A rule-based data-informed cellular consensus map of the human mononuclear phagocyte cell space

TL;DR: A rule-based data-informed approach to build next generation cellular consensus maps, using the human dendritic-cell and monocyte compartment in peripheral blood as an example, and providing a generalizable method for building consensus maps for the life sciences.

Abstract: Single-cell genomic techniques are opening new avenues to understand the basic units of life. Large international efforts, such as those to derive a Human Cell Atlas, are driving progress in this area; here, cellular map generation is key. To expedite the inevitable iterations of these underlying maps, we have developed a rule-based data-informed approach to build next generation cellular consensus maps. Using the human dendritic-cell and monocyte compartment in peripheral blood as an example, we performed computational integration of previous, partially overlapping maps using an approach we termed ‘backmapping’, combined with multi-color flow-cytometry and index sorting-based single-cell RNA-sequencing. Our general strategy can be applied to any atlas generation for humans and other species. Graphical Highlights Defining a consensus of the human myeloid cell compartment in peripheral blood 3 monocytes subsets, pDC, cDC1, DC2, DC3 and precursor DC make up the compartment Distinguish myeloid cell compartment from other cell spaces, e.g. the NK cell space Providing a generalizable method for building consensus maps for the life sciences

Summary

Introduction

• Such single-cell technologies allow for a fully data-driven analysis to establish cell maps of an organism, such as those proposed by the Human Cell Atlas consortium (Rozenblatt-Rosen et al., 2017).
• Reliable consensus maps are a prerequisite to reconcile conflicting data that might have been generated based on different data generating approaches (Edney, 2019; Monmonier, 2015).
• In order to establish a consensus map of the human mononuclear myeloid cell compartment the authors allow for the integration of prior knowledge in that they define a priori criteria for the cellular compartment under study in order to increase resolution and to allow 5 building of a consensus map.

Results

• Integrated phenotypic characterization of the myeloid cell compartment in human peripheral blood CC-BY-NC-ND 4.0 International licensea certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
• The copyright holder for this preprint (which was notthis version posted June 3, 2019.
• To integrate the identified DC subsets in map 1 and map 2 with each other, the authors computed a UMAP topology from the original map 1 single-cell transcriptome data comprising the DC cell space and overlaid the signatures of the map 2 DC subsets (pDC, cDC1, cDC2, pre-DC) .
• This analysis showed that if the totality of the Lin-CD16+ compartment is mapped back onto the Lin- UMAP topology , NK cells (CD56+), monocytes (CD56-CD16+/-) and granulocyte fractions (CD16high) are included in this cellular compartment.

Discussion

• Consensus maps are an important instrument within an iterative process of producing cellular maps of all organs and tissues in different species, including humans.
• Because the authors propose to include prior knowledge in the respective scientific field into the algorithm for generating such consensus maps, they define the overall strategy as being ‘data-informed’, combining prior knowledge and data-driven technologies including single-cell omics.
• CC-BY-NC-ND 4.0 International licensea certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
• The copyright holder for this preprint (which was notthis version posted June 3, 2019.
• BioRxiv preprint 20 providing the next iteration of this particular subspace in the myeloid cell map of human peripheral blood.

Acknowledgments:

• The authors thank Jessica Tamanini for critical review and editing of the manuscript.
• This work was supported by the German Research Foundation to JLS (GRK 2168, INST 217/577-1, EXC2151/1), by the HGF grant sparse2big to JLS, the FASTGenomics grant of 5 the German Federal Ministry for Economic Affairs and Energy to JLS and the EU project SYSCID under grant number 733100, also known as Funding.
• F.G is an EMBO YIP awardee and is supported by Singapore Immunology Network (SIgN) and Shanghai Institute of Immunology core funding.
• The authors declare that there are no competing interests.

