HADES research resources

A lightweight companion hub for the HADES project, connecting the method paper, experiment-level datasets, analysis code, supplementary workflow video, the companion Data Descriptor, and reusable data-publication resources.

Resource status — 21 September 2026. The HADES method paper has been submitted. Exp32, Exp35, Exp44 and Exp68 have public Zenodo records. Exp62 has been submitted to the e!DAL Plant Genomics and Phenomics Research Data Repository (PGP) and is awaiting repository review; while review is pending, an anonymous SURFdrive preview/download link provides an exact copy of the content submitted to PGP.

Project

HADES (High-throughput Automated Device for End-to-end Screening) is a fully automated platform for sterile in vitro root phenotyping and plant–microbe interaction studies at the Netherlands Plant Eco-phenotyping Centre (NPEC), Utrecht University. It integrates automated media preparation, imaging-based seed selection and robotic sowing, refrigerated stratification, climate-controlled cultivation, programmable microbial inoculation and multimodal imaging within one robotic workflow.

The system can process up to 2,160 custom-designed 129 × 129 mm square Petri dishes (referred to as plates throughout this resource center) in parallel and combines transmitted-light RootCam imaging, multi-channel fluorescence and visible-to-near-infrared hyperspectral imaging with automated analysis. The resulting longitudinal datasets support quantitative study of root architecture, reporter activity, microbial colonization and metabolite-associated fluorescence under standardized conditions.

Method paper

Title: HADES: high-throughput end-to-end automation of multimodal phenotyping for root-microbe interactions
Status: Submitted; reference / DOI to add when available.

Abstract

High-throughput analysis of root system architecture and plant-microbe interactions is constrained by limited reproducibility, labour-intensive workflows, and low-throughput multimodal imaging under sterile conditions. Here, we present HADES, a fully automated phenotyping platform integrating media preparation, seed selection and sowing, robotic bacterial inoculation, environmental control, and time-resolved imaging of up to 2,160 Petri dishes in parallel. HADES combines bright-field, fluorescence, and hyperspectral imaging with deep-learning-based analysis to quantify root architecture, reporter activity, microbial colonisation, and metabolite-associated spectral signatures. We demonstrate its versatility by resolving coumarin-dependent differences in Arabidopsis thaliana root development and metabolite-associated fluorescence, simultaneously monitoring auxin signalling and colonisation by root-colonising Pseudomonas simiae WCS417 bacteria, and screening a bacterial transposon-mutant collection for altered plant-associated phenotypes. HADES provides a scalable, sterile, and flexible platform for reproducible functional genomics and plant-microbiome research, enabling high-throughput mechanistic dissection of plant-microbe interactions.

Experiment datasets

The experiment records and the method-paper figure sources serve different purposes. The public experiment datasets preserve acquisition-level records, PlantScreen Data Analyzer outputs, reviewed metadata, variable definitions, manifests and provenance. The large customized ROOT, fluorescence and hyperspectral downstream result trees used for the method-paper analyses are not duplicated in these dataset records; their code is maintained separately under Analysis code. The lightweight tables under /source are the direct quantitative inputs used for method-paper figures and are provided for transparent figure provenance.

The companion Data Descriptor focuses on the three Arabidopsis reference records Exp32, Exp62 and Exp68. Exp35 and Exp44 remain separate supporting HADES records because they represent a potato study and a large bacterial-mutant screen, respectively.

HADES Exp32 dataset: Arabidopsis root architecture and coumarin-associated fluorescence dynamics across coumarin biosynthesis genotypes under iron deficiency

Method-paper relationship: Figure 2, longitudinal root-system architecture; Figure 5, coumarin-associated fluorescence; Supplementary Videos S2 and S5.

Dataset: DOI
Release scope: Full Exp32 experiment; Arabidopsis Data Descriptor reference record
Product form: Archive-optimized
Deposited data/ payload: 38.752 GB
Reviewed scope: 20 plates; 100 plant-position rows

Public-record description: Longitudinal Arabidopsis thaliana imaging dataset examining root development and coumarin-associated fluorescence across wild-type Col-0 and the f6′h1, cyp82c4 and s8h coumarin-biosynthesis mutants under iron-deficient conditions. The public record preserves RootCam transmitted-light and fluorescence acquisition records, corresponding PlantScreen Data Analyzer outputs where generated, reviewed metadata, package-scoped variable definitions, manifests, checksums, provenance and validation information.

