Single-cell analysis tools for drug discovery and development

Abstract

The genetic, functional or compositional heterogeneity of healthy and diseased tissues presents major challenges in drug discovery and development. Such heterogeneity hinders the design of accurate disease models and can confound the interpretation of biomarker levels and of patient responses to specific therapies. The complex nature of virtually all tissues has motivated the development of tools for single-cell genomic, transcriptomic and multiplex proteomic analyses. Here, we review these tools and assess their advantages and limitations. Emerging applications of single cell analysis tools in drug discovery and development, particularly in the field of oncology, are discussed.

Figure 1. Quantitative single cell transcriptomic methods

Two separate, but conceptually similar methods, with similar throughput capabilities, are illustrated in this figure, along with representative data. a. The CytoSeq method is based upon isolating individual cells within 30 micrometer diameter (20 picoliter volume) wells, and then placing into each well a single barcoded bead. b. Each barcoded bead is designed with the shown structure. Each bead contains tens to hundreds of millions of distinct oligonucleotide primers which are each comprised of a barcode that identifies the bead (and thus the single cell), plus a molecular index (UMI) that is associated with a particular mRNA capture sequence. After bead and cell co-localization within a well, cells are lysed and mRNAs are captured via hybridization onto specific bead-bound oligonucleotides. The beads are then all removed from the well-plate, and all amplification reactions are carried out in a single tube. Adapted from . c. The microdrop-based in Drop technique for single cell transcriptomics. For this method, single cells are entrained into a single droplet, along with a hydrogel microsphere. Each hydrogel microsphere contains photo-cleavable oligonucleotide primers that have a similar construction to the bead shown in part b, while the droplets contain the cell lysis buffers and reverse transcription (RT) reagents, so that the whole process from cell capture and lysis to signal amplification happens separately in each droplet. d. A snapshot of representative data from an inDrop study of the kinetics of differentiation of mouse embryonic stem (mES) cells following leukemia inhibitory factor (LIF) withdrawal. For this plot, data sets representing 5 time points are analyzed using principal component anaysis to reveal asynchrony in mES cell differentiation. Each dot represents a single cell. Adapted from ref .

Figure 2. Emerging single cell proteomics methods

a. Mass cytometry uses antibodies (Abs), encoded with transition metal containing mass tags, to label proteins of interest. Cells are fixed and permeabilized so as to permit Ab-staining of cytoplasmic proteins. Single cells are entrained into vapor and atomized. A mass filter separates the transition metal atoms, which are then mass analyzed. The abundance and identities of the transition metal atoms are traced back to the Ab staining reagents. b. The microengraving technique utilizes a microchip with many thousands of microwells, into which between 0 and a few cells of interest are loaded. An Ab coated coverslide is placed over the microchip to capture specific secreted proteins. Microchip addresses are correlated with regions on the coverslide and with microscopy images to associate a given cell with a given secretion profile. Captured proteins are detected using fluorescent secondary Abs, with different proteins identified using different fluorophores. The coverslide can be replaced during the time-course of an experiment to capture single cell secretion kinetics. Cells of interest may be removed for further analysis. c. Single cell barcode chips (SCBCs) contain up to a few thousand microchambers, into which between 0 and a few cells are loaded. An Ab-barcoded glass slide is patterned so that each microchamber contains a complete, miniaturized Ab array onto which secreted, or following cell lysis, cytoplasmic or membrane proteins are captured. Protein assays are developed using fluorescently-labeled secondary Abs, with different proteins identified according to the spatial location of the immunoassay within the barcode. If cells are not lysed (only secreted proteins detected), then the cells remain viable and may be further investigated. d. Single cell Westerns are miniaturized variants of traditional Western Blotting methods, with ∼10³ single cells analyzed per microchip.

Figure 3. Single cell analysis traces the lineage of a colon cancer

The work flow proceeds from left. A biopsy of a healthy colon is analyzed using FACS to separate cells extracted from the crypt-like structures of the colon epithelium. The bottom regions of the crypts are enriched in stem cell-like populations, with those cells identified as EpCAM⁺/CD44⁺. More differentiated enterocyte and goblet cells are found near the top of the crypts, and are defined as EpCAM⁺/CD44^-/CD66a^high. Single cell, multiplex transcriptomics is used to develop a 53 gene expression classifier. Principal component analysis (PCA) of the single cell data resolves the major cellular subpopulations. The genes that define those subpopulations are plotted with respect to how they are represented within the two dominant principal components. The plot reveals how the classifier resolves immature progenitors (top left of graph), enterocyte-like cells (top right), and goblet-like cells (bottom left). Classifers of these populations, also identified from hierarchical clustering of the single cells transcriptome data, provide the color coding for each mRNA on the plot. Once established, the classifier was used to analyze cells collected from a patient colon cancer tumor, and to show that the tumor cells (drawn with a red border) are largely goblet-like and immature progenitors. A single immature progenitor tumor cell is sorted from the patient tumor using FACS, and implanted into a mouse model to grow a monoclonal tumor. Analysis of that tumor reveals a cellular composition reminiscient of the original patient tumor, implying that the tumor cellular heterogeneity can originate from expansion and lineage differentiation of a single progenitor-like cell. The principal component plot is adapted from reference .

Figure 3. Single cell analysis traces the lineage of a colon cancer

The work flow proceeds from left. A biopsy of a healthy colon is analyzed using FACS to separate cells extracted from the crypt-like structures of the colon epithelium. The bottom regions of the crypts are enriched in stem cell-like populations, with those cells identified as EpCAM⁺/CD44⁺. More differentiated enterocyte and goblet cells are found near the top of the crypts, and are defined as EpCAM⁺/CD44^-/CD66a^high. Single cell, multiplex transcriptomics is used to develop a 53 gene expression classifier. Principal component analysis (PCA) of the single cell data resolves the major cellular subpopulations. The genes that define those subpopulations are plotted with respect to how they are represented within the two dominant principal components. The plot reveals how the classifier resolves immature progenitors (top left of graph), enterocyte-like cells (top right), and goblet-like cells (bottom left). Classifers of these populations, also identified from hierarchical clustering of the single cells transcriptome data, provide the color coding for each mRNA on the plot. Once established, the classifier was used to analyze cells collected from a patient colon cancer tumor, and to show that the tumor cells (drawn with a red border) are largely goblet-like and immature progenitors. A single immature progenitor tumor cell is sorted from the patient tumor using FACS, and implanted into a mouse model to grow a monoclonal tumor. Analysis of that tumor reveals a cellular composition reminiscient of the original patient tumor, implying that the tumor cellular heterogeneity can originate from expansion and lineage differentiation of a single progenitor-like cell. The principal component plot is adapted from reference .