Figure Legends

• Generating a new consensus map of the mononuclear myeloid cell compartment in human peripheral blood.
• (B) Visualization of ~1.4 mio. live CD45+Lin(CD3, CD19, 5 CD20, CD56)- cells after UMAP dimensionality reduction of the flow cytometry panel introduced in A (left panel), mononuclear myeloid cell compartment (second panel), overlay of index-sorted cells (third panel), UMAP topology of the index-sorted cells based on the single-cell transcriptome data .
• (B) Heatmap of 10 most significant marker genes for each of the 11 clusters identified and visualized in Figure 2A.
• (G) UMAP topology of scRNA-seq data derived from the map1 DC and mono subsets (left panel) and overlay of the NK cell signature onto this UMAP topology.
• 20 25 .CC-BY-NC-ND 4.0 International licensea certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

Tables S1:

• Cell types classified in the respective studies Data Table S1: 5 Data Table S1.csv.
• Gene signatures of the 11 clusters identified in their new scRNA-seq consensus map.

Data Table S2:

• CC-BY-NC-ND 4.0 International licensea certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
• The copyright holder for this preprint (which was notthis version posted June 3, 2019.
• Cell types classified in the respective studies .
• CC-BY-N -ND 4.0 Internatio al licensea certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

• Peripheral blood mononuclear cells (PBMC) Buffy coats or venipuncture blood were obtained from healthy donors (University hospital Bonn, local ethics vote 203/09) after written consent was given according to the Declaration of Helsinki.
• 10 Peripheral blood mononuclear cells (PBMC) were isolated by Pancoll (PAN-Biotech) density centrifugation from buffy coats.

METHOD DETAILS

• Whole blood or buffy coat was diluted in room temperature PBS (1:2 or 1:5, respectively) and layered onto polysuccrose solution (Pancoll; PAN Biotech, Germany) for the enrichment of mononuclear cells by density gradient centrifugation according to the manufacturer's instructions.
• Washed cells were incubated with L/D Marker DRAQ7 (BioLegend, USA) for 5 min at room temperature before acquisition and sorting of the cells using a BD FACSARIA III (BD BioSciences, USA).
• CC-BY-NC-ND 4.0 International licensea certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
• The authors new index-sorted single cell transcriptome dataset was based on the Smart-Seq2 protocol (Picelli et al., 2013).
• CDNA was diluted to an average of 200pg/µl and 100pg cDNA from each cell was tagmented by adding 1µl TD and 0.5µl ATM from a Nextera XT DNA Library Preparation Kit to 0.5µl diluted cDNA in each well of a fresh 384-well plate.

Cytospin preparation and May-Grünwald/Giemsa staining

• Cell populations of interest were sorted into 1.5 ml reaction tubes containing 200 µl FACS-buffer 5 using a BD FACSARIA III (BD BioSciences, USA).
• Whole blood was diluted in room temperature PBS (1:2) and layered onto polysuccrose solution (Pancoll; PAN Biotech, Germany) for the enrichment of mononuclear cells by density gradient 15 centrifugation according to the manufacturer's instructions.
• Sequenced single-cell data was demultiplexed using bcl2fastq2 v2.20.
• Based on the pseudoalignment estimated by Kallisto, transcript levels were quantified as transcripts per million reads (TPM).

Quality control

• Concerning their new index-sorted and Smart-Seq2-based single cell transcriptome dataset the following quality control scheme using various meta information was performed to obtain highquality transcriptome data: 1) We removed genes that are detected in less than 6 cells (0.2 percent of cells), 2) and removed cells that have less than 1,000 uniquely detected genes.the authors.the authors.
• Next, 25 .CC-BY-NC-ND 4.0 International licensea certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
• BioRxiv preprint 36 the authors filtered further outlier cells with 3) less than 50,000 unique reads, 4) less than 30% pseudoalignment of reads to the transcriptome, 5) a lower rate of endogenous-to-mitochondrial count rate of 2, 6).
• To reduce the influence of variation of sequencing depth among samples the authors applied a lognormalization to the data and scaled each cells gene expression profile to a total count of 10,000.
• The residuals of this regression are scaled and centered and used for further downstream analysis.

