Description

Abstract (English)

Flow cytometry captures subtle changes in immune cell distributions that reflect human disease, but its full potential in clinical decision-making is presently limited by technical variability across instruments, sites, and time. To accelerate research into generalizable predictions of disease, we created a large benchmark dataset of 482 clinical flow cytometry samples from 381 healthy donors, collected over 17 months using six cytometers at four international laboratories. Of special value, a subset of 329 samples from 228 donors were measured in parallel on at least 2 instruments. This resource captures real-world methodological, temporal, and instrument-dependent variability. We propose an alignment strategy, so-called relativisation, which normalizes fluorescence intensities to median CD3 expression per channel as an internal reference, enabling consistent clustering and classification across platforms. Using this approach, we predict donor age, sex, and CMV-IgG serostatus with superior precision. We validate our method with conventional and spectral flow cytometers, and show that relativisation minimises systematic measurement errors. Annotated datasets at each pre-processing step are publicly available in ImmPort. This resource provides a carefully structured dataset and alignment strategy to support reproducible, generalizable and automated analysis of flow cytometry in multi-centre clinical diagnostics.

Methods (English)

Overall methods

Patients and Ethics

Blood samples were collected from healthy donors at the University Hospital Regensburg, Germany, the University Hospital Marqués de Valdecilla, Santander, Spain, and Germans Trias I Pujol Research Institute (IGTP), Badalona, Spain. The study was conducted in multiple phases: R1 and R2 in Regensburg (July 18, 2023-January 23, 2024, and May 15, 2024-July 16, 2024), SAN in Santander (April 29, 2024-May 28, 2024), FORT in Regensburg (January 14, 2025 - January 24, 2025) and BAD in Badalona (March 18, 2025 - March 19, 2025). The study was approved by the Ethics Committees of the University of Regensburg (22-2780-101), Hospital Universitario Marqués de Valdecilla (CS24-116; 2024_6) and IGTP (PI-23-272). The study was conducted in accordance with the principles of the Declaration of Helsinki and all other relevant national and international laws and guidelines. All donors provided written, fully informed consent to sample collection and publication of anonymized results. Complete descriptions of each cohort (Supplementary Note 1) and clinical investigations (Supplementary Note 2) are provided as Supplementary Material.

Flow cytometry measurements

Step-by-step protocols can be accessed at Protocol Exchange. Briefly, blood was collected into EDTA-vacutainers by peripheral venepuncture and then delivered to the responsible lab at ambient temperature. Samples were stored at 4°C for up to 4h before processing. Whole blood samples were stained as previously described using the DURAClone IM T Cell Subsets Tube (Beckman Coulter, B53328) and single-staining controls. DURAClone IM T cell subsets compensation tubes were run every two weeks. Daily quality control (QC) checks were run on all cytometers using Flow-Check Pro Fluorospheres (Beckman Coulter, A63493) to ensure proper function.

Data were collected in Regensburg using a Navios™ cytometer running Navios™ Cytometry List Mode Acquisiton Analysis Software, Version 1.3 (Beckman Coulter) or a CytoFLEX LX™ cytometer running CytExpert, Version 2.4.0.28 (Beckman Coulter) or a BD LSRFortessa X-20' running BD FACSDiva software v9.0. In Santander, data were collected using a DxFLEX cytometer running CytExpert for DxFLEX, Version 2.2.0.7. In Badalona, data were collected using both a Cytek Aurora™ 5L spectral flow cytometer running SpectroFlo® software (Cytek Biosciences), Version 3.1.0, or a BD LSRFortessa 4L flow cytometer running BD FACSDiva™Software (BD Biosciences), Version 6.2. Settings for each cytometer were established by independent experienced operators without exchange of reference samples, calibration materials or example data.

