Comparison of single-cell RNA-seq methods to enable transcriptome profiling of neutrophils in clinical samples

Summary

Monitoring neutrophil gene expression is a powerful tool for understanding disease mechanisms, developing diagnostics, enhancing therapies, and optimizing clinical trials. Neutrophils are sensitive to the processing, storage, and transportation steps that are involved in clinical sample analysis. This study evaluates the capabilities of technologies from 10× Genomics, PARSE Biosciences, and HIVE (Honeycomb Biotechnologies) to generate single-cell RNA sequencing (scRNA-seq) data from human blood-derived neutrophils. Our comparative analysis shows that all methods produced high-quality data, importantly capturing the transcriptomes of neutrophils. Here, we establish a reliable scRNA-seq workflow for neutrophils in clinical trials: we offer guidelines on sample collection to preserve RNA quality and demonstrate how each method performs in capturing sensitive cell populations in clinical practice.

Graphical abstract

Highlights

• •Comparison of single-cell RNA-seq methods for clinical biomarker studies
• •Flex, Evercode, and HIVE capture neutrophil transcriptomes
• •Flex and Evercode show strong concordance with flow cytometry cell populations
• •Flex has a simplified sample collection protocol suitable for clinical site collection

Motivation

Neutrophils are crucial in the clinical setting, as they are key immune responders and important markers in various diseases. Transcriptional profiling of neutrophils by single-cell RNA sequencing (scRNA-seq) has remained challenging due to the low levels of mRNA and high levels of RNases. We compare current scRNA-seq methods to evaluate the efficiency of the preservation of neutrophils in samples and resulting data quality with the goal of identifying an scRNA-seq method with a simple sample collection that is suitable for implementation on clinical samples and ensures the best data quality.

Hatje et al. evaluate current fixed single-cell technologies to determine their suitability for measuring the neutrophil transcriptome and their potential for implementation in clinical trials.

Introduction

Neutrophils are innate immune effector cells that comprise approximately 60% of leukocytes in circulation. They mediate the body’s first response to invading microorganisms through degranulation, phagocytosis, and the production of neutrophil extracellular traps (NETs).¹ Neutrophil dysregulation, particularly NET formation, is strongly implicated in human diseases ranging from sepsis and autoimmunity to cancer metastasis and inflammatory diseases.¹ In the clinic, neutrophils and neutrophil expression signatures are increasingly being used as biomarkers. Notably, the neutrophil-to-lymphocyte ratio, when combined with tumor mutation burden, is being used to forecast the effectiveness of immune checkpoint inhibitors in cancer treatment.^2,3 Additionally, biomarkers derived from neutrophils are being investigated for their potential to predict major adverse cardiac events.⁴

Single-cell RNA sequencing (scRNA-seq) has helped to improve our understanding of the different transcriptional states of neutrophils and suggests a future role for the neutrophil gene expression signatures as clinical biomarkers. Four distinct and stable transcriptomic states observed during the maturation and activation of neutrophils have been described, Nh0, Nh1, Nh2, and Nh3, suggesting that a deeper understanding of these transcriptomes could provide disease biomarkers.⁵ Expanding on this, Montaldo et al. have described the transcriptome of neutrophils in a steady state and upon stress using both a bulk RNA-seq approach and scRNA-seq. In that study, they describe how the different activation statuses of neutrophils are predictive biomarkers for organ transplant success.⁶ A recent study has highlighted how there is a high level of transcriptional heterogeneity in neutrophils isolated from different types of cancer. Originally, high levels of invading neutrophils were thought to be a poor prognostic indicator. More recent findings suggest that neutrophils with an antigen-presenting transcriptional program are associated with a positive outcome in most cancers.⁷ Understanding of neutrophil biology will help develop biomarkers for identifying patients who may experience cytokine release syndrome in response to T cell-engaging therapies—for example, T cell bispecifics.⁸ Taken together, these papers suggest a future requirement for profiling neutrophils in clinical samples.

Neutrophils contain lower RNA levels than other cell types in the blood.⁵ Classical methods using gel emulsion beads (10× Genomics) have proved challenging to capture neutrophils and granulocytes.² However, several groups have demonstrated that generating neutrophil scRNA-seq data is feasible. 10× Genomics recommends the addition of protease and RNase inhibitors to the standard protocol to increase the capture of neutrophils. Additionally, GEM X technology from 10× Genomics (not evaluated in this study) promises better recovery of sensitive cell types. Wigerblad et al. attributed successful collection of neutrophils to changes in the bioinformatics pipeline and not the addition of protease and RNase inhibitors.⁵ We have previously shown that the microwell-based scRNA-seq BD Rhapsody effectively captured the transcriptome from neutrophils. The percentage of neutrophils retrieved from samples was comparable to flow cytometry using CD16, CD11b, and CD62L as markers.⁸ A direct comparison between the BD Rhapsody and Chromium Single-Cell 3′ Gene Expression suggested that RNA capture is significantly more sensitive in the microwell-based method, leading to more sensitive detection of cells with low RNA content.⁹

scRNA-seq is a powerful tool in drug discovery. However, its potential for use with clinical samples is limited by the requirement to use fresh cells. For peripheral blood mononuclear cells (PBMCs), protocols have been developed whereby cell separations can be performed at the clinical sites and then the cells cryopreserved and banked for later analysis at a central testing facility.¹⁰ For more sensitive cell types such as neutrophils, this is not possible, as a high proportion of these cells die, and the remaining cells are morphologically and functionally altered in the freeze-thaw process.^11,12,13 For these cell types, the analysis must be performed at the clinical site, which reduces the number of clinical sites that are able to collect samples for scRNA-seq analysis. Taken together, the biological importance and sensitive nature of neutrophils and the complexity of global clinical trial settings call for an easy-to-use stabilization protocol for subsequent scRNA-seq.

