key: cord-0807911-p5k4zqb0
authors: Geanon, Daniel; Lee, Brian; Gonzalez‐Kozlova, Edgar; Kelly, Geoffrey; Handler, Diana; Upadhyaya, Bhaskar; Leech, John; De Real, Ronaldo M.; Herbinet, Manon; Magen, Assaf; Del Valle, Diane; Charney, Alexander; Kim‐Schulze, Seunghee; Gnjatic, Sacha; Merad, Miriam; Rahman, Adeeb H.
title: A streamlined whole blood CyTOF workflow defines a circulating immune cell signature of COVID‐19
date: 2021-02-16
journal: Cytometry A
DOI: 10.1002/cyto.a.24317
sha: 147ed3c9f1c4c6f5302402cb22c614da28dd4808
doc_id: 807911
cord_uid: p5k4zqb0

Mass cytometry (CyTOF) represents one of the most powerful tools in immune phenotyping, allowing high throughput quantification of over 40 parameters at single‐cell resolution. However, wide deployment of CyTOF‐based immune phenotyping studies are limited by complex experimental workflows and the need for specialized CyTOF equipment and technical expertise. Furthermore, differences in cell isolation and enrichment protocols, antibody reagent preparation, sample staining, and data acquisition protocols can all introduce technical variation that can confound integrative analyses of large data‐sets of samples processed across multiple labs. Here, we present a streamlined whole blood CyTOF workflow which addresses many of these sources of experimental variation and facilitates wider adoption of CyTOF immune monitoring across sites with limited technical expertise or sample‐processing resources or equipment. Our workflow utilizes commercially available reagents including the Fluidigm MaxPar Direct Immune Profiling Assay (MDIPA), a dry tube 30‐marker immunophenotyping panel, and SmartTube Proteomic Stabilizer, which allows for simple and reliable fixation and cryopreservation of whole blood samples. We validate a workflow that allows for streamlined staining of whole blood samples with minimal processing requirements or expertise at the site of sample collection, followed by shipment to a central CyTOF core facility for batched downstream processing and data acquisition. We apply this workflow to characterize 184 whole blood samples collected longitudinally from a cohort of 72 hospitalized COVID‐19 patients and healthy controls, highlighting dynamic disease‐associated changes in circulating immune cell frequency and phenotype.

correlates of disease progression. Mass cytometry allows for an evaluation of over 40 parameters in a single sample, thereby enabling comprehensive characterization of immune cells in limited samples. However, the need for large numbers of reagents, long sample processing workflows and expensive and complicated CyTOF hardware present challenges that limit wide adoption of mass cytometry assays. These challenges are further amplified in studies involving clinical sample collection at multiple sites that may differ in their available sample processing resources and levels of technical expertise.

To address these challenges, we have optimized a standardized, streamlined sample processing workflow to allow whole blood to be easily stained, stabilized, preserved, and transferred to a central CyTOF core for final processing and data acquisition. This whole blood sample processing workflow leverages the commercially available Fluidigm MaxPar Direct Immune Profiling Assay (MDIPA), which incorporates a dry tube containing a 30-marker broad immunophenotyping panel. Pre-mixed dry antibody panels offer significant advantages over a conventional liquid antibody for flow cytometry workflows by significantly reducing the processing time, technical variation and potential errors associated with pipetting multiple individual antibodies, and these advantages are amplified when using larger antibody panels. 1 Applying this assay to whole blood rather than PBMCs offers further sample sparing advantages by allowing the assay to be performed with only 270 μl of whole blood, eliminating the labor and technical variation associated with PBMC isolation, and allowing analysis of granulocyte subsets that are otherwise removed by density centrifugation. The MDIPA workflow has previously been found to show excellent intra-and inter-site reproducibility using both whole blood and PBMCs. 2 Despite these advantages, the standard MDIPA protocol as provided by Fluidigm still requires approximately 2 h of upfront sample processing time including several centrifugation steps, followed by an overnight incubation and suggests data acquisition within 48 h of sample staining. This workflow generally limits sample collection to labs that have the necessary technical expertise and resources for sample processing and on-site mass cytometry instrumentation.

