent versus LED growth lights), light intensity and temperature (S1 Table). Supporting this, the data loggers revealed variability in measured temperatures (S2a Fig) and photoperiod (S2b Fig). Image analysis of scanned roots (S3 Fig) revealed that SynCom17 caused a consistent decrease in root development observed after 14 DAI onwards (Fig 2c).

2.4. Reproducible microbiome assembly

SynComs were prepared using optical density at 600 nm (OD₆₀₀) to colony-forming unit (CFU) conversions (S2 Table) to ensure equal cell numbers (final inoculum 1 × 10⁵ bacterial cells per plant) and shipped on dry ice to each laboratory as 100× concentrated stocks in 20% glycerol. The cells were resuspended and added to 10-day-old B. distachyon seedlings in the EcoFAB 2.0. After 22 days of growth, the roots and media were sampled, shipped back to the organizing laboratory, sequenced (see S3 Table for read counts), and compared to the original inoculum. For both SynComs (SynCom16 and SynCom17), the community composition at 22 DAI differed from the inoculum (Fig 3). As hypothesized, the root microbiome inoculated with SynCom17 was dominated by Paraburkholderia sp. OAS925 across all laboratories (98 ± 0.03% average relative abundance ± SD). In its absence (SynCom16), other isolates showed high relative abundance in the root microbiome with increased variability across laboratories, namely Rhodococcus sp. OAS809 (68 ± 33%), Mycobacterium sp. OAE908 (14 ± 27%), and Methylobacterium sp. OAE515 (15 ± 20%). The most dominant microbial isolates detected in root samples were also typically present in the media samples (S4a Fig). Ordination plots showed clear separations between SynCom16 and SynCom17 microbiomes for both root and media, with generally higher variability between samples for the SynCom16 microbiome (S4b Fig). There was a minimal contribution of unknown reads in all samples, consistent with the observed sterility of the controls. The three samples with the highest proportion of unknown reads (>2.5%) contained operational taxonomic units (OTUs) associated with human microbes (S4 Table), suggesting introduction during handling. Furthermore, SynCom17 treatment in laboratory D did not show unknown reads (S3 Table), suggesting that the failed sterility test (Fig 2a) was likely caused by plate downstream contamination.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Root microbiome.

Microbiome composition of Brachypodium distachyon roots and starting inoculum of plants inoculated with SynCom16 or SynCom17 (±Paraburkholderia sp. OAS925). Letters indicate different laboratories, with each biological replicate shown (n = 7). The inoculum shows technical replicates (n = 3). The data underlying this figure can be found at https://doi.org/10.6084/m9.figshare.26401315 or in S3 Table.

https://doi.org/10.1371/journal.pbio.3003358.g003

2.5. Reproducible rhizosphere metabolome

The spent medium from each fabricated ecosystem was filtered and shipped to the organizing laboratory for LC–MS/MS analysis (S5 Table), followed by targeted and untargeted metabolomics to determine the root exudate composition and metabolite profiles in the presence of different SynComs in the rhizosphere. The targeted analysis identified 60 metabolites spanning diverse metabolite classes (S6 Table). Hierarchical clustering revealed general clustering by treatment and not laboratory (Fig 4), consistent with the experimental reproducibility observed with plant growth phenotypes and root microbiome composition. Furthermore, the metabolite clustering showed several treatment-dependent metabolite changes. The first large cluster included diverse metabolites increased in the SynCom17 treatment. A second large cluster consisted of metabolites with lower relative concentrations in the SynCom17 treatment, represented mainly by amino acids. A third, much smaller cluster consisting primarily of nucleosides(tides) increased in the SynCom16 or both SynCom treatments. This finding highlights the prominent impact of the community dominated by Paraburkholderia sp. OAS925 on modulation of metabolite composition in the rhizosphere. This was further supported by untargeted metabolomics on 833 detected features that showed a clear separation between rhizosphere metabolomes of axenic plants and SynCom17, which was reproducible across all laboratories (S5 Fig). These changes may be due to metabolite production or uptake by the microbes or plant roots or the activity of extracellular enzymes [17–19].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Targeted metabolomic analysis of rhizosphere.