We selected three technologies to compare: Evercode WT Mini v.2 from Parse Biosciences (Evercode), Chromium Single-Cell 3′ Gene Expression Flex from 10× Genomics (Flex), and the Honeycomb Biotechnologies HIVE scRNA-seq v.1 device (HIVE). The selection criteria were based on the ability to stabilize cells rapidly prior to library prep, the requirement to process large numbers of cells, and the availability of a commercial product that can be distributed easily to clinical sites. Evercode works on a principle of combinatorial barcoding, where fixed cells are given a sample barcode with the reverse transcription step; samples are then pooled and split before a further three successive barcoding steps, including the addition of a unique molecular barcode.¹⁴ This approach allows for up to 96 multiplexed samples and has been reported to detect more genes expressed at low levels than Chromium Single-Cell 3′ Gene Expression.¹⁵ HIVE works on the principle that cells are distributed into nano-wells and stabilized. The samples can be stored at −80°C prior to the library preparation steps. HIVE has successfully been used to isolate neutrophils from red blood cell (RBC)-depleted samples.¹⁶ In Flex, fixed and permeabilized cells are incubated with a set of 18,532 probes covering the entire transcriptome prior to library preparation steps. The use of probe hybridization allows for the capture of smaller fragments of RNA, which are found in formalin-fixed, paraffin-embedded (FFPE) tissue. This method was successfully used on FFPE tissues and xenograft models.^17,18

Neutrophils are reported to have a short half-life ex vivo, and the methods of isolation can lead to activation or apoptosis. Therefore, we used the findings of previous studies on neutrophil isolation to define the conditions for this study. Previous reports have demonstrated that neutrophils suitable for functional characterization can be isolated from blood stored at room temperature or at 4°C for 24 h or for up to 72 h when stored at 37°C.^19,20,21 Incubators for sample storage are not always available at clinical sites; therefore, we opted to look at the impact of storage at 4°C for 24 h. Currently, there is little information exploring the effect of time from blood draw to analysis or fixation on neutrophil transcriptome stability. This work aims to evaluate single-cell technologies to determine their suitability for (1) measuring the neutrophil transcriptome and (2) their potential for implementation in clinical trials, which require minimal sample processing and sample stabilization.

Results

Study design

To compare the different technologies, blood was drawn from healthy donors and then divided into different aliquots, which were tested using Flex, Evercode, and Chromium Single-Cell 3′ Gene Expression v.3.1 (Figure 1A; Table S1). An aliquot for each donor was run on the flow cytometer to characterize cells into the major cell types to compare with the results from the scRNA-seq clustering. We evaluated the HIVE in a separate experiment using the same format (Figure 1B). To compare directly across the technologies, we limited our analysis to the 18,532 genes captured in the Flex probe set. We used our established BESCA pipeline to analyze our single-cell experiments.²² The knee plots (Figures S1A and S1B) reveal a clear separation between cells and empty droplets for PBMC isolation, aiding in cutoff determination. However, RBC-depleted samples lack this distinct separation due to low gene expression in granulocytes. To ensure the inclusion of neutrophils, we applied a minimum threshold of 50 genes and 50 unique molecular identifiers (UMIs) across all samples, in line with Wigerblad et al., and distinguished empty droplets from cells in the clustering and cell-type annotation step.⁵

Figure 1.

Diagram showing the experimental design to test the four different technologies

(A) Evercode, Flex, Chromium Single-Cell 3′ Gene Expression v.3.1, and flow cytometry were tested on the same three blood samples.

(B) HIVE was tested on a different set of blood samples from three different donors at a different date. The samples were also profiled using flow cytometry.

Comparing the quality of scRNA-seq from the different methods

We compared the quality of the data using the following parameters: UMI counts, the number of genes detected, and the percentage of mitochondrial genes (Figure 2). These parameters are used to discriminate low-quality cells where the cells are stressed or cell leakage occurs during processing.²³ Across all the scRNA-seq technologies, the mitochondrial gene expression levels were low, between 0% and 8%, except for Chromium Single-Cell 3′ Gene Expression v.3.1, which showed very high levels up to 25%. Evercode shows the lowest levels of mitochondrial gene expression, followed by Flex. Chromium Single-Cell 3′ Gene Expression and HIVE, which both used non-fixed cells as input, had higher levels of mitochondrial genes detected.

Figure 2.

Quality-control comparison across technologies

(A–C) Violin plots showing the (A) mitochondrial gene expression levels, (B) UMI counts, and (C) gene counts for Chromium Single-Cell 3′ Gene Expression v.3.1, Flex, Evercode, and HIVE for RBC-depleted samples across the different cell types.

(D) Expression of data-derived stable genes from low-expression to high-expression RBC-depleted samples, separated by technology or cell type.

For Chromium Single-Cell 3′ Gene Expression v.3.1, Flex, and HIVE, we observed a bimodal distribution in the violin plots for the RBC-depleted samples. We propose that these are two populations of cells with different overall gene expression levels: PBMCs with high gene expression and granulocytes with low gene expression.⁵ In comparison, for PBMC isolations, we observe only the population with high gene expression levels (Figures 2B, 2C, and S1C–S1E). For Evercode, the bimodal gene expression was not observed in the RBC-depleted samples (Figures 2B and 2C).