Dimensionality reduction and clustering

• This resulted in a total of 2491 genes, which were used as input for a principal component (PC) analysis.
• To test for cellular heterogeneity, the authors used a shared nearest neighbor (SNN)-graph based clustering algorithm implemented in the Seurat package.
• The authors used the first 10 principal components for constructing the SNN-graph and set the resolution to 1.
• Monocle was used to infer differentiation trajectories by using the Louvain clustering method, umap dimensionality reduction and the SimplePPT algorithm (Qiu et al., 2017) 25 .CC-BY-NC-ND 4.0 International licensea certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
• The copyright holder for this preprint (which was notthis version posted June 3, 2019.

Additional analysis

• Differentially expressed (DE) genes were defined using a Wilcoxon-based test for differential gene expression built in the Seurat pipeline (v.2.3.4) (Data Table S1).
• Top10 DE genes have been visualized using heatmap of hierarchical clustered gene expression 5 profiles.
• Gene signature enrichment analysis Single-cell RNA-Seq data is inherently sparse and a high-dropout rate is limiting the use of single marker genes to identify cell populations.
• In order to increase the power, the authors use both up and downregulated gene signatures for the calculation of the gene expression scores.
• The difference between these two is scaled and visualized.

To assess the single-cell RNA-Seq data of human dendritic cells and monocytes publicly available

• Under the Gene Expression Omnibus accession number GSE94820, the authors applied the processing 25 .
• CC-BY-NC-ND 4.0 International licensea certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
• The copyright holder for this preprint (which was notthis version posted June 3, 2019.
• Next, the authors followed the general data analysis scheme described at the Seurat package webpage 15 (https://satijalab.org/seurat/get_started_v1_4.html).
• Briefly, the authors used the filtered cell-gene matrix provided by 10x Genomics and imported the data and performed the analysis with the Seurat package.

Backmapping

• In order to compare the transcriptome profiles of monocytes isolated from the dataset derived 5 from GSE94820 (Villani et al., 2017) with the comprehensive PBMC dataset, the authors used the previously introduced canonical correlation alignment to combine datasets (Butler et al., 2018).
• The authors determined the mutual highly variable genes as the overlap of the 4.000 genes from each dataset with highest dispersion.
• The authors treated the different batches of the HCA dataset 25 as individual datasets and normalized them and the expression table of the consensus map .
• CC-BY-NC-ND 4.0 International licensea certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
• First, the authors repeated the steps above but without integration of the new consensus map data.

Data visualization

• In general, the ggplot2 package was used to generate figures (Wickham, 2016).
• 25 .CC-BY-NC-ND 4.0 International licensea certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
• The copyright holder for this preprint (which was notthis version posted June 3, 2019.

QUANTIFICATION AND STATISTICAL ANALYSIS

• Statistical analysis was performed using the R programming language.
• Statistical tests used are described in the figure legend or methods part, respectively.
• Differentially expressed genes have been identified using a Wilcoxon-based test for differential gene expression.
• If not otherwise stated a significance level of 0.1 was applied to adjusted p-values (Benjamini Hochberg).

DATA AND SOFTWARE AVAILABILITY

• Processed and raw scRNA-seq datasets are available through the Gene Expression Omnibus (GSE126422).
• Additional Data tables are provided in form of EXCEL Tables (Data S1, S2) Data Table S1: Data Table S1.csv 10 Gene signatures of the 11 clusters identified in their new scRNA-seq consensus map.

ADDITIONAL RESOURCES

• In addition, the authors provide an interactive web tool to visualize the single-cell RNA-Seq data together with the flow cytometry data at https://paguen.shinyapps.io/DC_MONO/ (external database S1).
• .CC-BY-NC-ND 4.0 International licensea certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
• The copyright holder for this preprint (which was notthis version posted June 3, 2019.