We show mean values for 60 identified metabolites for each lab/treatment combination (n = 7), row-normalized to the average sum peak height per lab. Row colors indicate metabolite classes. Cluster 1: Abundant in SynCom17; Cluster 2: Low in SynCom17; Cluster 3: Abundant in SynCom16 or both SynComs. The data underlying this figure can be found at https://doi.org/10.6084/m9.figshare.26401315 or in S6 Table.

https://doi.org/10.1371/journal.pbio.3003358.g004

2.6. Statistical evaluation of variation between labs

Ordination analysis using non-metric multidimensional scaling (NMDS) for microbiome (S4b Fig) and metabolome data (S5 Fig) showed a clear separation between treatments, but some separation between labs. To explore this in more detail, we quantified the lab- versus treatment-associated variation by analysis of variation (ANOVA) (S7 Table), pairwise tests (S8 Table), and coefficient of variation (CV).

ANOVA analysis of microbial abundance showed that more microbes were significantly different due to the treatments versus lab doing the experiment, across both roots and media (S6a Fig). When investigating the source of variation, we observed microbe-dependent results, with the lab accounting for <18% of the variation, while SynCom treatment had a larger contribution in several microbes, especially Praburkholderia, Rhodococcus, and Methylobacterium (S6b Fig). CV analysis showed consistent trends for both root and media samples, with SynCom16 displaying higher within-lab variability than SynCom17 in all labs (S6c Fig). Between-lab CVs were similar across treatments. Overall, the observed microbial abundance levels were more dependent on treatment than lab.

Metabolite levels were more affected by the lab-associated factors than the microbial abundances. However, treatment was still the primary influence in the majority of the metabolites. Both lab and treatment significantly affected the majority of the 60 identified metabolites (S7a Fig). Comparing these results between the various labs, we found that 7–23 metabolites were significantly different in specific pairwise lab comparisons, with lab pair A–B being the most different, and labs A–D, C–D, and B–D most similar to one another (S7b Fig). This said, most metabolites were primarily influenced by treatment rather than lab, though a subset, especially salsolinol, tyramine, and dopamine showed notable lab effects (S7c Fig). Metabolites in Axenic and SynCom17 treatments had lower CV than SynCom16 in all labs, and the within-lab CV was approximately the same as the between-lab CV (S7d Fig). Together, these results support the conclusion that metabolite abundances were largely treatment-driven, though some metabolites were sensitive to lab-specific differences.

2.7. Colonization by Paraburkholderia

Given the reproducible impacts of Paraburkholderia sp. OAS925 on our fabricated ecosystems, including plant growth phenotypes, microbiome structure, and rhizosphere exometabolites, we performed additional analyses to gain insights into potential mechanisms that may explain its dominance. Comparative genomic analysis showed that the Paraburkholderia sp. OAS925 genome (IMG/M Taxon ID: 2931840637) uniquely includes acid resistance genes such as glutamate and arginine transporters and decarboxylases (S8 Fig) and a gene module coding for a Type 3 Secretion System (T3SS), which was not found in any other member of the SynCom (S9 Table).

We inoculated B. distachyon with a red fluorescent protein (RFP) expressing Paraburkholderia sp. OAS925 to investigate spatial-temporal root colonization in EcoFAB 2.0. Clear RFP signals were detected at the root tip and in the maturation zone at 1 DAI, with increased biofilm formation observed at 3 DAI (S9 Fig). We noted both sessile colonies on the rhizoplane (S1 Video) and active swimming surrounding root cells (S2 Video).

The biofilm formation and motility observed during microscopy motivated the follow-up in vitro assays to further assess these characteristics across isolates. Paraburkholderia sp. OAS925 exhibited the sixth-highest biofilm formation and the most liquid-culture growth (S10 Fig), highlighting its potential to outgrow many other bacteria in SynCom17 [20].

