Advances and Applications of Single Cell Sequencing Technologies

Abstract

Single cell sequencing (SCS) has emerged as a powerful new set of technologies for studying rare cells and delineating complex populations. Over the past 5 years, SCS methods for DNA and RNA have had a broad impact on many diverse fields of biology, including microbiology, neurobiology, development, tissue mosaicism, immunology and cancer research. In this review, we will discuss SCS technologies and applications, as well as translational applications in the clinic.

Introduction

The fundamental unit of an organism is a single cell. Homo sapiens are composed of approximately 3.72 × 10¹³ single cells that live harmoniously in tissues among their neighbors (Bianconi et al., 2013). However in diseases such as cancer, the greed and avarice of a single cell can lead to the downfall of the entire organism. Despite the complexity of tissues, most genomic studies to date have focused on analyzing bulk tissue samples, which are composed of millions of cells. In these averaged datasets, it is difficult to resolve cell-to-cell variations and identify rare cells that may play an important role in disease progression. The recent development of single-cell sequencing (SCS) methods has led to a paradigm shift in the field of genomics, away from bulk tissue analysis, and towards detailed and comprehensive studies of individual cells.

Our fascination with single cells dates back to the invention of the first microscopes in the 1660’s, which allowed early microscopists to observed single prokaryotic cells moving around in droplets of water. Subsequent work by early pathologists, such as Rudolf Virchow, in the late 1850’s established the link between abnormalities in single cells and human diseases. In the late 1900’s the development of cell staining techniques and cytological methods galvanized the field, enabling scientists to directly visualize genetic differences on chromosomes in single cells. However, many cytogenetic and immunostaining methods were limited to measuring targeted genes and proteins. In the 1990’s quantitative microarray technologies were developed for measuring genome-wide DNA and RNA information, but required too much input material for single cell analysis. Although PCR methods had been developed, they were only capable of amplifying small targeted regions of the genome. To overcome this limitation whole-transcriptome-amplification (WTA) (Van Gelder et al., 1990) and whole-genome-amplification (WGA) (Dean et al., 2002; Telenius et al., 1992) methods were developed to amplify genome-wide DNA and RNA. Another important milestone occurred in 2005 with the development of the first next-generation sequencing (NGS) technologies, which enabled genome-wide sequencing of DNA and RNA (Mardis, 2011).

The culmination of these technologies led to the invention of the first genomewide single-cell DNA (Navin et al., 2011) and RNA (Tang et al., 2009) sequencing methods for mammalian cells. These initial studies (and work by other groups) led to the establishment of a new field of biology: single cell sequencing (SCS). The field has shown tremendous growth over the last 5 years (Figure 1A) and impacted many diverse areas of biological research (Figure 1B–C, Supplementary Table 1). In this review, we will discuss the advances and limitations of SCS technologies, and the myriad of applications that they have had in biological research and medicine.

Figure 1. Timeline of Milestones in Single Cell Sequencing.

(A) Timeline of SCS Milestones (B) Histogram of the growing number of publications in SCS over the past 5 years (C) Prevalence of publications categorized by fields.

Single Cell Isolation Methods

In order to sequence a single cell, it must first be captured. While the methods for isolating single cells from abundant populations have been well-established, the isolation of rare single cells (<1%) remains a formidable technical challenge. To isolate a single cell randomly from an abundant population, several approaches can be employed: mouth pipetting, serial dilution, robotic micromanipulation, flow-assisted cell sorting (FACS) and microfluidic platforms (Table 1). Many of these approaches require cells or nuclei in suspension, and therefore cannot preserve their spatial context in tissues. This limitation can be overcome using Laser-capture-microdissection (LCM), which can also be used to isolate rare cells. In contrast, the isolation of rare single cells (< 1%) is far more challenging. Many commercial platforms have been developed for isolating circulating tumor cells (CTCs), which occur at very low frequencies (1 in 1 million) in the blood of cancer patients (Cristofanilli et al., 2004). The CellSearch system is an FDA approved clinical system that uses magnets with ferrofluid nanoparticles conjugated to antibodies for EpCAM and CD45 to isolate CTCs (Yu et al., 2011). Nagrath et al. developed another method that uses a nanopost microchip technology with EpCAM antibodies (Nagrath et al., 2007). The Magsweeper (Illumina Inc.) is a technology that involves dipping a rotating magnet with EpCAM antibodies to isolate CTCs (Powell et al., 2012). Other methods are more generally applicable to rare cell populations. The DEP-Array system (Silicon Biosciences) uses a microchip with dielectropheretic cages to navigate individual cells by charge (Altomare et al., 2003). The CellCelector (Automated Lab Solutions) uses robotic micromanipulation capillary system to identify single cells for isolation (Choi et al., 2010). An alternative approach uses nanofilters to isolate rare cells by size exclusion rather than surface markers (Adams et al., 2014). The advantages and limitations of these methods are summarized in Table 1 and have been reviewed in detail in other review articles (Navin and Hicks, 2011; Navin, 2014; Shapiro et al., 2013).