To facilitate broader adoption of this assay in studies involving multiple clinical sample collection sites we have adapted this dry MDIPA antibody panel as part of a protocol utilizing commercially available SmartTube proteomic stabilizer to significantly reduce sample processing time and hardware requirements at the site of sample collection ( Figure S1 ). By minimizing initial sample processing requirements and transferring downstream processing steps to a central core, this workflow facilitates robust, highly standardized CyTOF immune monitoring in studies involving multiple clinical sites that cannot accommodate complex sample processing workflows and that do not have their own mass cytometry instrumentation. We further demonstrate the application of this workflow to study a large number of blood samples collected from hospitalized COVID-19 patients, highlighting its utility in facilitating large-scale standardized immune monitoring initiatives. Table 1 . Patient COVID-19 disease severity was based primarily on respiratory symptoms and need for supplemental oxygen therapy. Laboratory values for cardiovascular, renal, and hepatic function were also considered and adjusted for according to patient weight, age, and biological sex. The need for renal replacement therapy was also considered. This severity scoring system accounts for similar variables as SOFA, MODs, and APACHE II. ECMO was not used in this cohort. In experiments to evaluate the performance of additional antibody clones (Table 2) pre-and post-Prot1 fixation, replicate aliquots of the same blood samples were either stained as fresh whole blood and then fixed and frozen or were fixed and frozen as unstained aliquots and then thawed and lysed as described above. The thawed samples were then resuspended in a volume of Cell Staining Buffer equal to the volume of the original blood aliquot for staining.

SmartTube fixation and thaw/lyse results in partial permeabilization of cells and, consistent with prior results, 6 we found that addition of 100 U/ml of heparin was critical to prevent non-specific staining of eosinophils when adding antibodies to SmartTube-fixed whole blood.

However, this was not found to be necessary when staining fresh whole blood.

It is important to note that the standard SmartTube thawing protocol as per the manufacturer's instructions worked well for all healthy donor samples used in our initial validation experiments; however, when applying this protocol to blood collected from hospitalized COVID-19 patients we observed several instances in which the stabilized samples appeared to be partially clotted and exhibited high amounts of debris after thawing and lysis, which we suspect may be related to polymerized fibrin or other plasma factors related to COVID-19 disease-associated coagulopathy. If not addressed, this debris contributed to overall poor sample and staining quality and in some cases precluded analysis of samples. We found that following the red blood cell lysis washes with three additional large volume washes using 10 ml of PBS + 0.2% BSA with centrifugation at 250 rcf and followed by filtration through a 70 micron filter depleted the majority of this debris and permitted effective analysis of blood samples that would otherwise have been discarded.

Immediately prior to data acquisition, samples were washed with Cell staining on non-B cells, which occurred to a varying degree across subjects but was particularly notable in some samples. We have determined that this artifact is caused by donor-specific serum factors interacting with specific reagents in specific lots of the MDIPA. 9 While these artifacts could lead to erroneous data interpretation and negatively impact unbiased clustering approaches, they could easily be overcome in manual gating analyses by avoiding CD19 as an exclusion parameter for non-B cells and instead using CD20, which did not show any evidence of artifactual staining. After gating the data, the impact of each tested condition on relative staining quality was evaluated in two ways: (1) in cell frequencies, we derived the percentage of variance from each covariate (i.e., age, timepoint, severity) that explained changes in celltype frequencies by using the algorithm variancePartition. 14, 15 Next, to more precisely identify changes in cellular populations we performed differential abundance analysis using the limma package. 16 Briefly, the annotated cell frequencies were used as input into a linear model fit using Severity as the outcome variable. Additionally, we modeled within-patient variability by correlating biological replicates per cell type into a consensus correlation, which is a robust average of the individual correlations and allows us to adjust the linear model for individuals. Since all samples were stained using a fixed starting blood volume of 270 μl, to verify the observations using cell frequencies, we used the total CD45+ cell numbers as input, adjusted for sample size, and log2 normalized the cell frequencies before using linear modeling.