Next, we examined the dynamic range of Flex, Evercode, and HIVE alongside Chromium Single-Cell 3′ Gene Expression v.3.1. To do this, we selected stable genes at different expression levels across all technologies by identifying the gene with the lowest variance within defined bins from low expression to high expression. We performed this separately for RBC-depleted samples (Figures 2D and S2) and PBMCs (Figures S3A–S3C). Evercode samples had lower expression of the highly expressed stable genes. From these data, we conclude that Evercode samples show a different dynamic range from the other technologies, which explains the lack of bimodal distribution in the Evercode RBC-depleted samples.

scRNA-seq clustering

For clustering, we combined the data from all four technologies and observed that the cells clustered based on their cell type and the technology (Figure 3A). Within those clusters, we observed that the cells separated based on the cell separation method used (PBMC versus RBC depletion) (Figure 3A). Neutrophil clusters in all technologies were associated with lower gene counts and UMI counts, which is in line with the low levels of RNA and gene expression in this cell population (Figure 3B). The Chromium Single-Cell 3′ Gene Expression v.3.1 and HIVE clusters showed higher percentages of mitochondrial gene expression (Figures 2A and 3B). Looking at the cell-type clustering for each individual technology (Figures 3A and S4A), we see that the major cell types can be identified clearly in the four different technologies. In all technologies, we also observed artifact clusters, which are composed of empty droplets due to the lower cutoffs used, doublets, or cell types that cannot be assigned to any group (Figure S4A).

Figure 3.

Cell-type clustering and assignment comparison across technologies

(A) Uniform manifold approximation and projection (UMAP) cell separations, with color coding denoting technology, cell separation (tissue), and cell types.

(B) The mitochondrial gene expression and number of genes detected for the different cell types from UMAPs.

(C) Boxplots showing the percentages of B cells, T cells, neutrophils, monocytes, and natural killer cells determined by flow cytometry, Flex, and Evercode or flow cytometry and HIVE. Boxes denote interquartile range and bars show the range of the data.

(D) UMAP of the extracted neutrophils colored by unsupervised clustering for Flex, Evercode, and HIVE, plus a dot plot showing the expression of the top 5 marker genes per cluster for those clusters.

Percentage of cell populations determined by scRNA-seq compared to flow cytometry

To determine how well the technologies captured neutrophils, we compared the percentages of neutrophils from each technology with flow cytometry on the same sample. Flex, Evercode, and HIVE all successfully isolated neutrophils from the RBC-depleted samples. Chromium Single-Cell 3′ Gene Expression v.3.1 also captured neutrophils in RBC-depleted samples; however, the percentage captured was lower compared to the other technologies (Figure 3C). The percentages of neutrophil populations using Flex were the closest to those determined by flow cytometry (Figures 3C and S4C; Tables S2 and S3). Flex and Evercode also compared favorably with flow cytometry results for the isolation of T cells, B cells, monocytes, and natural killer (NK) cells. Indeed, the performance was comparable on the PBMC isolations with the Chromium Single-Cell 3′ Gene Expression v.3.1 methods and flow cytometry (Table S4).

Identification of neutrophil populations

We extracted the neutrophils and re-clustered them within each technology. Then, we used the Wilcoxon rank test to identify the marker genes of each cluster (Figure 3D). We broadly defined clusters of immature neutrophils with the key markers S100A8/A9 and MNDA in Flex (cluster 1) and HIVE (clusters 1 and 2). In the Evercode data, the immature neutrophils defined by S100A8/A9 and MNDA are not identifiable (Figure 3D). In all three technologies, we observe interferon-stimulated neutrophil clusters with the key markers IFIT genes and ISG15 in Flex (clusters 4 and 5), Evercode (clusters 6 and 9), and HIVE (clusters 7 and 9) (Figure 3D). We then clustered our data based on these signatures as defined in Wigerblad et al. and Xie et al. (Figures S5A and S5B).^5,15 We did not observe the clearly defined subtypes described in these publications in any of the technologies tested. Unsupervised clustering (Figure 3D) demonstrated that HIVE and Flex detected a population of immature neutrophils that broadly aligned with Nh0 cells/hG5a neutrophils, and a population of mature neutrophils that aligned with Nh2/hG5b neutrophils could be detected in all the technologies.

Time course for optimum sampling of neutrophils

Neutrophils are a particularly sensitive cell type with a reported short half-life in vivo and in vitro.^24,25 To determine the maximum time that samples could be stored prior to processing, we tested cells at different time points after blood draw (immediate processing and 2, 4, 6, 8, and 24 h after blood draw), prior to fixing and measuring the transcriptome with Flex. The selection of Flex for these experiments was based on the performance and ease of sample processing. After the cell isolation step, we performed a cell count prior to stabilization (data not shown). We observed little overall cell death or decrease in cell count over the 24 h after the blood draw. In concordance with this, the general quality of the scRNA-seq data was unchanged across the time course. There was a little increase in the expression of mitochondrial genes, with all samples having expression levels of <1% for mitochondrial genes, number of counts, or UMI counts across the time course (Figures S6A–S6C). There were also no differences in the percentage of cell types over time since blood draw, as determined by the scRNA-seq (Figures S6D–S6F), indicating that there is no apoptosis of specific cell types taking place over the 24 h. We compared the transcriptional profile of neutrophils at different time intervals after the blood draw, and we observed that the number of genes differentially regulated compared to the 0 h time point started to significantly increase 4 h post blood draw, with the number of genes up- and down-regulated increasing at each time point (Figures 4A and 4B). The most significantly changed pathway was associated with cell stress, defined as stress MP5 by Gavish et al.²⁶ In our results, we observed that stress signatures were up-regulated at all time points 2 h after blood draw, showing the importance of prompt sample processing (Figure 4C). These data are in concordance with the previous report,²⁷ which found markers of neutrophil activation, apoptosis, and degranulation 4 h post blood draw. Therefore, despite live, functionally active neutrophils being present in blood samples 24 h post blood draw, the transcriptome of neutrophils is significantly changed 4 h post blood draw. Our results indicate that immediate fixation of neutrophils is required if the transcriptome is being analyzed. Indeed, we also tested B lymphocytes, myeloid leukocytes, NK cells, and T cells (data not shown). They all show an activation of the stress response, and this is not specific to neutrophils.