Data were pre-processed by a single experienced, blinded operator who performed: (1) sample-wise manual recompensation; (2) manual gating; and (3) rescaling with a suitable arcsinh cofactor. Data were then exported as FCS files for upload to the ImmPort repository (Accession_ID). These FCS files contain the uncompensated data and three compensations: 1) single-stain compensation for each sample, 2) compensation from DURAClone IM T Cell Subsets compensation tubes, 3) manually recompensated single-stain compensation for each sample. The manual gatings per sample are provided as FlowWorkspace gating sets in h5 format and should be applied to the manually recompensated data. An example gating strategy is provided. In addition, we provide manually recompensated, pre-gated T cell FCS files for all fully stained samples.

Flow cytometry analyses

All computations were performed in R using the Bioconductor packages flowCore and flowWorkspace, alongside our own convenience tools, cytobench , cycompare and otcyto. We gathered FCS files from all instruments, and carried out pre-processing, clustering, classification, and optimal transport analyses.

Pre-processing included applying compensation (or unmixing in the case of spectral cytometry), manual gating to identify CD45+ CD3+ singlet T cells in panel samples and lymphocytes in single-stained controls, arcsinh transformation of fluorescence intensities using manually selected cofactors, random cell subsampling to standardise sample sizes and optionally CytoNorm for normalisation. Alternatively, we applied data relativisation as an alignment strategy before arcsinh transformation.

For reproducibility, R1 and SAN donors were randomly assigned to training, validation, and test sets only once. This donor-level split was preserved for all downstream analyses and serves as the foundation for model development and evaluation. Ideally, future benchmarking efforts using our data should report on those splits for direct comparability of results.

Methods (English)

Methods CORE cohort

Step-by-step protocols can be accessed at Protocol Exchange [Kronenberg et.al. (2021) DOI:10.21203/rs.3.pex-757/v1], except we here generally used 30 minutes of incubation time. Briefly, blood was collected into EDTA-vacutainers by peripheral venepuncture and then delivered to the responsible lab at ambient temperature. Samples were stored at 4°C for up to 4h before processing. Whole blood samples were stained according to manufacturer's instructions (but with 30 min incubation time) using the DURAClone IM T Cell Subsets Tube (Beckman Coulter, B53328, https://www.mybeckman.de/reagents/coulter-flow-cytometry/antibodies-and-kits/duraclone-panels/duraclone-im-t-cell-subsets/B53328), single-staining controls and an unstained control. DURAClone IM T cell subsets compensation tubes were run on all three cytometers every two weeks. 

Data were collected in Regensburg using both a Navios cytometer running Navios Cytometry List Mode Acquisiton & Analysis Software, Version 1.3 (Beckman Coulter) or CytoFlex LX cytometer running CytExpert, Version 2.4.0.28 (Beckman Coulter). In Santander, data were collected using a DxFLEX cytometer running CytExpert for DxFLEX, Version 2.2.0.7. 

QC-control beads were used to monitor instrument performance.

Within the .fcs files, we supply multiple compensations following the Kaluza Analysis Software workflow: 

1. singlestain_auto: For each sample, single-stain controls were used to calculate the compensation matrix.
2. singlestain_manual: For each sample, single-stain controls were used to calculate the compensation matrix. An experienced operator performed blinded analyses for sample-wise recompensation.
3. DCtubes_auto: DURAClone IM T Cell Subsets Tube compensation kits were used every two weeks to calculate the compensation matrix. All samples from that day until the "next" compensation kit were compensated with the same matrix.

From Kaluza, we exported the autofluorescence vector and the spillover matrix. To ensure the same results as in our publication, please follow the Kaluza workflow: 

1. Subtract autofluorescence from the data multiplied with the maximum value of the channel. 
2. Apply compensation as usual.
3. Add the removed autofluorescence back to the data.

Notably, the DxFLEX flow cytometer used in Santander was set up independently from the cytometers in Regensburg by an independent application specialist from Beckman Coulter who did not have any knowledge about the study design or objectives. 

All files are named according to the following scheme, split by underscores ("_"). Logically connected parts are separated by hyphens ("-").