Method paper figure source

Direct source tables used for the Exp32 method-paper figures are stored under /source:

HADES Exp35 dataset: Potato root colonisation dynamics by Pseudomonas simiae WCS417 and Pseudomonas capeferrum WCS358 across host genotypes

Method-paper relationship: Supplementary Figure S3, fluorescence-based quantification of bacterial colonisation dynamics on potato roots; Supplementary Video S4.

Dataset: DOI
Release scope: Separate potato public record; outside the three-record Data Descriptor scope
Deposited data/ payload: 23.6 GB
Reviewed scope: 42 plates; 42 valid plant-position rows (one plant per plate)

Public-record description: Longitudinal potato root-imaging dataset examining bacterial root-colonisation dynamics across host and microbial genotypes. The experiment contains diploid hybrid true potato seed genotypes HYB007 and RP043 under mock treatment or inoculation with mCherry-labelled Pseudomonas simiae WCS417 or Pseudomonas capeferrum WCS358.

Method paper figure source

Direct source tables used for the Exp35 method-paper figure are stored under /source:

HADES Exp44 dataset: High-throughput longitudinal Arabidopsis screen of 480 GFP-labelled Pseudomonas simiae WCS417 transposon-library isolates

Method-paper relationship: Figure 4, high-throughput longitudinal screening of GFP-labelled WCS417 transposon-library isolates for altered bacterial colonisation and root-development-associated phenotypes; Supplementary Table S4 provides annotations for sequenced candidate isolates.

Dataset: DOI
Release scope: Separate WCS417 transposon-screen public record; outside the three-record Data Descriptor scope
Deposited data/ payload: 28.3 GB
Reviewed scope: 100 plates; 500 plant positions (480 bacterial-isolate treatments and 20 mock controls; five seedlings per plate)

Public-record description: HADES Exp44 is a high-throughput longitudinal Arabidopsis thaliana imaging dataset generated using the HADES automated phenotyping platform to screen a GFP-labelled Pseudomonas simiae WCS417 transposon library for plant-associated phenotypes. Bacterial treatments are assigned at the individual plant-position level, so multiple library isolates can occur within the same physical plate.

Seven-day-old seedlings were inoculated at the root-shoot junction with GFP-labelled WCS417 library isolates using the HADES robotic liquid-handling system. The companion method paper describes application of 10-µL bacterial suspensions at OD600 = 0.1 and daily imaging over seven consecutive days after inoculation. Longitudinal RootCam acquisition provides transmitted-light records of plant and root development together with fluorescence imaging of the GFP-labelled bacterial treatments.

The public release preserves the source HADES acquisition records and curated experimental metadata, including mappings between physical plates, individual plant positions, bacterial treatments and available mutant identifiers. Custom HADES segmentation, fluorescence alignment, derived phenotype tables and figure-generation outputs are downstream products and are not included in the dataset itself.

A subset of 96 screened bacterial isolates was selected for transposon insertion-site sequencing. Available metadata link sequenced isolates to screening-well identifiers where the relationship could be recovered. Usable locus assignments were obtained for approximately half of this subset; absence of a locus assignment should not be interpreted as missing phenotyping data or automatically as confirmation of a wild-type genotype.

Method paper figure source

The direct source table used for the Exp44 method-paper screening figure is stored under /source:

Sequence data

GFP-labelled Pseudomonas simiae WCS417 mariner transposon mutants

EC00167691.zip

Raw Sanger sequencing data for the 96 GFP-labelled P. simiae WCS417 mariner transposon mutants. Individual sequence files are labelled according to the position of each mutant in the 96-well screening plate. The corresponding transposon-disrupted genes and mutant annotations are provided in Supplementary Table S4.

HADES Exp62 dataset: Multimodal Arabidopsis phenotyping with Boxeed seed imaging and VNIR hyperspectral fluorescence across eight genotypes under iron deficiency

Method-paper relationship: Figure 6, VNIR hyperspectral discrimination of coumarin-associated emission profiles in the Col-0, f6′h1, cyp82c4 and s8h subset; Supplementary Figure S2, Boxeed seed-imaging and selection component.