Swimming motility assays on soft agar revealed that Paraburkholderia sp. OAS925 had the highest motility within the first 24 h (S11a Fig). Additionally, compared to other isolates with similar motility phenotypes (S11b Fig), it maintained fast swimming in acidic conditions (S11c Fig) in the range of the hydroponic medium (pH 5.5–6.0 at the start of the experiment). KEGG mapping (map02040) showed the presence of flagellar assembly genes, suggesting that the observed motility is due to flagella.

4.1. Preparation of synthetic bacterial communities

The bacterial isolates originate from the switchgrass rhizosphere and are available at DSMZ as individual strains [9] or as a collection (DSM 200000SY, inquire at contact@dsmz.de). Glycerol stocks had their 16S gene (27F - 1492R) sequenced to confirm isolate identity and purity. Each verified isolate was streaked on a preferred medium (S2 Table), and a single colony was inoculated into liquid culture. After 2–8 days at 27 °C, cultures were centrifuged at 5,000 g and washed with ½ Murashige and Skoog (MS) basal salts medium. The washed cultures were used to create SynComs by measuring OD₆₀₀, converting to CFU, and combining them to 2e7 cells/ml each (1:1 ratio) in 20% glycerol for cryopreservation [9]. We used SynComs composed of equal numbers of cells for the constituent strains, which is a widely used approach [9,36,37]. A 20% glycerol in ½ MS basal salts was a mock solution. Solutions were aliquoted (100 µL) in microcentrifuge tubes and stored at −80 °C. The participants diluted the solutions 100-fold, assuming 50% cell survival during freezing-thawing, resulting in a final theoretical CFU of 1e5/plant. Serial dilution established the OD₆₀₀ to CFU conversions, and a hemocytometer was used to measure Gottfriedia sp. OAE603.

The inoculums were distributed to five participants: Lawrence Berkeley National Laboratory (LBNL), USA (organizer); the University of Melbourne, Australia; the University of North Carolina at Chapel Hill, USA; Forschungszentrum Jülich, Germany; and the Max Planck Institute for Plant Breeding Research, Germany. Empty tubes were shipped first to estimate speed, dry ice sublimation, and select vendors. Shipments included content declaration, receiver cover letter, and country-specific permits. Identifying the closest phylogenetic species for each isolate helped avoid customs delays. Shipments to Germany were classified as biosafety level 1 (BSL1) and sent via FedEx International Priority with 8.2 kg of dry ice as a delivered-at-place with a proforma invoice detailing freight charges and goods value. The SynCom import to Australia required a permit for conditionally non-prohibited goods under the Biosecurity Act 2015 Section 179 (1) and was sent via Aeronet Worldwide with 22.68 kg of dry ice, which was refilled during the 9-day transit. The shipment to North Carolina used FedEx Standard Overnight with 9 kg of dry ice.

4.2. Experimental setup and plant growth conditions

Participating laboratories assembled EcoFAB 2.0 devices and followed the experimental protocol (https://dx.doi.org/10.17504/protocols.io.kxygxyydkl8j/v1) [15]. In summary, B. distachyon Bd21-3 seeds were surface-sterilized, plated on ½ MS basal salts with 1.5% phytoagar, and stratified for 3 days at 4 °C. The plates were then moved to the growth chamber with a 14 h photoperiod (16 h for lab C) at 26 °C and 10 h dark at 20 °C with photosynthetic photon flux density (PPFD) at 110–140 µmol/m²/s and 70% humidity if tunable. Each lab used two data loggers (HOBO MX2202 and UA-002-64) to track illuminance in lux normalized to maximum value per lab (for dark duration confirmation) and temperature. After 3 days, germinated seeds were transferred to sterile EcoFAB 2.0 devices with 9 mL of ½ MS basal salts (pH 5.5–6.0) and placed back in the chamber. After 4 days, plants were inoculated with Mock, SynCom16, or SynCom17, resuspended from 100 µL stocks to 10 ml using sterile ½ MS basal salts. Each EcoFAB 2.0 device received 1 ml of inoculum, resulting in a glycerol concentration of 0.02% in 10 ml of hydroponic medium. The treatment groups included: i) Mock-inoculated axenic plant control, ii) SynCom16-inoculated plants, iii) SynCom17-inoculated plants, and iv) Mock-inoculated technical control (plant-free). The sterility of the uninoculated devices was tested at 0 and 22 DAI by incubating hydroponic medium on LB agar plates for 22 days. Evaporated water was re-supplied and roots imaged by scanning every seven days. Some laboratories shifted their root imaging timepoints based on scanner availability. Plant harvest and sampling happened at 22 DAI.