Finally, the results were adjusted for multiple observations using the Benjamini-Hochberg method, and the figures were produced with the ggplot2 package, 15 to represent fold change and false discovery rate (FDR). We performed one analysis evaluating the relative frequencies of more broadly defined cell subsets as a percentage of all CD45+

immune cells, and a second analysis where neutrophils were excluded and the frequencies of more granularly defined subsets were evaluated as a percentage of all non-neutrophils.

3.1 | SmartTube stabilization prior to MDIPA staining negatively impacts several antibodies SmartTube proteomic stabilizer allows whole blood to be fixed and preserved with the addition of a single buffer followed by a 10-min incubation, after which blood can be transferred to long-term storage at −80 C and shipped to a remote core for downstream sample staining and processing. This workflow offers the advantage of a very rapid workflow that entails minimal sample processing at the site of collection, and has previously been effectively used to facilitate complex immune monitoring studies. 17, 18 However, a limitation of this workflow is that the whole blood stabilization entails fixation, which is expected to impact some antibody epitopes. We evaluated this by taking two parallel aliquots of whole blood, processing one with the conventional MDIPA workflow, and fixing and freezing the second with SmartTube Prot1 stabilizer, followed by staining with the MDIPA panel. The data were analyzed by manual gating all major immune cell subsets and evaluating expression patterns of all 30 markers across all the gated populations ( Figure 1A ). We found that staining of some markers was preserved post-fixation, while others were significantly compromised in comparison to the conventional MDIPA workflow as shown by a significant reduction in the correlation between marker expression across populations ( Figure 1B) . Notably, most chemokine receptors were dramatically affected, in many cases showing high non-specific background staining resulting in an inability to distinguish true positive and negative populations, and an overall loss of staining index ( Figure 1C ). These results are consistent with our prior observations of the effect of formaldehyde fixation on chemokine receptor expression. 19 We also observed a reduced staining index for CD25 and CD127, two markers used to resolve CD4+ T regulatory cells.

Thus, while immediate SmartTube fixation offers advantages in rapid blood processing, it is not compatible with all the antibodies used in the MDIPA panel. to a remote site for subsequent processing. However, the fact that the cells were being fixed without much dilution of the staining antibodies raised the potential concern of elevated staining background and reduced staining quality due to cross linking of free antibodies to the cells. To evaluate this, we stained parallel aliquots of blood from three donors using the MDIPA antibody panel, after which one aliquot was processed according to the standard MDIPA protocol, while the other was fixed, frozen with SmartTube Prot1 and subsequently processed following the SmartTube thaw/lyse protocol. The data were manually gated and markers were compared across populations as in Figure 1 . When evaluating marker expression between these two protocols, we found that the SmartTube workflow resulted in nearly identical staining patterns to the conventional MDIPA workflow ( Figure 2 ). Almost all markers showed correlations of over 98% between the two protocols, and in many cases staining indexes were slightly higher with the SmartTube workflow indicating better resolution of cell populations ( Figure 2B ,C). In addition, we found that the SmartTube-based workflow resulted in an approximately 35% greater recovery of CD45+ cells from the same starting volume of blood ( Figure 3A ). Overall relative cell population frequencies were highly correlated between both protocols ( Figure 3B ), and both protocols were able to clearly show consistent inter-individual differences in the frequency of CD4 T cell memory subsets defined by differential chemokine receptor staining ( Figure 3C ).