Table 1.

Single-Cell Isolation Methods for Abundant and Rare Populations

Isolation Methods for Abundant Cells
Isolation Methods	Description	Advantages	Disadvantages	Cost
Serial dilution	serial dilution to about one cell per microliter	simple approach; low cost	high probability of isolating multiple cells	$
Mouth pipetting	isolate single cells with glass pipettes	simple approach; low cost	technically challenging	$
Flow sorting	microdroplets with single cells are isolated by electric charge at high pressure	high-throughput; fluorescent markers can be used to isolate subpopulations	expensive equipment; requires operator	$$
Robotic micromanipulation	robotic-controlled micropipettes isolate single cells	high accuracy; fluorescence can be used	low throughput	$$$
Microfluid platforms	microfluidic chips isolate single cells in flow channels	high-throughput; reactions can be performed on-chip; reduced reagent costs	cell size must be uniform; expensive consumables	$$$
Isolation Methods for Rare Cells
Isolation Methods	Description	Advantages	Disadvantages	Cost
Nanofilters	size discrimination on nanofabricated filters	cells are selected by size exclusion	cells can adhere to filters during backwash	$
MagSweeper	rotating magnet with EpCAM antibodies	high enrichment of rare cells	biased toward markers used for isolation	$$
Laser-capture microdissection	cells are cut from a tissue section slide with lasers under a microscope	spatial context is preserved	cell slicing; UV damage to DNA/RNA	$$$
CellSearch	magnets with nanoparticles conjugated to antibodies enrich surface markers	high throughput	biased toward markers used for isolation	$$$
CellCelector	robotic capillary micromanipulator	high-throughput	expensive system and large footprint	$$$
DEP-Array	microchip with dielectropheretic cages	high sensitivity for isolating rare cells	time-consuming; low-throughput; cells are deposited into large final volumes	$$$$

This table summarizes the advantages and disadvantages of single-cell isolation methods from abundant populations and rare populations.

Single Cell DNA Sequencing Methods

The development of DNA SCS methods has proven to be more challenging than RNA. This is due to the fact that a single cell contains only 2 copies of each DNA molecule, but thousands of copies of most RNA molecules. The limited amount of input material for WGA results in a number of technical errors, including: coverage non-uniformity, allelic dropout (ADO) events, false-positive (FP) errors and false-negative (FN) errors (Table 2). The most salient technical errors occur during the initial rounds of amplification and are subsequently propagated by all daughter molecules. FP errors accumulate at random sites due to the infidelity of the WGA polymerase and lead to single nucleotide errors (Dean et al., 2002; Lasken, 2007). However, by far the greatest source of WGA error comes from allelic dropout events at 10–50% of the mutation sites (Fiegler et al., 2007; Hou et al., 2012; Lasken, 2007; Talseth-Palmer et al., 2008; Zong et al., 2012).

Table 2.