Figure 4.

Transcriptional profile of neutrophils after blood draw

(A) Graph showing the number of genes differentially regulated (up or down) in neutrophil pseudobulks at 2, 4, 6, 8, or 24 h after blood draw compared to the sample processed immediately after blood draw. The data show results for the 3 donors at each time point.

(B) Volcano plots for the neutrophil pseudobulk of the genes up- or down-regulated compared to 0 h for each time point tested.

(C) Pathway enrichment for cell stress for neutrophil pseudobulks over the time course. False discovery rates (FDR) are denoted on heat map using ∗, ∗∗, and ∗∗∗.

Discussion

The recent advances in scRNA-seq that allow the rapid stabilization and storage of cells prior to library prep will enable the broader implementation of scRNA-seq analysis in clinical trials. Flex, Evercode, and HIVE protocols support a model where cells are stabilized at the clinical site, enabling storage and transport to the analytical labs. Practically, the HIVE offers a straightforward protocol for use at a clinical site, with the cells simply pipetted into the device after the cell separation step. For Flex, samples need to be centrifuged after the cell separation and then resuspended in fixative. Since these experiments were completed, 10× Genomics has modified its protocols to allow whole blood to be stabilized with paraformaldehyde and then stored and transported at −80°C. This modification allows cell separation and analysis to be performed at the analytical site, presenting a simple procedure for cell stabilization at the clinical site. In contrast, the Evercode protocol for fixation of cells requires several consecutive cell straining and centrifugation steps, making it the most time-consuming and practically the hardest protocol to perform at a clinical site.

Our experiments indicate that all the technologies successfully captured neutrophils from RBC-depleted samples. The three technologies (Flex, Evercode, and HIVE) produced high-quality data, with low levels of mitochondrial gene expression. Flex and Evercode, which use fixed cells, exhibited lower levels of mitochondrial gene expression than Chromium Single-Cell 3′ Gene Expression v.3.1 and HIVE, which use live or frozen cells as input. This difference may be due to the release of cytoplasmic RNA upon the fixation or permeabilization of the cell.²⁸ Interestingly, for Evercode, we observed a different dynamic range from the other technologies. Specifically, we observed that the fully combinatorial barcoding approach¹⁴ led to a lower representation of highly expressed genes. To further understand these observations, we compared the proportion of the different white blood cell (WBC) components captured by each method to those obtained through flow cytometry. Our analysis suggested that Flex and Evercode captured the different WBC components in the samples at a similar percentage to flow cytometry.

HIVE and Chromium Single-Cell 3′ Gene Expression v.3.1, while capturing neutrophil profiles, did not align as closely with the flow cytometry results. For cell-type assignment, Flex had the closest alignment with flow cytometry, followed closely by Evercode. Both HIVE and Chromium Single-Cell 3′ Gene Expression v.3.1 showed greater deviation from the cell-type proportions estimated by flow cytometry. These results agree with a previously published study on PBMCs, where Flex was closely aligned with cytometry by time of flight (CyTOF) for the same sample, whereas greater differences were observed with Evercode and HIVE.²⁸

Unsupervised clustering of neutrophils across all the tested methods defined two distinct groups of neutrophils, which agrees with subtypes that have been previously published.^5,15 The immature neutrophil subtype that aligns with immature neutrophils, Nh0 cells/hG5a, was observed in Flex and HIVE. The second group, consisting of mature interferon-stimulated neutrophils (Nh2/hG5b), was observed in all technologies tested. However, we did not observe all the previously described neutrophil subtypes. This may be due to the neutrophil numbers that were collected in this study, as the less-abundant populations were absent from our analysis. The populations described by Wigerblad et al. were described from >70,000 captured neutrophils.⁵ For our experiments, the neutrophil populations included 18,000 cells for Flex and 9,000–10,000 cells for Evercode and HIVE.

This study compared three methods to determine the viability of implementing scRNA-seq at clinical sites for profiling neutrophils. All three methods produced high-quality data and were able to capture neutrophils from peripheral blood samples. We determined that Flex exhibited the best performance, with the proportions of neutrophils captured in blood samples comparable to those observed by flow cytometry. A closer analysis of the neutrophil subtypes using Flex revealed two distinct populations of neutrophils that have previously been described in the literature. The Flex workflow proved to be amenable to sample collection at the clinical site, making it the best choice for implementation in clinical trials. We recommend limiting the time between blood draw and fixation to 2 h, as after this time, we observe an increase in differential gene expression regulation associated with stress. This study presents a route for scRNA-seq implementation in clinical trials, offering a powerful tool for biomarker development and a deeper understanding of neutrophil biology.