4.3. Root phenotyping

To automate root analysis from flatbed scanner images, we used RhizoNet, a workflow for precise root segmentation, ideal for tangled roots in EcoFAB 2.0 devices [38]. Lab A’s scans faced issues with condensation and reflections, causing low contrast; thus, SmartRoot V4.21 was used instead [39]. The raw root measurements from both methods were normalized to the maximum value for each lab.

4.4. Samples collection and shipment

Sample collection and initial processing were performed independently in each lab. We collected 50 µL of the unfiltered growth medium for amplicon sequencing. The remaining spent medium was filtered via a 0.2 µm PES filter for metabolomic analysis, collected in polypropylene tubes (lab A: VWR 21008-103, labs B–E: VWR 93000-026) and stored at −80 °C before shipment. The root and shoot were separated during plant harvest, fresh weights were measured, roots were frozen for microbiome analysis by amplicon sequencing, and shoots were lyophilized for dry weight measurement. The filtered medium, unfiltered medium, and frozen roots were shipped to LBNL on dry ice with gel packs. The loggers were shipped separately at room temperature. To import intact frozen B. distachyon roots to the USA required a Controlled Import Permit (PPQ Form 588), supplier declaration, and TSCA Certification. Samples from Germany were shipped via DHL Medical Express (2 days) on 10 kg of dry ice, from Australia via Cryopdp (5 days) on 24 kg dry ice with a refill request, and from North Carolina via FedEx Priority Overnight on 9 kg dry ice (1 day).

4.5. Analysis of rhizosphere metabolites

At the LBNL, hydroponic medium samples were lyophilized (Labconco, Kansas City, MO). Empty tubes were used as extraction controls. During extraction, samples were kept on dry ice, and Solvents were chilled at −20 °C. The dried material was suspended in 1mL methanol (MX0486, Sigma), vortexed, and transferred to a microcentrifuge tube, followed by residue collection with an additional 0.5mL methanol. Samples were then sonicated (97043-944, VWR) in ice water for 15 min, centrifuged at 10,000g for 5 min at 10 °C, and supernatants transferred to new tubes. Supernatants were dried overnight by vacuum concentration (SpeedVac, Thermo). The following morning, samples were resuspended in 150 µL methanol containing internal standard mix (S5 Table), vortexed, and centrifuged at 10,000g for 5 min at 10 °C. Supernatants were filtered with 0.22 µm PVDF filters (UFC30GV, Millipore), transferred to amber glass vials with inserts (5,188−6,592, Agilent), and sealed with screw caps (5,185−5,820, Agilent). Samples were analyzed by LC-MS/MS, with polar metabolites separation using hydrophilic liquid interaction chromatography (column 683775-924, Agilent) on an Agilent 1290 HPLC system, followed by detection in positive ion mode on a Thermo Orbitrap Exploris 120 Mass Spectrometer (S5 Table). Samples were injected in positive mode, with methanol blanks between each sample; internal and external controls were used for quality control.