Structure: <cohort>_<donor>_<sample>_<processing>_<appliedGating>_<subsample method>_<subsample size>_<device>_<sample-number>-<antigen>-<fluorochrome>.fcs

Example: R2_d052_s054_raw_ungated_none_Inf_navios_01-CD3-FITC.fcs
R2: Regensburg cohort 2
d052: donor 52
s054: sample 54
raw: raw data
ungated: no gating applied
none: No subsampling applied
Inf: How many cells were subsampled (Inf = Infinity, so all measured cells)
navios: Navios cytometer
01-CD3-FITC: First cytometry-sample of this biosample. Here CD3 was stained with FITC.

Specials: 
- 01-CD3-FITC: Single-stained CD3 with FITC
- 02-CD3-PE: Single-stained CD3 with PE
- 03-CD3-ECD: Single-stained CD3 with ECD
- 04-CD3-PC5.5: Single-stained CD3 with PC5.5
- 05-CD3-PC7: Single-stained CD3 with PC7
- 06-CD3-APC: Single-stained CD3 with APC
- 07-CD3-AF700: Single-stained CD3 with AF700
- 08-CD3-AF750: Single-stained CD3 with AF750
- 09-CD3-PB: Single-stained CD3 with PB
- 10-CD3-KrO: Single-stained CD3 with KrO
- 11-none: No staining
- 12-panel: DURAClone IM T Cell Subsets Tube
- 13-01..11-SingleNone: All single-stain controls and unstained control concatenated in one file

Kronenberg, Katharina, Paloma Riquelme, and James A. Hutchinson. "Standard protocols for immune profiling of peripheral blood leucocyte subsets by flow cytometry using DuraClone IM reagents." (2021). (https://doi.org/10.21203/rs.3.pex-757/v1)

Methods

Methods BAD + FORT cohort

All files are named according to the following scheme, split by underscores ("_"). Logically connected parts are separated by hyphens ("-").

Structure: <cohort>_<donor>_<sample>_<processing>_<appliedGating>_<subsample method>_<subsample size>_<device>_<sample-number>-<antigen>-<fluorochrome>.fcs

Example: BAD_d01_s01_raw_ungated_none_Inf_fortessa_01-CD3-FITC.fcs
BAD: Badalona cohort
d01: donor 01
s01: sample 01
raw: raw data
ungated: no gating applied
none: No subsampling applied
Inf: How many cells were subsampled (Inf = Infinity, so all measured cells)
fortessa: Fortessa (flow cytometer), alternative; Aurora (spectral cytometer
01-CD3-FITC: First cytometry-sample of this biosample. Here CD3 was stained with FITC.

Specials: 
- 01-CD3-FITC: Single-stained CD3 with FITC
- 02-CD3-PE: Single-stained CD3 with PE
- 03-CD3-ECD: Single-stained CD3 with ECD
- 04-CD3-PC5.5: Single-stained CD3 with PC5.5
- 05-CD3-PC7: Single-stained CD3 with PC7
- 06-CD3-APC: Single-stained CD3 with APC
- 07-CD3-AF700: Single-stained CD3 with AF700
- 08-CD3-AF750: Single-stained CD3 with AF750
- 09-CD3-PB: Single-stained CD3 with PB
- 10-CD3-KrO: Single-stained CD3 with KrO
- 11-none: No staining
- 12-panel: DURAClone IM T Cell Subsets Tube

Technical info (English)

NOTE for CORE.02-COMPENSATED.7z:
Unfortunately, as of now we were unable to upload the complete compensated CORE dataset to zenodo as one. Instead, we split it and leave it up to you to merge it together. For this, please download:

CORE.02-COMPENSATED.7z.aa
CORE.02-COMPENSATED.7z.ab
CORE.02-COMPENSATED.7z.ac
CORE.02-COMPENSATED.7z.ad

into one directory, then use 

cat CORE.02-COMPENSATED.7z.a* > CORE.02-COMPENSATED.7z

to create the full 7-zip file. Check the md5sum to ensure that it worked correctly with 

md5sum CORE.02-COMPENSATED.7z   # (93933ee1c29723242d2acac598bfd65a)