Limitations of the study

Our study’s primary objective was to identify an scRNA-seq method that effectively captures the entire neutrophil population and is robust enough for use in clinical trial settings, with a strong emphasis on the ease of sample collection at clinical sites. This led us to select the Flex method, which utilizes hybridized probes for gene expression measurement. While the Flex probe panel provides focused and reliable gene expression profiles crucial for our clinical biomarker goals, we acknowledge that this approach inherently means we do not capture information on splice variants. Consequently, the broader, unrestricted gene panel advantages offered by Evercode and HIVE were not extensively explored in this specific analysis, as our immediate focus was on direct clinical biomarker utility. Furthermore, we recognize that our evaluation did not include a comparison to the GEM-X technology from 10× Genomics. This promising technology, which offers improved capture of sensitive cell types, was unfortunately not available for assessment during the time frame of our study.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Emma Bell (emma.bell@roche.com).

Materials availability

All materials used in this study were commercially available.

Data and code availability

• •The data for technology comparison and time course are available at https://doi.org/10.5281/zenodo.15373124 and https://doi.org/10.5281/zenodo.13750776, respectively. These are also mentioned in the key resources table.
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We would like to acknowledge the Flow 360 Labs team at Roche for their support with flow cytometry data acquisition.

Author contributions

Experimental design, K.H., K.S., S.D., F.K., N.G., M.M., T.B., and E.B.; data analysis and interpretation, K.H., K.S., L.J., D.M., P.K., J.D.Z., and E.B.; manuscript preparation, K.H., K.S., S.D., F.K., L.J., D.M., P.K., M.M., T.B., A.G., M.S., and E.B.; experimental execution, S.D., F.K., N.G., L.J., and E.B.

Declaration of interests

The authors declare no competing interests.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work, the authors did not use any AI or AI-assisted technology.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
CD3, BUV805 intracellular	BD Bioscience	AB_2870183
CD4 BUV737 intracellular	BD Bioscience	AB_2870079
CD14 BUV563 surface	BD Bioscience	AB_2870860
CD8 BUV496 surface	BD Bioscience	AB_2870759
CD45 BUV395 surface	BD Bioscience	AB_2869519
HLA-DR BV785 surface	Biolegend	AB_2563461
CD56 BV750 surface	BD Bioscience	AB_2871824
CD11c BV711 surface	BD Bioscience	AB_2738019
CD19 BV650 surface	Biolegend	AB_2564255
CD25 BV605 surface	BD Bioscience	AB_2740127
CD123 BV421 surface	Biolegend	AB_10962571
CD45RA PerCP-Cy5.5 surface	Biolegend	AB_893357
CD15 Pe-Cy7 surface	BD Bioscience	AB_10563901
CD20 Pe-Cy5 surface	Biolegend	AB_314256
CD193 PE-CF594	BD Bioscience	AB_2737660
FOXP3 PE intracellular	Biolegend	AB_492986
CCR7 APC-Cy7 surface	Biolegend	AB_10916390
CD16 APC surface	Biolegend	AB_314212
human IgG	Jackson ImmunoResearch, West Grove, Pennsylvania	AB_2337043
Biological samples
Six healthy volunteer donor blood samples	Roche medical center	N/A
Chemicals, peptides, and recombinant proteins
Phosphate buffered saline (PBS)	ThermoFisher, Massachusetts	Catalog # 10010023
2% FBS	VWR International, Pennsylvania	89510–194
True Stain Monocyte Blocker	Biolegend	426101
Fixable Green Dead Cell Stain	ThermoFischer Scientific, Waltham, Massachusetts	L23101
Ficoll	GE healthcare	N/A
EasySep™ RBC Depletion Reagent	Stemcell technologies	18170
Critical commercial assays
Next GEM Single Cell 3′ Reagent Kits v3.1	10x Genomics, Pleasanton California	PN-1000121
Chromium Next GEM Chip G Single Cell Kit, 16 rxns	10x Genomics, Pleasanton California	PN-1000127
Single Index Kit T Set A, 96 rxns	10x Genomics, Pleasanton California	PN-1000213
Chromium Next GEM Single Cell Fixed RNA Sample Preparation Kit	10x Genomics, Pleasanton California	PN-1000414
Chromium Fixed RNA Kit, Human Transcriptome,	10x Genomics, Pleasanton California	PN-1000476
Chromium Next GEM Chip Q Single Cell Kit	10x Genomics, Pleasanton California	PN-1000422
Dual Index Kit TS Set A 96 rxns	10x Genomics, Pleasanton California	PN-1000251
HIVE collectors	Honeycomb	N/A
HIVE scRNA-seq v1 Sample Capture Kit	Honeycomb	N/A
Parse Evercode cell fixation v2	Parse biosciences	N/A
Evercode™ WT Mini v2 kit	Parse biosciences	N/A
S1 Reagent Kit v1.5 (100 cycles)	Illumina	20028319
Deposited data
FACS, 10 × 3′, Parse, 10X Flex, and Hive data and analysis	This paper	https://doi.org/10.5281/zenodo.15373124
10X Flex time course (after blood draw) data and analysis	This paper	https://doi.org/10.5281/zenodo.13750776
Software and algorithms
Scanpy	Wolf et al.29	https://scanpy.readthedocs.io/en/stable/generated/scanpy.tl.score_genes.html
BESCA	Madler et al.22	https://github.com/bedapub/besca/blob/master/besca/datasets/genesets/CellNames_scseqCMs6_sigs.gmt
Uniform Manifold Approximation and Projection (UMAP) algorithm)	Becht et al.30	https://github.com/lmcinnes/umap
Leiden algorithm	Traag et al.31	https://github.com/vtraag/leidenalg
Msigdb	Gene set enrichment analysis	https://www.gsea-msigdb.org/gsea/msigdb/human/collections.jsp
Limma	Ritchie et al.32	http://bioinf.wehi.edu.au/limma/
Reactome pathways	Gavish et al.26	https://reactome.org
cellranger 6.0.2 mkfastq	10X Genomics	https://www.10xgenomics.com/support/software/cell-ranger/latest/analysis/inputs/cr-mkfastq
bcl-convert 3.8.4	Illumina	https://emea.support.illumina.com/sequencing/sequencing_software/bcl-convert/downloads.html
split-pipe v1.0.5p: available to PARSE customers on request	Parse	https://support.parsebiosciences.com/hc/en-us/articles/20403758539924-How-Do-I-Access-the-Parse-Biosciences-Pipeline
cellranger	10X Genomics	https://www.10xgenomics.com/support/software/cell-ranger/latest/release-notes/cr-release-notes
Other
10mL EDTA KCL vacutainer blood collection tubes	BD	366643
SepMate-50 tubes	Stemcell technologies	85450
50 mL Falcon tube	Fischer Scientific	10788561
EasySep Easy Eights magnet	Stemcell technologies	18103
50% Brilliant Stain Buffer	BD Biosciences	659611