For untargeted metabolomics, the mzML files were processed via MZMine 3.0 [40] to create a feature list using a custom batch process (S1 File). The features with MS2 spectrum were then annotated in GNPS2 using spectral metabolite libraries [41]. Features with MS2, retention time (RT) > 0.6 min, and exudate sample max peak height >10× of extraction and technical controls were included, resulting in 833 features across all samples and labs. For targeted metabolomics, metabolites were identified (S6 Table) by analyzing the data with an in-house library of m/z, RT, and MS2 fragmentation information from authentic reference standards using Metabolite Atlas (https://github.com/biorack/metatlas) [42]. Only metabolites with a maximum exudate sample peak height >3× of extraction and technical controls were included. The identified metabolites were manually classified using the PubChem Classification Browser (https://pubchem.ncbi.nlm.nih.gov/classification/).

4.6. Microbiome analysis by amplicon sequencing

DNA was extracted from ground roots and media using the DNeasy PowerSoil Pro Kit (Qiagen), following the manufacturer’s instructions with minor modifications. Ground roots were mixed with a resuspension buffer, transferred to bead-beating tubes, and frozen at −80 °C. Samples were then thawed at 60 °C and bead-beaten using the FastPrep-24 system for 30 s at setting 5.0 (MP Biomedicals). The elution buffer pre-heated to 60 °C. Samples were extracted in batches with water-only samples as negative process controls. Glycerol stocks for 16- and 17-SynCom and a glycerol-only mock were included for time-zero data.

PCR amplification of V4 amplicons was performed in two steps. Library amplicons were generated using the Illumina i7 and i5 index/adapter sequences with V4 primers 515F (Parada) (GTGYCAGCMGCCGCGGTAA) and 806R (Apprill) (GGACTACNVGGGTWTCTAAT) [43]. Amplification used pooled primers on a Bio-RAD CFX 384 Real-Time PCR system with QuantiNova SYBR Green PCR kit (Qiagen) in 10 µL reactions with primers at 4 µM and mitochondrial and chloroplast PNA blockers (PNA Bio) at 1.25 µM. Amplification was initiated at 95 °C for 3 min, followed by the cycle: 95 °C for 8 s, 78 °C for 10 s, 54 °C for 5 s, and 60 °C for 30 s, followed by fluorescence measurement. Root samples were amplified for 22 cycles and media samples for 30 cycles with at least 1 replicate. To allow direct comparison, the SynCom16 and SynCom17 inocula were each amplified for both 22 and 30 PCR cycles, mirroring the amplification conditions used for root and media samples, respectively. Libraries were purified at least twice to remove excess primers using the Mag-Bind TotalPure NGS beads (0.8×), and targets were quantified using the QuantiFluor dsDNA System (Promega). Libraries were then pooled (i.e., root and media samples separately) and purified again. The concentration of the pooled library was determined using the Qubit dsDNA Assay Kit (BR or HS, Invitrogen). The flowcell was sequenced on the Illumina MiSeq sequencer using MiSeq Reagent kits, V3 (600-cycle), following a 2 × 00 indexed run recipe. MiSeq reads were processed using Usearch (v11.0.667) [44]. Reads were merged, trimmed, and short sequences removed using the ‘fastq_mergepairs, fastx_truncate, and fastq_filter’ commands. An initial OTU table was generated with the ‘fastx_uniques, cluster_otus, and otutab’ functions, and OTUs were assigned to SynCom members using the ‘annot’ function with V4 reference sequences. The three samples with the most unknown reads were analyzed via NCBI Microbial Nucleotide BLAST with Representative Genomes database and MegaBLAST algorithm to identify potential matches. The relative abundance of each SynCom member was calculated as the proportion of total microbial reads (excluding plant reads) per sample.