Technical Errors Associated with Single-Cell Sequencing

	Technical Artifact	Amplification Method	Error Type	Description
WGA	chimeric molecules	MDA	false-positive inversions	3′ and 5′ ends of newly synthesized molecules hybridize together during MDA leading to inversions
	coverage nonuniformity	MDA, DOP-PCR, MALBAC	copy number aberrations, false-negative SNVs	Under and over amplifications of the genome can lead to erroneous copy number abberations and false-negative SNVs
	FP amplification error	MDA, DOP-PCR, MALBAC	SNV, indel	DNA polymerase introduces random FP errors
	allelic dropout	MDA, DOP-PCR, MALBAC	False-negative errors	Heterozygous (AB) variants undergo dropout during WGA leading to homozygous (AA or BB) genotypes
	pileup regions	DOP-PCR	copy number amplifications	Massive over-amplifications of focal genomic regions occur during DOP-PCR
WTA	amplification distortion	dt-anchor, Template-Switching	erroneous expression values	over/under amplification during WTA leads to erroneous expression values
	transcript dropout	dt-anchor, Template-Switching, UMI	false-negative unexpressed genes	failure to amplify a transcript during WTA
	3′ bias	dt-anchors	failure of RT polymerase to fully synthesize the first cDNA strand	strong bias toward amplification of 3″ end of RNA transcripts

This table lists the common technical errors that arise during WGA and WTA in single-cell sequencing experiments.

Importantly, WGA is not a single technique, but encompasses a wide variety of experimental methods. The most common WGA methods are degenerative-oligonucleotide-PCR (DOP-PCR) and multiple-displacement-amplification (MDA) (Figure 2A–B). DOP-PCR generates low physical coverage (~10%) of a single cell genome, but accurately retains copy number levels during WGA. In the first SCS method developed for genomic DNA, DOP-PCR was combined with flow-sorting of nuclei and NGS to generate high-resolution copy number profiles from single mammalian cells (Baslan et al., 2012; Navin et al., 2011). However, the low physical coverage of DOP-PCR makes it a poor approach for measuring mutations at base-pair resolution. MDA using either the Phi29 or Bst polymerases has been widely reported to achieve high physical coverage (>90%) from a single cell genome or exome (Hou et al., 2012; Xu et al., 2012; Yong Wang, 2014; Zong et al., 2012) (Figure 2B). However, MDA generates non-uniform coverage and causes distortions in read depth making it a poor method to measure DNA copy number (Navin, 2014). Phi29 is the ideal polymerase for MDA reactions, since it has a very low FP error rate (1e-7) compared to Bst (1e-5), which does not have proofreading activity (Dean et al., 2002; Lasken, 2013). Another DNA SCS method is multipleannealing- and-looping-based-amplification-cycles (MALBAC), which uses the Bst polymerase to form circular DNA fragments followed by adapter ligation PCR (Zong et al., 2012). This method can obtain both copy number information and single nucleotide variants (SNVs), but generates extremely high FP error rates, making it more suitable for copy number profiling. Another method, called NUC-SEQ or single nucleus exome sequencing (SNES) takes advantage of G2/M nuclei which have duplicated the amount of genomic DNA in a single cell (12 picograms) prior to MDA, which reduces many technical error rates during single cell sequencing of exomes and genomes (Yong Wang, 2014)(Leung et al. 2015).

Figure 2. WGA and WTA Methods for Single Cell Sequencing.

(A–B) Whole-genome-amplification methods. (C-D) Whole-transcriptome-amplification methods. (A) Degenerative-Oligonucleotide-Primer PCR (B) Multiple-displacementamplification. (C) oligo dT-Anchor Approach (D) Template switching

After WGA the amplified DNA is used to construct libraries for NGS. While several sequencing platforms are available, Illumina has emerged as the primary tools in most studies due to low cost per base, high-throughput and low error rates. To further save costs and increase throughput, single cell libraries are often barcoded and pooled together for multiplexed sequencing. In many studies, the barcoded libraries are used for targeted capture (exome or gene panels) to sequence only regions of interest and obtain higher coverage depth in these areas. For a more comprehensive review of single cell DNA sequencing methods please refer to the following articles (Blainey and Quake, 2014; de Bourcy et al., 2014; Navin and Hicks, 2011; Navin, 2014; Van Loo and Voet, 2014).

