No detectable alloreactive transcriptional responses under standard sample preparation conditions during donor-multiplexed single-cell RNA sequencing of peripheral blood mononuclear cells

Abstract

Background: Single-cell RNA sequencing (scRNA-seq) provides high-dimensional measurements of transcript counts in individual cells. However, high assay costs and artifacts associated with analyzing samples across multiple sequencing runs limit the study of large numbers of samples. Sample multiplexing technologies such as MULTI-seq and antibody hashing using single-cell multiplexing kit (SCMK) reagents (BD Biosciences) use sample-specific sequence tags to enable individual samples to be sequenced in a pooled format, markedly lowering per-sample processing and sequencing costs while minimizing technical artifacts. Critically, however, pooling samples could introduce new artifacts, partially negating the benefits of sample multiplexing. In particular, no study to date has evaluated whether pooling peripheral blood mononuclear cells (PBMCs) from unrelated donors under standard scRNA-seq sample preparation conditions (e.g., 30 min co-incubation at 4 °C) results in significant changes in gene expression resulting from alloreactivity (i.e., response to non-self). The ability to demonstrate minimal to no alloreactivity is crucial to avoid confounded data analyses, particularly for cross-sectional studies evaluating changes in immunologic gene signatures.

Results: Here, we applied the 10x Genomics scRNA-seq platform to MULTI-seq and/or SCMK-labeled PBMCs from a single donor with and without pooling with PBMCs from unrelated donors for 30 min at 4 °C. We did not detect any alloreactivity signal between mixed and unmixed PBMCs across a variety of metrics, including alloreactivity marker gene expression in CD4+ T cells, cell type proportion shifts, and global gene expression profile comparisons using Gene Set Enrichment Analysis and Jensen-Shannon Divergence. These results were additionally mirrored in publicly-available scRNA-seq data generated using a similar experimental design. Moreover, we identified confounding gene expression signatures linked to PBMC preparation method (e.g., Trima apheresis), as well as SCMK sample classification biases against activated CD4+ T cells which were recapitulated in two other SCMK-incorporating scRNA-seq datasets.

Conclusions: We demonstrate that (i) mixing PBMCs from unrelated donors under standard scRNA-seq sample preparation conditions (e.g., 30 min co-incubation at 4 °C) does not cause an allogeneic response, and (ii) that Trima apheresis and PBMC sample multiplexing using SCMK reagents can introduce undesirable technical artifacts into scRNA-seq data. Collectively, these observations establish important benchmarks for future cross-sectional immunological scRNA-seq experiments.

Keywords: Alloreactivity; PBMCs; Sample multiplexing; Sample preparation; scRNA-seq.

Fig. 1

Schematic overview of experimental design. PBMCs from 8 healthy HLA-mismatched donors (tubes on left) were barcoded with MULTI-seq LMOs (black double-helix hybridized to red DNA barcode) and BD single-cell multiplexing kit (SCMK) antibodies (black antibody conjugated to teal DNA barcode). Cells were then strategically pooled to directly assess whether mixing HLA-mismatched PBMCs during scRNA-seq causes alloreactivity

Fig. 2

MULTI-seq and SCMK classifications largely match in silico genotyping, with lower SCMK classification efficiency and bias against activated CD4+ T cells. a Sample classification results from three demultiplexing pipelines (e.g., deMULTIplex, souporcell, and demuxEM) projected onto MULTI-seq (top) and SCMK (bottom) sample barcode space. n = 4032 cells from microfluidic lane #3. b Classification frequencies across all PBMC cell types following SCMK sample demultiplexing. c SCMK unclassified cells in CD4+ T cell gene expression space. n = 6879 CD4+ T cells

Fig. 3

Mixing HLA-mismatched PBMCs does not cause allogenic response during multiplexed scRNA-seq sample preparation. a Sample classification results plotted as densities in PBMC gene expression space (top left) grouped according to unmixed donor A PBMCs (bottom left), mixed Donor A PBMCs (bottom right), and Donors A-C PBMCs (top right). b Expression of genes known to be upregulated (e.g., IFNG and CD40LG) or downregulated (e.g., DUSP1 and FOS) by CD4+ T lymphocytes during an allogenic response across unmixed (black) and mixed (red) donor A CD4+ T cell subsets. c Representative CD4+ T cell, CD8+ T cell, CD14+ monocyte, CD16+ monocyte, and NK cell gene expression embeddings following iterative subsetting to select equal numbers from each JSD comparison group. Cells are colored according to donor ID (e.g., a, b, c) and mixing status (e.g., − = unmixed, + = mixed). n = 1336 CD4+ T cells, 448 CD8+ T cells, 864 CD14+ monocytes, 136 CD16+ monocytes, and 224 NK cells. d JSD analysis summary. Bar plots denote average JSD between PBMC donors (white), mixed/unmixed donor A cells (beige), technical replicates (gray), and donor A cells following label permutation (red). Difference in JSD scores between mixed/unmixed and technical replicates depicted in black. Error bars denote +/− 1 standard deviation. n = 100 iterations