4.7. Growth and biofilm formation assays for bacterial isolates

Lab A conducted the biofilm formation assays. The crystal violet assay for biofilm formation was modified from a previous method [45]. Isolates were grown in R2A, washed, and resuspended in a 30 mM phosphate buffer. They were inoculated into the plates with NLDM medium [20] at a 1:10 (v/v) ratio (final volume of 100 µL and initial OD₆₀₀ of 0.02) and incubated statically at 30 °C for 3 days (n = 4–5) and growth was measured at OD₆₀₀. After incubation, wells were washed 3× with MilliQ water, air-dried, and stained with 125 µL of crystal violet solution (0.1% v/v crystal violet, 1% v/v methanol, and 1% v/v isopropanol in Milli-Q water) for 30 min at RT. Wells were rinsed 3× with MilliQ water, destained with 125 µL of 30% acetic acid for 30–60 min at RT, and absorbance at 550 nm (OD₅₅₀) of the destaining solution was measured.

4.8. Genomic analysis of bacterial isolates

All isolates were grown in R2A except Bradyrhizobium sp. OAE829, which was grown in 1/10 R2A. High molecular weight genomic DNA was extracted from bacterial pellets with the Monarch HMW DNA Extraction Kit (New England Biolabs) or the MasterPure Complete DNA Purification Kit (Lucigen). The genomic DNA was submitted to the JGI for sequencing using PacBio Sequel II or Illumina NovaSeq S4. Genomes were stored in the Genomes OnLine Database (GOLD) [46], followed by submission to the Integrated Microbial Genomes and Microbiomes (IMG/M) (https://img.jgi.doe.gov/) for annotation [47]. The annotated genomes can be accessed via Hugging Face (https://doi.org/10.57967/hf/5885) [16] or IMG/M under the listed Taxon ID or GOLD Project ID (S2 Table): Arthrobacter sp. OAP107 (2931867202, Gp0588953), Gottfriedia sp. OAE603 (2931797537, Gp0588949, formerly known as Bacillus sp. OAE603 [11]), Bosea sp. OAE506 (2931782253, Gp0589672), Bradyrhizobium sp. OAE829 (2931808876, Gp0589676), Brevibacillus sp. OAP136 (2931855177, Gp0588951), Paraburkholderia sp. OAS925 (2931840637, Gp0589681, formerly known as Burkholderia sp. OAS925 [9]), Chitinophaga sp. OAE865 (2931817136, Gp0589677), Lysobacter sp. OAE881 (2931823763, Gp0589678), Marmoricola sp. OAE513 (2931787146, Gp0589673), Methylobacterium sp. OAE515 (2931791092, Gp0589674), Mucilaginibacter sp. OAE612 (2931861231, Gp0588952), Mycobacterium sp. OAE908 (2852593896, Gp0440934), Niastella sp. OAS944 (2931847253, Gp0589682), Paenibacillus sp. OAE614 (2931801854, Gp0589675), Rhizobium sp. OAE497 (2931775946, Gp0589671), Rhodococcus sp. OAS809 (2931833612, Gp0589680), Variovorax sp. OAS795 (2931827682, Gp0588950).

Comparative genomics was conducted using genome statistics and KEGG pathways protein-coding gene abundance from IMG/M. Based on the genes involved in acid resistance of Escherichia coli, including Glutamine/Glutamate and Arginine membrane transporters (GadC and AdiC), glutaminase (YbaS), and decarboxylases (GadA, GadB, and AdiA) [29], we searched for genes annotated with similar functions (GltIJKL, HisPMQ-ArgT, GlnHPQ, glsA, GAD, AdiA). KEGG module coverage was compared between Paraburkholderia sp. OAS925 and the other 16 genomes using the “Statistical Analysis” tool from IMG [48] using Fisher’s exact test and modules with a corrected p-value <10⁻⁰⁵ were manually inspected for a potential link to plant root colonization.

A phylogenomic tree of the 17 SynCom members was constructed using the GTDB-tk workflow [49], incorporating 120 marker proteins for multiple sequence alignment. Fasttree [50] generated the tree, which included two closely related genomes for context, and it was visualized with Interactive Tree Of Life (iTOL) [51]. The phylum-level taxonomic classification is indicated for each member. Bootstrap values, derived from 100 replicates, are displayed for nodes with over 50 bootstrap support values. Chloroflexus aggregans served as an outgroup to root the tree.