Single Cell RNA Sequencing Methods

The development of single cell RNA sequencing methods has shown considerable progress over the past 5 years. To sequence a single cell transcriptome, the RNA must first be amplified by whole-transcriptome-amplification (WTA). This step is necessary because a typical mammalian cell contains only 10 picograms of total RNA and 0.1 picograms of mRNA. Initial WTA methods utilized the T7 RNA polymerase to perform linear amplification of cDNA by in vitro transcription (IVT) (Van Gelder et al., 1990). These methods were further developed using oligo d(T) primers conjugated to adapter sequences for reverse transcription and selective amplification of polyadenylated mRNA by PCR (Tang et al., 2009) (Figure 2C). However, these WTA methods were plagued by strong 3’ mRNA bias. To mitigate this bias, a WTA method called SMART-Seq was developed that amplifies only full-length mRNA transcripts using an Moloney Murine Leukemia Virus (MMLV) reverse transcriptase (Ramskold et al., 2012). MMLV has both template-switching and terminal transferase activity, which results in the addition of nontemplated cytosine residues to the 5’ end of the cDNA. By adding a poly-G template with an adapter sequence, MMLV can switch templates and transcribe the other strand, leading to full-length cDNA transcripts that are amplified by PCR (Figure 2D). Another technical artifact of single-cell RNA sequencing is amplification bias, in which mRNAs levels are distorted during WTA. To reduce these errors, a recent method developed unique molecular indexes (UMIs) to label the original pool of RNA molecules prior to WTA (Islam et al., 2014). After WTA, the resulting cDNA libraries are barcoded and pooled for multiplexed NGS. For a more detailed discussion of RNA SCS methods please refer to the following review articles (Macaulay and Voet, 2014; Saliba et al., 2014; Sandberg, 2014).

Single Cell Epigenomic Sequencing Methods

Epigenomic profiling of single cells remains to be one of the greatest technical challenges in the field. The problem is that standard epigenomic sequencing methods require a pool of DNA that is split into two separate fractions for treatment with bisulfide or methylation restriction enzymes prior to sequencing. The other technical barrier is that epigenetic DNA modifications cannot be amplified with DNA polymerases. Despite these technical hurdles, two recent studies have made initial progress. The Hi-C approach was recently adapted for single cell profiling at megabase resolution to identify physical chromatin interactions in single cells (Nagano et al., 2013). In another study, reduced single cell representation bisulfite sequencing (scRRBS) was developed to measure cytosine methylation modifications at 1.5 million CpG sites in a single cell, which is equivalent to about 10% of the genome (Guo et al., 2013). While these studies are clearly pioneering, they were also challenged by limited coverage (2.5% and 10%), low resolution and many technical errors. However, future refinement of these technologies are likely to lead to accurate whole-genome epigenomic profiling of single cells.

Distinguishing Technical Errors from Biological Variations

At the current state of technology, most SCS methods introduce extensive technical errors and variability into datasets. Unfortunately, naïve users often interpret these technical errors as extensive biological heterogeneity. To eliminate these errors and confirm that a mutation or transcript is truly heterogeneous in a population of cells, orthogonal validation using a targeted approach is critical. To validate heterogeneous DNA variants or mutations, targeted validation can be performed using deep-sequencing with molecular barcodes (Kennedy et al., 2014; Schmitt et al., 2012; Wang et al., 2014) or digital droplet PCR (Bio-Rad, Raindance Inc.). To validate heterogeneous RNA expression changes, targeted validation can be performed with single cell RT-qPCR or with digital droplet PCR. Unfortunately, many published studies to date have incorrectly reported extensive biological heterogeneity without orthogonal validation, which is more likely to be explained by technical errors.

In summary, there has been tremendous progress in the development of single cell DNA and RNA sequencing methods. However, all SCS methods generate technical errors during WGA and WTA and thus orthogonal validation using a targeted approach is critical at the current state of technology. We now turn to a detailed discussion of the many broad fields of biology that have been impacted by SCS methods over the last 5 years (Figure 3).

Figure 3. Broad Applications of SCS in Biological and Biomedical Research.

