Nuclei multiplexing with barcoded antibodies for single-nucleus genomics

Abstract

Single-nucleus RNA-Seq (snRNA-seq) enables the interrogation of cellular states in complex tissues that are challenging to dissociate, including frozen clinical samples. This opens the way, in principle, to large studies, such as those required for human genetics, clinical trials, or precise cell atlases of large organs. However, such applications are currently limited by batch effects, sequential processing, and costs. To address these challenges, we present an approach for multiplexing snRNA-seq, using samplebarcoded antibodies against the nuclear pore complex to uniquely label nuclei from distinct samples. Comparing human brain cortex samples profiled in multiplex with or without hashing antibodies, we demonstrate that nucleus hashing does not significantly alter the recovered transcriptome profiles. We further developed demuxEM, a novel computational tool that robustly detects intersample nucleus multiplets and assigns singlets to their samples of origin by antibody barcodes, and validated its accuracy using gender-specific gene expression, speciesmixing and natural genetic variation. Nucleus hashing significantly reduces cost per nucleus, recovering up to about 5 times as many single nuclei per microfluidc channel. Our approach provides a robust technique for diverse studies including tissue atlases of isogenic model organisms or from a single larger human organ, multiple biopsies or longitudinal samples of one donor, and largescale perturbation screens.

CellTag Indexing: genetic barcode-based sample multiplexing for single-cell genomics

“…Current multiplexing methods are based on transient transfection or temporary molecular interactions with the cell or nucleus surface [7][8][9][10][11][12] . Although relative to CellTag Indexing, this offers faster labeling of cells, it does not support long-term labeling.…”

“…Recently, several "label-and-pool" approaches have been developed to mark individual cells of the same sample with a distinct barcode prior to pooling and processing in the same single-cell RNA-sequencing (scRNA-seq) run [7][8][9][10][11][12] . For example, cells can be tagged with barcoded antibodies 9,12 , chemically labeled with DNA oligonucleotides 8,10 , or transiently transfected with DNA oligonucleotides 11 , such that sample identifiers for each cell can be read, in parallel with their transcriptomes. Similarly, several other methods exist to couple genetic perturbations with barcodes [13][14][15][16][17] , although these have not been demonstrated to support reliable, large-scale sample multiplexing.…”

Cumulus: a cloud-based data analysis framework for large-scale single-cell and single-nucleus RNA-seq

“…Cumulus demultiplexes cell hashing and nucleus hashing data using the DemuxEM algorithm, which we recently described 11 .…”

“…Cumulus supports analysis starting from a variety of input modalities, such that scientists can use it as a single framework for diverse data types, all of which share a single cell/nucleus transcriptome as a core readout ( Supplementary Table 1). These include: droplet-based 2,7 (3' or 5' ends, with UMIs) and plate-based 8 (full length, no UMI) sc/snRNA-seq (Methods); CITEseq 9 , which simultaneously measures mRNA expression and the abundance of oligo-tagged surface antibodies in single cells (Methods), data from both cell 10 or nucleus 11 hashing experiments, which are lab techniques that reduce batch effects and cell/nucleus profiling costs, using a probabilistic demultiplexing algorithm 11 (Methods); and Perturb-seq methods for pooled CRISPR screens [12][13][14][15][16] with scRNA-seq readout (Methods). Other mainstream, non-Cloud based analysis packages, such as Cell Ranger 7 , Seurat 17 , or SCANPY 18 , currently support only some of these input data types, posing a potential burden for users (Supplementary Table 1).…”

“…Pegasus utilizes SCANPY's AnnData data structure18 to store gene-count matrices and analysis results. More implementation details are discussed in the subsequent sections and Cumulus supports feature-count matrix generation of CITE-seq 9 , cell hashing 10 , nucleus11 hashing and Perturb-seq[12][13][14][15][16] protocols, using either 10x Genomics V2 or V3 chemistry. Each feature-count matrix generation job runs parallelly on a separate compute node with 1 CPU, 32 GB memory and 100 GB disk space, and calls 'generate_count_matrix_ADTs', a fast C++ program we implemented, to extract the matrix from sequence reads in FASTQ files.…”