Repository target: e!DAL - Plant Genomics and Phenomics Research Data Repository (PGP)
Status: Submitted 3 September 2026; repository review in progress
Temporary anonymous preview / download: SURFdrive — exact copy of the content submitted to e!DAL-PGP
Dataset-specific e!DAL link / DOI: pending repository review
Licence: CC BY 4.0
Release scope: Full eight-genotype Exp62 batch; Arabidopsis Data Descriptor reference record
Product form: Repository-browsable
Submitted data payload: 579.4 GB
Reviewed scope: 149 plates; 745 plant-position rows

Dataset description: Multimodal Arabidopsis thaliana phenotyping dataset combining Boxeed seed imaging/selection records with VNIR hyperspectral fluorescence acquisition under iron-deficient conditions across eight genotypes. The method paper uses the Col-0, f6′h1, cyp82c4 and s8h subset for its hyperspectral demonstration, while the public release preserves the complete eight-genotype experiment.

Method paper figure source

Direct source tables used for the Exp62 method-paper hyperspectral figure are stored under /source:

HADES Exp68 dataset: Arabidopsis auxin-responsive fluorescence and Pseudomonas simiae WCS417 colonisation dynamics across bacterial and sucrose conditions

Method-paper relationship: Figure 3, dual-channel fluorescence imaging of auxin-responsive DR5v2::mTurquoise2 signalling and WCS417-mCherry colonisation; Supplementary Video S3. The method-paper demonstration uses the no-added-sucrose, non-pvd WCS417-mCherry subset and corresponding controls.

Dataset: DOI
Release scope: Complete 12-condition Exp68 batch; Arabidopsis Data Descriptor reference record
Product form: Archive-optimized
Deposited data/ payload: 78.604 GB
Reviewed scope: 30 plates; 150 plant-position rows

Public-record description: Longitudinal Arabidopsis thaliana imaging dataset examining plant development, auxin-responsive fluorescence and bacterial-associated fluorescence across plant, bacterial and media conditions. The complete batch contains Col-0 and DR5v2::mTurquoise2 plants, mock/WCS417-mCherry/WCS417_pvd-mCherry bacterial conditions, and media with or without 5% sucrose, preserving a broader design than the subset analysed in the method paper.

Method paper figure source

Direct source tables used for the Exp68 method-paper dual-channel fluorescence figure are stored under /source:

Analysis code

The experiment deposits preserve acquisition and vendor-generated records. The customized analysis code used for the method paper is maintained separately so that segmentation, registration and figure-level products can be regenerated without permanently duplicating large image-derived result trees.

ROOT / root architecture analysis pipeline

Repository (fixed frozen branch): github.com/NPEC-NL/pyphenotyper/tree/hades-paper-frozen
Frozen release page: hades-method-submission-2026-08-26
Frozen archive DOI: DOI

Processes monochromatic backlit RootCam images through plate detection/cropping, U-Net-based seedling segmentation, skeletonization and graph-based root reconstruction. Dijkstra's shortest-path algorithm reconstructs the primary root between the root-shoot junction and primary-root tip. Outputs include shoot size, primary-root length, lateral-root length, lateral-root number, nodes and tips, together with structural masks reused by the fluorescence and hyperspectral workflows.

Fluorescence analysis — HADES_FC

Repository: github.com/valerian-meline/HADES_FC
Frozen release page: hades-method-submission-2026-08-26
Frozen archive DOI: DOI

Uses masks from corresponding monochromatic RootCam images to quantify fluorescence without re-segmenting fluorescence frames. Geometric registration aligns modalities; region-specific measurements cover the root system, primary and lateral roots, root tip, nodes, shoot and a dilated peri-root/rhizosphere region. Mean intensity, total signal and pixel counts support reporter, microbial-colonization and coumarin-associated fluorescence analyses.

VNIR hyperspectral analysis — HADES_HSI

Repository: github.com/valerian-meline/HADES_HSI
Frozen release page: hades-method-submission-2026-08-26
Frozen archive DOI: DOI

Processes VNIR fluorescence cubes using dark-reference correction, spectral smoothing and spatial filtering. RootCam masks are registered in two stages using a 670-nm leaf reference and a 410-nm root-enhanced reference. Registered root and peri-root regions are used to derive wavelength-resolved mean, summed and standard-deviation spectra plus pixel-level spectra and spatial coordinates.

Software reproducibility demo

A small real-data software-verification workflow is provided for reviewers who want to test the complete custom analysis chain on a representative HADES acquisition without downloading a full experiment record.