Panels illustrating the diverse fields of biology that have been impacted by SCS technologies over the past 5 years. Image credits: neurobiology, Zeynep Saygin (Cell Picture Show); germline transmission, Wang and Navin; organogenesis, Mikael Häggström (Wikimedia Commons); cancer biology, NIH; clinical diagnostics, Wang and Navin; immunology, Olivier Schwartz and the Electron Microscopy Core Facility, Institut Pasteur (Cell Picture Show); microbiology, NIAID; tissue mosaicism, Wang and Navin; embryology, Seth Ruffins, Russell Jacobs, and the Caltech MRI Atlas of Mouse Development (Cell Picture Show); prenatal genetic diagnosis, Shutterstock. All images used with permission.

Microbiology

A formidable challenge in microbiology has been that over 99% of microbial species on planet earth cannot be cultured and expanded in the lab (Ishoey et al., 2008; Lasken and McLean, 2014). Single-cell DNA and RNA sequencing methods provide a powerful new approach to resolve microbial genomes and delineate cell-to-cell diversity within diverse populations. However, bacteria and other microorganisms often have only femtograms of DNA and RNA, making it even more challenging to amplify than mammalian cells (Lasken, 2007). In an early study, MDA was used to amplify DNA from the marine cyanobacterium Prochlorococcus for pyrosequencing and de novo assembly (Rodrigue et al., 2009). In another study Woyke et al. used FACS and MDA to perform NGS and assemble two marine flavobacteria genomes to 90% coverage (Woyke et al., 2009). Blainey et al. also used MDA to sequence and assemble the genomes of 5 single cells from an Ammonia-oxidizing archaea (Blainey et al., 2011). Another study performed SCS of five segmented filamentous bacterial cells to gain insight into their life cycles (Pamp et al., 2012). While initial studies were often limited to sequencing just a few microbial cells, a subsequent large-scale study was conducted on 201 uncultivated archaeal and bacterial cells from nine diverse habitats. In this study SCS revealed 29 uncharted branches of the tree of life, revealing ‘microbial dark matter’ and challenging the canonical three domains of life (Rinke et al., 2013). These studies show that SCS is complimentary to metagenomic deep-sequencing methods, and can open up new avenues of investigation into microbial genomes that cannot be cultured in the lab.

Neurobiology

Neurons represent one of the most morphologically diverse populations of cells. Traditional classification has relied mainly on morphological features (De Carlos and Borrell, 2007), however single-cell RNA sequencing provides a powerful unbiased approach to classify neurons based on their transcriptional profiles. In a study by Qiu et al. single neuron RNA-sequencing was combined with electrophysiology to obtain transcriptional profiles from embryonic mouse hippocampus and neocortical neurons (Qiu et al., 2012). In another study, single cell RNA-seq was performed in situ in spatially defined neuronal regions, which identified cell-to-cell transcriptional variation in hippocampal neurons (Lovatt et al., 2014). Pollen et al. used low-coverage single cell RNA sequencing and microfluidics to analyze single cells from 11 brain populations, and identified Notch signaling as an important regulator of brain development (Pollen et al., 2014). In another study, Usoskin et al. used single-cell RNA sequencing to profile 622 sensory neurons in mice, which revealed 11 novel expression classes of sensory neuron cell types (Usoskin et al., 2014).

Several studies have also begun to investigate DNA heterogeneity in neurons. SCS was recently used to study LINE-1 retrotransposition in the cerebral cortex (Evrony et al., 2012) and found that each cortex neuron had an average of 0.6 somatic insertions events. In another study, SNS (Baslan et al., 2012; Navin et al., 2011) was used to identify copy number variants (CNVs) in neurons from three normal and two pathological brain samples (Cai et al., 2014). The authors reported that large (>1mb) clonal CNVs arose in about 5% of neurons during normal development. In contrast, another study used SNS to profile neuronal copy number diversity in the prefrontal cortex of postmortem brains, which identified many de novo CNVs in neurons that were not clonal between different single cells (McConnell et al., 2013). In another study, SCS using microwells identified copy number changes in a normal postmortem brain and a patient with Down syndrome (Gole et al., 2013). These initial studies show that SCS provides a novel approach to classify neuronal cell types and identify an unexpected amount of DNA diversity in neuronal populations.