Experimental model and study participant details

Whole blood was drawn from 6 healthy human donors. We do not have any information on the sex, gender, ancestry, age, race or ethnicity. These factors are not anticipated to affect the findings of the study which is to compare methods of scRNA-seq as we are not looking at biological variation.

Method details

Samples

Whole blood from healthy donors was drawn into 10mL EDTA KCL or sodium Heparin tubes. For the technology comparison experiments, the blood was divided into 2 mL aliquots, then processed immediately for PBMC isolation or red blood cell removal. In order to determine the time limit for processing and fixing samples after the blood draw we set up a time course. 10 mL of blood was drawn from three donors, a 2 mL aliquot was processed immediately, then subsequent aliquots were processed (RBC-depletion) 2, 6, 8 and 24 h after the original blood draw. In the interim, the sample was stored at 4°C.

PBMC isolation

PBMC were isolated from whole blood using Ficoll (GE healthcare, Illinois) separation according to manufacturer’s instructions. Briefly 15 mL Ficoll was added to SepMate tubes (Stemcell technologies, Vancouver, Canada). Blood was diluted 1:1 with phosphate buffered saline (PBS) (ThermoFischer, Massachusetts) + 2% FBS (VWR International, Pennsylvania). The diluted blood was then added to the SepMate tube. Tubes were then sealed and centrifuged at 1200xg for 10 min at room temperature. The top layer of cells, containing enriched mononuclear cells was decanted into a new 50 mL Falcon tube. The isolated cells were then washed three times with PBS+2%FBS.

RBC-depletion

RBCs were removed from blood samples using the EasySep RBC Depletion Reagent (Stemcell technologies) and the associated EasySep Easy Eights magnet, according to the manufacturer’s instructions. Briefly, 5 mLs of whole blood was diluted with PBS-2%FBS solution. The EDTA and RBC depletion solution were added to the samples prior to gently pipette mixing. The samples were transferred to the magnet and left for 5 min. The supernatant was transferred to a new 14 mL tube and the process repeated at least 3 times, or until no RBCs were visibly remaining.

Cell counting and viability

After the cell isolations and before fixation or library preparation, the cell viability and number were measured using a Cellaca Cell counter (Nexcellom Bioscience, Massachusetts). The viability of all samples was high and ranged between 100 and 96% viability. The number of cells required as input for each different technology is summarized in Table S1. For the library preps for fixed cells, the cell number and viability were counted before fixation and then the fixed cells were counted prior to library prep. For PBMC samples between 250 000–400 000 cells were fixed per sample, and for the RBC-depleted samples 500 000–850 000 cells per sample were fixed.

Chromium single cell 3′ gene expression v3.1

An input of 8000 live cells per sample were used for the chromium single cell 3’ gene expression v3.1 libraries. The libraries were prepared using the Next GEM Single Cell 3ʹ Reagent Kits v3.1. This involves GEM generation on the Chromium instrument, where gel beads were mixed with the live cells and partitioned using oil droplets, resulting in droplets containing a single cell and a gel bead containing barcodes and primers. After barcoding, samples were transferred to a PCR machine for the reverse transcriptase step. Full length, barcoded cDNA from poly adenylated mRNA is then purified using magnetic beads, before PCR amplification. After fragmentation and size selection, P5 and P7 illumina adapters alongside sample barcodes were added to create the final libraries. For sequencing, we targeted 50,000 reads per cell. Therefore, we sequenced the samples on a single NovaSeq S1 flow cell.

Chromium single cell gene expression flex

Live cells were fixed using the Chromium Next GEM Single Cell Fixed RNA Sample Preparation Kit (10X Genomics) according to the manufacturer’s instructions. For PBMC samples between 250 000–400 000 cells were fixed per sample, and for the RBC-depleted samples 500 000–850 000 cells per sample were fixed. Briefly, cells were centrifuged and supernatant removed. Cells were resuspended in the kit supplied fixation buffer, then incubated for 16 h at 4°C. The cells were centrifuged and supernatant removed before resuspension in the Quenching buffer. The cells were washed, counted and the number of cells adjusted to that of the lowest sample. The Chromium Next GEM Single Cell Fixed RNA Sample Preparation Kit and Chromium Fixed RNA Kit, Human Transcriptome, Chromium Next GEM Chip Q Single Cell Kit were used according to manufacturer’s instructions.