Guide: SOFTWARE_DEMO.md
Release: v1.0
Demo archive: demo.zip

The demo uses one Exp62 Col-0 plate containing five plants. ROOT1/FC1 and VNIR2 records come from the same experimental day, with independent acquisition-round numbers for the different sensors. The distributed archive contains inputs only; downstream ROOT, fluorescence and hyperspectral analysis directories are regenerated locally.

ROOT1 raw TIFF
    -> historical vendor-equivalent 8-bit RGBA PNG
    -> PyPhenotyper
    -> ROOT1 masks and morphology measurements
    -> HADES_FC (FC1 / F483)
    -> registered fluorescence measurements and QC overlays
    -> HADES_HSI (VNIR2)
    -> registered hyperspectral measurements and QC overlays

This is a software reproducibility demo, not a substitute for the full experiment record and not a one-command recreation of every method-paper figure. The demo.zip release asset is prepared from Exp62 specifically for verification; scientific reuse should cite the full Exp62 dataset.

The detailed guide documents the tested Windows CPU workflow, recommended Python environments, deterministic ROOT TIFF-to-PNG conversion, expected outputs, measured runtimes, troubleshooting, and a concise reviewer verification path.

Supplementary video

Supplementary Video S1 — Automated workflow of the HADES platform

YouTube: youtu.be/F-a306otM4A

Demonstration of the fully automated operation of the HADES platform. The video shows the sequential workflow from automated preparation of plant growth plates and imaging-based seed selection with robotic sowing, through refrigerated stratification and climate-controlled cultivation, to high-throughput phenotyping using RootCam and hyperspectral imaging, followed by sterile robotic microbial inoculation. Robotic transfer systems connect all hardware modules, enabling continuous, unattended plant cultivation, phenotyping, and experimental manipulation. Corresponding hardware stations are shown in Fig. 1.

Supplementary Video S2 — Automated root segmentation and longitudinal tracking in HADES

YouTube: https://www.youtube.com/watch?v=rqZrqshpGcw

Time-lapse RootCam imaging of Arabidopsis thaliana wild-type Col-0 seedling grown on iron-deficient Hoagland medium lacking Fe-EDTA and solidified with 0.8% Gelrite. Deep-learning-based image analysis automatically segments the primary and lateral root system and assigns key developmental landmarks, including the top and tip of the primary root, as well as the junctions and tips of individual lateral roots. The extracted root architecture is tracked throughout seedling development, enabling automated, high-throughput quantification of root growth dynamics and architectural traits over time. Corresponding images are shown in Fig. 2.

Supplementary Video S3 — Dual-channel auxin signalling and WCS417 colonisation in Arabidopsis

YouTube: https://www.youtube.com/watch?v=S5Xl6s9WQcM

Representative time-lapse fluorescence imaging of Arabidopsis thaliana Col-0 and DR5v2:mTurquoise2 reporter seedlings grown on Hoagland medium solidified with 0.8% Gelrite and inoculated with mCherry-labelled Pseudomonas simiae WCS417 (10⁵ CFU mL⁻¹). The video simultaneously visualises bacterial colonisation (mCherry channel) and auxin-responsive reporter activity (mTurquoise2 channel) during seedling development. Deep-learning-based root segmentation enables longitudinal quantification of fluorescence along the root system, allowing bacterial colonisation dynamics to be correlated with spatial and temporal changes in auxin-responsive signalling. Corresponding images and quantitative analyses are shown in Fig. 3.

Supplementary Video S4 — Longitudinal bacterial colonisation of potato roots

YouTube: https://www.youtube.com/watch?v=oWgu6D1o7_U

Representative time-lapse RootCam imaging of true potato seed genotype HYB007 grown on Hoagland medium solidified with 0.8% Gelrite following inoculation with mCherry-labelled Pseudomonas capeferrum WCS358r at the root-shoot junction. The video shows simultaneous visualization of root architecture and bacterial fluorescence. In combination with deep-learning-based root segmentation, it enables automated longitudinal quantification of bacterial colonisation along the root system throughout seedling development. Corresponding images and quantitative analyses are presented in Supplemental Fig. S3.