Accurate immunophenotyping requires active steps to identify and minimize potential sources of non-specific antibody staining. One well-known source of non-specific antibody staining is binding by Fcreceptors. While this can be mitigated using Fc-blocking reagents, this is generally not needed in whole blood staining workflows since endogenous serum antibodies effectively occupy and block Fc-receptors. Another source of non-specific antibody staining that is more specific to CyTOF whole blood workflows is a charge-based interaction between cationic granule proteins which can result in nonspecific antibody binding by eosinophils. This non-specific interaction can be effectively blocked using heparin, a strongly anionic compound. 6 While this problem primarily presents when performing intracellular staining on fixed whole blood, we were concerned that SmartTube fixation of the blood with relatively minimal antibody dilution may result in non-specific binding of MDIPA antibodies. The majority of our initial tests were conducted using blood collected in sodium heparin vacutainers, which would be expected to prevent any such issues; however, we wanted to evaluate whether higher levels of non-specific eosinophil antibody binding would occur when staining non-heparinized blood and, if so, whether heparin supplementation could mitigate such artifacts. To evaluate this, we collected and stained three aliquots of blood collected from the same individual in either a sodium heparin tube, an EDTA tube, or a sodium citrate CPT tube. Samples were stained with the MDIPA panel and fixed with SmartTube stabilizer and analyzed as described above. We found that overall staining quality was very similar between the tube types, with no evidence of elevated eosinophil background staining in EDTA and citrate CPT tubes relative to heparin tubes ( Figure S3A ). Cell-type specific marker expression patterns ( Figure S3B -C) and frequencies ( Figure S3D ) were highly correlated between the blood samples collected in all three tube types. However, we did observe some changes in relative staining intensity for some markers, most notably CD8a, which showed an almost 10-fold reduction in staining intensity in blood collected in EDTA tubes relative to that collected heparin tubes ( Figure S3C ). Furthermore, supplementing additional heparin into the EDTA tubes did not have any measurable impact on staining quality ( Figure S4 ). Overall, these data show that while staining quality is optimal in heparin tubes, overall staining patterns and relative population frequencies (data not shown) are largely preserved across tube types, indicating that this SmartTube stabilization workflow is broadly applicable to perform CyTOF immunophenotyping on blood samples collected in multiple tube types.

In addition to collection tube type, we also considered delays in Major immune cell subsets were defined based on the core phenotyping markers, and cell-type specific marker expression patterns for each of the tested antibodies was evaluated on the paired pre-and post-fixation. While the majority of antibodies showed highly correlated expression patterns, some showed extremely low correlations indicating a severe impact of fixation ( Figure 4A ). In most cases, the low correlations reflected a loss of detectable expression postfixation which was the case for NKG2A, NKG2C, CD88, and CD40;

F I G U R E 4 Evaluation of fixation-stable markers to supplement the core MDIPA panel on SmartTube-fixed whole blood. Aliquots of whole blood were stained with several panels of antibodies either prior to or following SmartTube fixation to evaluate the fixation sensitivity of different antibody clones. Each panel shared a core set of fixation-stable markers, which were used to gate major immune cell subsets. however, in the case of IgA and IgM, expression was only detectable on the post-fixed samples ( Figure 4B,C) . This is likely due to interference from soluble serum IgA and IgM antibodies in fresh whole blood, which are effectively washed away during the Prot1 fixation and lysing protocol. Even amongst antibodies that showed highly correlated expression patterns between the pre-and post-fixed samples, we observed some that showed a reduced staining index post fixation (e.g., CD71) and others that showed increased staining index (e.g., CD29). However, some of these differences could be minimized by re-titration of the antibodies on the fixed blood samples, and all of these antibodies ultimately represent reasonable potential candidates to supplement MDIPA-stained blood samples at the site of central downstream processing. Given that the high abundance of neutrophils was a major driver of relative cell frequency, we evaluated differences in more granularly-defined immune cell subsets as a percentage of nonneutrophils. Differential abundance analysis adjusting for covariates highlighted several differences relating to both COVID status and disease severity ( Figure 5C ). Expanding on the overall reduction in T cells, this analysis showed reduction in several T cell subsets including naïve CD4+ and CD8+ T cells, γδ T cell, and most dramatically CD161+ MAIT-like CD8+ T cells. We also observed reductions in CD56hi CD16low NK cells and Innate Lymphoid

Cells. Reciprocally, we observed elevated frequencies of CD16low monocytes, plasmablasts and CD38 + HLADR+ activated CD4+ and CD8+ T cells, though the magnitude of the changes in these latter subsets varied considerably between patients ( Figure S7 ), consistent with other reports. 23 In most cases, the most dramatic and significant fold changes were observed when comparing healthy controls to either of the hospitalized COVID-19 cohorts.