In summary, samples were incubated with the human transcriptome probes to allow hybridization for 16 h at 42°C. The samples were then washed and transferred to the Chromium X instrument for GEM generation, during this stage the cell is partitioned into an oil droplet containing a 10x barcoded gel bead, so that cell specific barcodes and UMIs are added to the hybridized probes. The probes are then extended and then PCR amplified prior to the addition of P5 and P7 illumina adapters and sample index barcodes. The samples were sequenced on a Novaseq 6000 to a depth of at least 15 000 reads per cell using an SP - 100cycles v1.5 reagent kit.

Honeycomb biotechnologies HIVE scRNA-seq v1 device

Eight HIVE collectors were loaded with freshly isolated cells. HIVE collectors contain more than 65,000 60μm-wide picowells that are pre-loaded with barcoded 3′ transcript capture beads. Each collector was loaded by centrifugation with a total of 15,000 cells according to HIVE scRNA-seq v1 Sample Capture Kit User Protocol (Revision A). Once loaded, 3 HIVE devices were incubated for 30 min at room temperature before direct processing, and the remaining HIVE devices were frozen at −20°C for later processing after 1 week or 3 weeks of storage. Upon thawing the HIVE devices were equilibrated for 60 min at room temperature before processing. All HIVE devices, whether processed directly or after storage at −20°C were processed the same way and according to the manufacturer’s instructions.

Briefly, the cells were lysed and hybridized to beads in the HIVE collectors. Beads were recovered by centrifugation into a bead collector and transferred to a filter plate set on a vacuum manifold allowing beads washing and buffer exchanges by aspiration. After the first strand synthesis (45 min at 37°C), 1X NaOH was added and beads washed 3 times before the 2nd strand synthesis (37°C for 30 min). Samples were washed and transferred to a deep well plate for whole transcriptome PCR amplification. After a double-sided SPRI clean-up, samples were indexed by PCR and a final SPRI was performed. Libraries size was checked the bioanalyzer on high sensitivity DNA chips (Agilent), concentration determined on Qubit 3.0 using the dsDNA High sensitivity kit and sequenced on the Novaseq 6000 with an S1 - 100cycles v1.5 reagent kit, using HIVE custom primers and 25-8-8-50 sequencing cycles.

Evercode WT mini v2 from parse biosciences

Live cells were fixed using the Evercode Cell Fixation v2 kit (Parse Biosciences, Washington). For PBMC samples between 250 000–400 000 cells were fixed per sample, and for the RBC-depleted samples 500 000–850 000 cells per sample were fixed according to the manufacturer’s instructions. Cells were spun down at 200 x g at 4°C for 10 min before resuspension in a prefixation buffer. The cells suspension was then passed through a 40 μm strainer to remove cell clumps. A fixation additive was then added to the suspension and the sample was placed on ice for 10 min, before the addition of a permeabilizing solution and a further 3-min incubation on ice. A neutralization buffer was added before a further centrifugation step for 10 min, 200 x g at 4°C. The supernatant was removed, and cells were resuspended in Cell Buffer and DMSO added to the samples. The fixed cells were then processed using the Evercode WT Mini v2 kit according to the manufacturer’s instructions. This protocol uses successive pooling and barcode steps. In the first step, well barcodes are added, and reverse transcription of mRNA takes place within the cell. After this step the samples are then pooled and redistributed, and further barcodes ligated a further two times. In the third step UMI’s are added to the cDNA. In the last step, cells are lysed, the cDNA isolated and sequencing adapters and sub-library barcodes added by PCR. The resulting libraries were sequenced on the Nova-seq 6000 using an S1 flow cell and S1 Reagent Kit v1.5 (100 cycles) with PE 74bp_6 index_86bp reads.

Flow cytometry

250 000 live cells were stained for flow cytometry; cells were centrifuged at 320 x g for 5 min at 4°C. The supernatant was removed, and cells resuspended in 50μL of blocking solution containing 1:100 dilution of human IgG (Jackson ImmunoResearch, West Grove, Pennsylvania), 1:50 dilution True Stain Monocyte Blocker (Biolegend, California) and 1:800 dilution of Fixable Green Dead Cell Stain (ThermoFischer Scientific, Waltham, Massachusetts) in PBS. The samples were incubated at 4°C for 20 min. The samples were centrifuged at 320 x g and washed with FACS buffer (PBS supplemented with 2% FCS and 2mM EDTA). The cells were resuspended in 50uL of a solution containing a panel of antibodies for the following surface markers: CD14, CD8, CD45, HLA-DR, CD56, CD11c, CD19, CD25, CD123, CD45RA, LD, CD15, CD20, CD193, CCR7 and CD16. The mix was prepared in a buffer containing 50% FACS buffer and 50% Brilliant Stain Buffer (BD Biosciences, Franklin lakes, New Jersey). The antibodies (previously titrated) are summarized in key resources table. The cells were then incubated for 20 min at 4°C. Then centrifuged and washed with FACS buffer before incubating for 30 min with Foxp3 Fixation/Permeabilization working solution (Thermofischer Scientific) at room temperature in the dark. The cells were washed in permeabilization buffer, before the addition of the intracellular antibodies: CD3, CD4 and FOXP3 prepared in permeabilization buffer/Brilliant Stain Buffer (1:1 vol/vol) (details in Table S3) and incubated for 30 min at room temperature. The samples were further centrifuged at 620 x g and resuspended in FACS buffer before running on a FACSymphony A5 Cell Analyzer (BD Bioscience). The gating strategy is outlined in Figure S7.

