Single-cell genomics: coming of age.

Abstract

Single-cell genomics is the study of the individuality of cells using omics approaches. Although young, the field has now entered its teenage years and is beginning to show clear signs of maturity. Its origins can be traced back to pioneering experiments that allowed the detection of gene expression in single cells by microarrays (reviewed in [1]). However, it was with the emergence of “next-generation” DNA sequencing that single-cell genomics really took off [2–4]. Although initial experiments were modest in size and resulted in noisy and incomplete data, they immediately revealed the great potential for biological discoveries. It soon became clear that the substantial technical and biological variability required data from many single cells in order to allow meaningful data mining and interpretation of the data [5]. Thus, the following years were spent pursuing a few lines of development: improvements in the accuracy and scope of single-cell methods and increasing throughput and reducing cost. Today, we are in a position to routinely measure gene expression in tens of thousands of single cells with high accuracy in terms of quantification of gene expression (albeit sensitivity in terms of detection of mRNAs varies significantly depending on protocol and sequencing depth). The costs are at least manageable and continue to decrease. While single-cell RNA-seq is now mature and almost routine, technological development has shifted to other modalities: DNA, protein, chromatin modifications, and more. Single-cell whole-genome DNA sequencing is challenging because loss of material causes dropouts in the sequence and because sequencing errors are difficult to distinguish from real mutations. Despite these challenges, single human cortical neurons have been used to reconstruct lineages based on somatic mutations that had accumulated during development [6]. Similarly, clonal evolution within solid tumors can be revealed by detecting somatic copy number variations in single cells (reviewed in [7]). Another trend is the extension of single-cell analysis to measure epigenetic states such as DNA accessibility [8–10], methylation [11], and chromosome conformation [12]. Generally these methods pose similar challenges to DNA sequencing but offer access to pure cellular epigenetic states that are simply inaccessible by bulk methods. Single-cell protein analysis occupies a different niche, where smaller numbers of proteins can be analyzed but in very large numbers of cells, classically using fluorescence-activated cell sorting (FACS) for up to eight targets but more recently with mass cytometry targeting up to hundreds of proteins [13]. A limiting factor for protein analysis remains the requirement for high-quality affinity reagents such as antibodies. Finally, a recent development (but see [14]) is the combination of methods to simultaneously measure two or more modalities in single cells. For example, genome and transcriptome [15, 16], transcriptome and methylome [17, 18], and RNA and protein [19]. In the near future, such experiments will be able to link the phenotypes of single cells evolving in tumors to their genotypes. Due to the speed with which single-cell genomics technologies are evolving, computational analysis methods are racing to keep up. Statistical and computational methods are at the heart of single-cell genomics and are critical to extracting meaningful information and biology from the data. Much work has focused on transcriptomic data analysis (e.g., reviewed in [20]) and in this special issue of Genome Biology there are examples of areas that benefit from bespoke computational approaches at the levels of both cells and genes. In terms of individual genes, a method to define significant differences in the cell-to-cell variation in gene expression (as opposed to mean expression levels) is reported [21] and one paper addresses expression states of long noncoding RNAs [22]. In terms of cell-to-cell variation at the DNA level, there is clearly tremendous scope for computational method innovation in the area of tumor heterogeneity, addressed by Beerenwinkel and colleagues [23], and Markowetz and Ross [24] in this issue.