Supplementary Video S5 — Coumarin-associated fluorescence dynamics in Arabidopsis

YouTube: https://www.youtube.com/watch?v=aG26yrAFsBg

Representative time-lapse RootCam imaging of Arabidopsis thaliana Col-0 and the coumarin biosynthesis mutants f6′h1, cyp82c4, and s8h grown on iron-deficient Hoagland medium lacking Fe-EDTA. The video visualises the temporal accumulation and spatial distribution of coumarin-associated autofluorescence during root development. Deep-learning-based root segmentation enables automated longitudinal quantification of fluorescence in the primary root, lateral roots, rhizosphere region, and shoot, revealing genotype-dependent differences in coumarin-associated fluorescence dynamics. Corresponding images and quantitative analyses are presented in Fig. 5.

Data Descriptor

Title: Longitudinal multimodal Arabidopsis root phenotyping data from an automated platform
Status: Pending submission.
Scope: Exp32, Exp62 and Exp68. Exp35 and Exp44 are linked supporting records but are not part of the three-experiment Data Descriptor release.

Abstract

Automated root phenotyping platforms generate large, time-resolved image collections whose reuse depends on experimental context, variable definitions and links between sensor records. This Data Descriptor presents three Arabidopsis experiments generated with HADES (High-throughput Automated Device for End-to-end Screening), covering longitudinal RootCam imaging, fluorescence, VNIR hyperspectral records, vendor-generated analysis products and, for Exp62, Boxeed seed records. HADES acquisition, vendor-side analysis and export are managed through PlantScreen, the vendor software environment used by the platform. Each experiment is released with reviewed metadata, a package-scoped HADES variable-registry snapshot, manifests, checksums and minimal raw-format loaders. The registry provides stable definitions, methods and scales for machine- and pipeline-generated measurements. psi_export_rebuilder, an open-source release tool developed for these records, converts readable PlantScreen exports into traceable research products while preserving mappings for controlled facility-side recovery. Large downstream segmentation, alignment and figure-generation trees are excluded because they largely duplicate acquisition images and can be regenerated from archived analysis code. Together, these resources support reuse of longitudinal multimodal root-phenotyping data outside the vendor environment.

Keywords: HADES; plant phenotyping; root imaging; fluorescence imaging; hyperspectral imaging; longitudinal imaging; FAIR data

psi_export_rebuilder

GitHub repository: github.com/NPEC-NL/psi-export-rebuilder
Frozen version used to create the reference datasets: 0.1.0
Current convenience release: v0.1.1 - adds an easy one-key run script for decompressing/reconstructing archive-optimized products; it does not change the deposited datasets or the release algorithm.
Frozen archive DOI: DOI

psi_export_rebuilder uses readable PlantScreen user exports as the research-facing release source. It inventories records, preserves Measurement/Analysis structure, converts supported tables, creates manifests and hashes, integrates reviewed metadata and package-scoped registry definitions, and retains content mappings to facility-side recovery representations.

The release software supports two repository-facing product forms used here:

Exp32 was also used for the complete facility round-trip test: the rebuilt product was reverse-transformed to the NPEC internal recovery layout, loaded into the NPEC PlantScreen environment, and used to regenerate a fresh PlantScreen Data Analyzer user-export tar with PlantScreen Data Analyzer 3.4.28.0.

System-specific source rules are isolated in adapters so the generic release workflow can be reused for other PlantScreen-derived installations after their local layouts and variables have been audited. A separate stewardship Article is in preparation to evaluate this broader architecture, storage strategy and adaptation beyond HADES; it is not required to interpret the HADES experiment records described in the Data Descriptor.

HADES variable registry

Registry: github.com/NPEC-NL/hades-registry
Frozen version used by the reference release: 1.0.0
Archive DOI: DOI

The registry is a versioned data dictionary for variables produced by PlantScreen Data Analyzer, Boxeed and the custom ROOT, fluorescence and VNIR pipelines. Entries describe stable identity, definitions, source software, type/scale, acquisition method, observation level and trait or ontology mappings. The registry provides the observed-variable layer used for MIAPPE-oriented export, while biological study design and treatment metadata are curated separately. Each experiment package contains the registry subset relevant to its deposited records; the complete registry is available from the repository and Zenodo release above.

Repository services and credits

Partners

NPEC logoNetherlands Plant Eco-phenotyping CentreUtrecht University logoUtrecht UniversityPSI logoPhoton Systems InstrumentsCropXR logoCropXR