Differences in relation to disease severity between the moderate and severe patients were more limited, though we noted a reduced frequency of conventional and plasmacytoid dendritic cells in severe disease, and a slight increase in the frequency of activated T cells.

In addition to changes in overall cell type frequencies, we also observed profound changes in the phenotype of circulating myeloid cells, and longitudinal sampling indicated that some of these changes were highly dynamic. For example, we also observed a striking increase in expression of CD169 (sialoadhesin) on monocytes in a subset of both moderate and severe COVID-19 patients ( Figure 6A ). CD169 is constitutively expressed on subsets of tissue resident macrophages 20 and is induced on circulating monocytes by type I interferon signaling. 21 Figure 6B ), consistent with other reports. 24, 25 Given that patients likely present to the hospital at different times relative to their initial date of infection, differences in CD169 may likely reflect differences in timing from onset of infection, which in many cases is difficult to accurately determine, suggesting caution in overinterpreting CD169 expression in relation to disease severity.

We also observed subpopulations of monocytes with reduced HLA-DR expression, which were notably increased in COVID-19

patients and associated with disease severity ( Figure 6C) , which was consistent with previous reports. [26] [27] [28] Longitudinal analyses showed that dysregulated HLA-DR expression was also dynamic and typically increased over the course of the hospital stay ( Figure 6D ). However, the kinetics of these changes were typically slower than those of CD169, and in some patients, particularly those with more severe disease, reduced HLA-DR expression persisted for the duration of the hospital stay.

In addition to changes in monocyte phenotype, we also observed dramatic changes in neutrophil phenotype, most notably elevated CD64 expression, which has previously been seen in the setting of sepsis and has been proposed as a biomarker of bacterial infection. 29 Once again, longitudinal analyses revealed the dynamic nature of these changes, with a progressive decrease in most patients. However, as with reduced HLA-DR expression on monocytes, elevated CD64 expression on neutrophils was more persistent and, in many cases, remained elevated above the levels seen in controls for the duration of the hospital stay. This is also consistent with our earlier report of elevated neutrophil CD64 expression in the absence of monocyte CD169 expression in children suffering from multisystem inflammatory syndrome >1 month after initial SARS-CoV-2 infection. 30 Together, these results highlight key features of COVID-19 associated immune dysregulation and broadly demonstrate the applicability of this assay to large-scale studies to characterize changes in circulating immune cell frequency and phenotype.

F I G U R E 6 (Continued)

Overall, we believe that the data provided here provide a validation of a streamlined, sample-sparing whole blood immune monitoring workflow ( Figure S1 ) and offer data to support important considerations in terms of vacutainer selection, storage duration prior to staining, and the incorporation of additional markers to supplement the core MDIPA panel. We also demonstrate and discuss important Furthermore, our data clearly illustrate the highly dynamic nature of these phenotypic changes and establish that it is critically important to consider the time from infection onset and the kinetics of disease progression in interpreting these changes. Studies that relate differences in CD169 to COVID-19 disease severity may be confounded by insufficient consideration of these kinetics. 31, 32 This protocol already offers a simple workflow that requires only However, despite the improvements that we have presented here to streamlined and simplify the workflow, the protocol still requires an appropriate BSL-approved environment to process the blood samples, and the ability to accurately pipette and time the addition of the necessary volumes, which may still not be possible at some resourcelimited clinical collection sites. Thus, a next logical step toward enabling even broader adoption of these workflows would be to prepackage the lyophilized antibody panel as part of a syringe based fixed-volume blood collection device, such as TruCulture system, 33 and to automate the timing and addition of the stabilization buffer and cryopreservation steps, using a device such as the SmartTube base station.