Quantification and statistical analysis

Bioinformatic analysis

In total, we sequence 29 samples for the technology comparison, six using Evercode, six using Flex, six using Chromium Single Cell 3’ Gene Expression v3.1, eight using HIVE, and three using Chromium Single Cell 3’ Gene Expression v3.1 from the HIVE donors. From the HIVE experiment we considered only the 3 HIVE samples at day 0. The reads from all technologies were mapped to the human genome (hg38) and we created one gene-by-cell count matrix per sample. FASTQ files from Parse were generated using bcl-convert 3.8.4 and the split-pipe v1.0.5p was utilized to generate the count matrices. FASTQ files from 10X Flex were generated using 10X Genomics cellranger 6.0.2 mkfastq and cellranger 7.1.0 multi was utilized to generate the count matrices. FASTQ files from 10X 3′ were generated using 10X Genomics cellranger 6.0.2 mkfastq and cellranger 6.0.2 count was utilized to generate the count matrices. FASTQ files from HIVE were generated using Bcl2fastq for demultiplexing and BeeNet (Honeycomb) for data pre-processing to generate the count matrices.

The count matrices were further processed using BESCA²² and Scanpy.²⁹ Low cutoffs for number of genes expressed and number of UMIs per cell were used in order to capture the neutrophils. This method has been used by previous studies,³⁰ as neutrophils typically have a lower levels of gene expression, and generally have low levels of RNA meaning that they would be filtered out using strict thresholds that have been developed for PBMCs. We applied the same cut-offs for number of genes and UMIs detected for all samples in order to achieve high comparability. Cells that expressed at least 50 and not more than 10,000 genes and included at least 50 and not more than 50,000 UMIs were kept for downstream analysis. In addition, we removed cells with high mitochondrial gene expression above 10% of UMIs mapping to mitochondrial genes. This resulted in 753,435 total cell barcodes.

We further processed cells from each technology separately and using the same methodologies. Normalization was performed for the entire dataset using count depth scaling to 10,000 total counts per cell, resulting in the cp10k (counts per 10,000) unit. Count values were log-transformed using natural logarithm: ln(cp10k + 1). To reduce dataset dimensionality before clustering, the highly variable genes within the dataset were selected. Genes were defined as being highly variable when they have a minimum mean expression of 0.0125, a maximum mean expression of 3 and a minimum dispersion of 0.5. Technical variance was removed by regressing out the effects of count depth and mitochondrial gene content and the gene expression values were scaled to a mean of 0 and variance of 1 with a maximum value of 10. The first 50 principal components were calculated and used as input for calculation of the 10 nearest neighbors. The neighborhood graph was then embedded into two-dimensional space using the Uniform Manifold Approximation and Projection (UMAP) algorithm) Cell communities are detected using the Leiden algorithm³¹ at a resolution of 1.

We performed cell type annotation for the clustering of cells from each technology individually and mapped it to the entire dataset. The entire dataset was integrated and reclustered using the methodology described above and utilizing BBKNN to correct for differences between samples.³² We assessed the cell types by calculating signature scores for all signatures provided by Besca (https://github.com/bedapub/besca/blob/master/besca/datasets/genesets/CellNames_scseqCMs6_sigs.gmt). The score is the average expression of a set of genes subtracted with the average expression of a reference set of genes, calculated by Scanpy’s score_genes function (https://scanpy.readthedocs.io/en/stable/generated/scanpy.tl.score_genes.html). For the cell type composition comparison we removed clusters of doublets and other artifacts before calculating cell type fractions per sample. This resulted in 322,323 total cells.

For each single cell sequencing technology, we calculated the mean absolute error (MAE) and root mean squared error (RMSE) of the neutrophil abundance compared to the abundance obtained by FACS. For each donor, we averaged the neutrophil abundance from three FACS experiments (technical replicates). Then, for each donor and technology, we determined the absolute difference between the FACS-derived abundance and the sequencing technology-derived abundance. The average of these differences across donors is the MAE for each technology. Similarly, we calculated the RMSE by first determining the squared differences between the FACS-derived and sequencing technology-derived abundances, then averaging these squared differences, and finally taking the square-root of this average for each technology.

For the time course experiment, we sequence 18 samples, all on the 10X Flex technology. The dataset was processed as described above, but different filtering criteria were applied to the cells: number of genes: 200 - 4,000; number of UMI counts 200 - 15,000; maximal mitochondrial fraction 1% (see Figures S8A–S8C). In order to identify differentially expressed genes in neutrophils over time, we generated a pseudobulk gene-by-sample matrix for the neutrophils identified. We tested each time point (2 h, 4 h, 6 h, 8 h, 24 h) versus immediate processing (0 h) by fitting the model: ∼0 + Donor + Time point. We filtered genes at a cut-off of average transcripts per million larger than 0.25. 12,726 genes were assessed for differential expression using limma+voom (source: http://bioinf.wehi.edu.au/limma/). We chose relaxed cut-offs to allow for high sensitivity to detect changes, fold-change larger than 1.5 or less than 0.666 and false discovery rate smaller than 10%. Afterward we counted the number of up- or down-regulated genes according to these cut-offs. We applied gene set enrichment analysis using Camera.³³ We used multiple geneset collections, including MSigDB hallmark (source: https://www.gsea-msigdb.org/gsea/msigdb/human/collections.jsp), Reactome pathways (source: https://reactome.org), and an internally curated set of cancer immuno-therapy (CIT) signatures. The latter includes a Stress MP5 (meta-program 5) signature²⁶ which showed highest enrichment.

Published: September 15, 2025