4.9. Fluorescent microscopy

Lab A conducted the plant growth and microscopy. The pGinger plasmid 23100 containing the RFP gene under the kanamycin (Kan) resistance marker was introduced into Paraburkholderia sp. OAS925, as described previously [52]. Briefly, 1mL of Paraburkholderia sp. OAS925 grown overnight at 30 °C in R2A medium was mixed with 1 mL of E. coli S17 dapE-harboring the pGinger plasmid grown overnight at 37 °C on LB medium with Kan at 50 µg/mL and diaminopimelic acid (DAP) at 300 µM. The mixture was pelleted for 1 min at 10,000g and then resuspended in 100 µL water with 300 µM DAP. This mixture was then placed onto an R2A agar plate and incubated overnight at 30 °C. The bacterial mix was then scraped, resuspended in water, and plated on R2A with Kan 50 µg/mL. Transconjugants were verified via fluorescent microscopy and colony PCR.

The Paraburkholderia sp. OAS925 expressing RFP was grown overnight in 7 mL of R2A medium with Kan 20 mg/L, shaken at 200 rpm at 27 °C. The cells were pelleted by centrifugation at 4,000g for 10 min, resuspended in ½ MS basal salts, and used to inoculate 3-day-old plants in EcoFABs 2.0 at a starting OD₆₀₀ of 0.01. A flatbed scanner captured root architecture to indicate microscopy locations at 1 and 3 DAI. The EVOS M5000 imaging system (Thermo Fisher) was used for inverted microscopy, merging txRED and bright-field images. Uninoculated plants served as controls for autofluorescence.

4.10. Motility assays

Lab A conducted the motility assays. Pre-cultures were grown in 8 ml of liquid medium shaken at 200 rpm, at 27 °C in the dark, for 5–8 days. The swimming motility was tested by observing colony spreading on R2A soft agar plates (0.3% w/v). Initially, motility was tested for all 17 isolates at pH 7.2. Then, Paraburkholderia, Gottfriedia, or Brevibacillus were tested at pH 4, 5, 6, 7, 8, or 9 (adjusted with 1M HCl or 0.5M NaOH). Each 10 cm Petri dish containing 30 ml of solidified medium was inoculated with 5 µL of culture at the center (n = 3). Plates were incubated at 27 °C in the dark, and the motility ring diameter was measured after 24 and 45 h.

4.11. Data management, statistical analyses, software, and data visualization

The participating laboratories uploaded plant biomass data, root scans, and photos into a shared Google folder with structured directories and spreadsheets. The heat maps were generated with RStudio version 4.0.5 using heatmap.2 in the ggplots package [53]. The NMDS plots were also generated in RStudio using vegan package version 2.5-7 [54]. GraphPad Prism 10 version 10.2.3 generated all other plots and statistical analyses. Biorender.com was used to create graphical overviews. Microsoft Excel version 16.78.3 was used to store and manipulate data frames.

Statistical analysis of cross-laboratory variation included a two-way ANOVA by measuring significance of main effects and their interactions, calculation of source of variation as sum of squares of main effects and total sum of squares, and CV that was measured as a ratio between the standard deviation σ to the mean μ. The statistical analysis was applied to microbe abundance data from 16S rRNA sequencing and raw peak heights for 60 identified rhizosphere metabolites. Additionally, metabolite intensities were also analyzed by pairwise comparisons. The ANOVA and pairwise tests were done in Python v 3.8 using the Pingouin package version 0.5.5. The default ANOVA command was used between lab and treatment. The default pairwise tests command was used between lab and treatment with a p-value adjustment of Benjamini/Hochberg FDR correction. The code can be found at https://doi.org/10.6084/m9.figshare.29700029.