Precision Immunology COVID-19 Repository effort

Helios mass cytometry instrumentation at the Human Immune Monitoring Center was obtained with support from S10OD023547. AUTHOR CONTRIBUTIONS

Data curation; formal analysis; investigation; methodology; writing-original draft; writing-review and editing. Brian Lee: Data curation

Gonzalez-Kozlova: Formal analysis; methodology; visualization; writing-original draft. Geoffrey Kelly: Data curation; investigation; methodology; writing-review and editing. Diana Handler: Data curation; investigation; methodology; writing-review and editing. Bhaskar Upadhyaya: Investigation; methodology; writing-review and editing

Ronaldo De Real: Data curation; investigation; methodology

Assaf Magen: Formal analysis; visualization. Diane Del Valle: Data curation; software; writing-review and editing. Alexander Charney: Funding acquisition; project administration. Seunghee Kim-Schulze: Project administration; supervision; writing-review and editing. Sacha Gnjatic: Funding acquisition

writing-review and editing. Miriam Merad: Funding acquisition

Adeeb Rahman: Conceptualization; data curation

funding acquisition; methodology; project administration; supervision; visualization; writing-original draft

Stabilization of preoptimized multicolor antibody cocktails for flow cytometry applications

Multi-site reproducibility of a human immunophenotyping assay in whole blood and peripheral blood mononuclear cells preparations using CyTOF technology coupled with Maxpar Pathsetter, an automated data analysis system

Sampling the host response to SARS-CoV-2 in hospitals under siege

Propagation, inactivation, and safety testing of SARS-CoV-2. Viruses

Average overlap frequency: a simple metric to evaluate staining quality and community identification in high dimensional mass cytometry experiments

Heparin reduces nonspecific eosinophil staining artifacts in mass cytometry experiments. Cytometry A

An updated debarcoding tool for mass cytometry with cell type-specific and cell sample-specific stringency adjustment

Clustergrammer, a web-based heatmap visualization and analysis tool for high-dimensional biological datajScientific Data

Donor-specific plasma factors contribute to aberrant staining artifacts with a commercial lyophilized mass cytometry antibody panel

SCANPY: large-scale single-cell gene expression data analysis

Dimensionality reduction for visualizing single-cell data using UMAP

Uniform Manifold Approximation and Projection for Dimension Reduction

Dream: powerful differential expression analysis for repeated measures designs

variancePartition: interpreting drivers of variation in complex gene expression studies

limma powers differential expression analyses for RNA-sequencing and microarray studies

Clinical recovery from surgery correlates with singlecell immune signatures

Application of phospho-CyTOF to characterize immune activation in patients with sickle cell disease in an ex vivo model of thrombosis

Development of a comprehensive antibody staining database using a standardized analytics pipeline

Whole blood immunophenotyping uncovers immature neutrophil-to-VD2 T-cell ratio as an early marker for severe COVID-19

Higher level of neutrophil-tolymphocyte is associated with severe COVID-19

Excessive neutrophils and neutrophil extracellular traps in COVID-19

Deep immune profiling of COVID-19 patients reveals patient heterogeneity and distinct immunotypes with implications for therapeutic interventions

Monocyte CD169 expression as a biomarker in the early diagnosis of COVID-19

A distinct innate immune signature marks progression from mild to severe COVID-19

Complex immune dysregulation in COVID-19 patients with severe respiratory failure

Early phases of COVID-19 are characterized by a reduction of lymphocyte populations and the presence of atypical monocytes

Monocyte HLA-DR measurement by flow cytometry in COVID-19 patients: an interim review

Neutrophil CD64 expression as a biomarker in the early diagnosis of bacterial infection: a meta-analysis

Mapping systemic inflammation and antibody responses in multisystem inflammatory syndrome in children (MIS-C)

Elevated calprotectin and abnormal myeloid cell subsets discriminate severe from mild COVID-19

Mass cytometry and artificial intelligence define CD169 as a specific marker of SARS-CoV2-induced acute respiratory distress syndrome

Standardized whole blood stimulation improves immunomonitoring of induced immune responses in multi-center study

Supporting Information section at the end of this article. How to cite this article: Geanon D

A streamlined whole blood CyTOF workflow defines a circulating immune cell signature of COVID-19

The authors have no conflicts to disclose.