Tissue Mosaicism

The traditional view of somatic tissues is that single normal cells have identical genomes, however this dogma is beginning to be challenged by increasing evidence of genetic mosaicism in normal tissues that arises during normal development (Biesecker and Spinner, 2013). To date most studies have analyzed bulk tissue samples, and therefore much controversy exists over the prevalence of mosaic mutations and whether they can simply be explained by technical error. SCS methods provide a novel approach to resolve cell-to-cell variations in normal tissues at an unprecedented genomic resolution. SNS was recently used to identify de novo CNVs in 13–41% of the neurons in the frontal cortex of postmortem brains, suggesting that CNV mosaic events are common in cortical neurons (McConnell et al., 2013). This unexpected amount of copy number diversity has previously not been appreciated in the brain. However, a recent study using the same SCS methods (SNS) challenged these data, by suggesting that somatic CNVs are extremely rare in neurons and other normal tissues (Knouse et al., 2014). In this study 96 single neurons were sequenced from mice and only a single somatic CNV was identified in one neuron. The authors also examined 89 single cells from 4 human patients frontal lobes, and found only 2 cells with aneuploid rearrangements (2.2%). In skin, the authors detected aneuploidy in only 2.7% of mouse keratinocytes and none in human cells. In liver cells they profiled 100 hepatocytes and found only 4% aneuploid cells. Thus, while both studies showed that copy number mosaicism is likely to exist in normal tissues, there is much debate regarding the prevalence of these rearrangements, and whether they might play an important role in human diseases.

Germline Transmission

Sperm cells and oocytes are single cells that fuse to form a zygote and transmit genomic material and evolution has engineered this process to generate genetic variation. Single-cell DNA sequencing provides a novel approach to study the mechanisms that generate germline variation. In one of the first studies on this topic, single sperm cells were sequenced, which revealed an average of 22.8 recombination events, 5–15 gene conversion events and 25–36 de novo mutations in each sperm cell (Wang et al., 2012). The authors also calculated copy number profiles, which showed that 7% of the single sperm cells had aneuploid genomes. Consistent with this study, another group used low-coverage whole-genome sequencing to delineate haplotypes in single sperm cells from one individual, which revealed an average of 25.3 recombination events per cell (Kirkness et al., 2013). In another study, Lu et al. applied MALBAC to sequence single sperm cells from an Asian individual, in which they reported aneuploidy in 4% of the cells and 26 recombination events per single sperm cell (Lu et al., 2012). While most germline studies have focused on sperm, a recent study used MALBAC to analyze fertilized oocytes (Hou et al., 2013). In this study oocytes from 8 individual females were analyzed, which identified 43 cross-over events per oocyte, a recombination rate that is 1.63X higher than sperm. Interestingly, this study also reported a much higher rate of aneuploidy in oocytes (17.6%) compared to sperm (4–7%). Taken together, these studies have confirmed previous recombination rates and revealed a striking amount of genomic diversity that arises in germ cells during the transmission of genetic material to offspring.

Embryogenesis

Extensive transcriptional regulation and epigenetic reprogramming occurs during the earliest stages of embryonic development, as the zygote forms the three major cell lineages (endoderm, ectoderm and mesoderm). The genomic regulation of these early events and maintenance of pluripotency has been challenging to study due to the limited amount of input material. To address this problem, RNA SCS was used to analyze transcriptional reprograming in vitro during the transition from the inner cell mass of blastocysts to pluripotent embryonic stem cells (Tang et al., 2010). In another study, RNA SCS was used to profile single cells from human pre-implantation embryos and embryonic stem cells which detected over 1000 heterogeneous transcripts within the same blastomere (Yan et al., 2013). In another study, RNA SCS was used to study transcriptome dynamics from oocyte to morula development in human and mouse embryos, which delineated a step-wise progression of pathways that regulate cell cycle, gene regulation, translation and metabolism (Xue et al., 2013). Another study used single cell bisulfite sequencing to measure cytosine DNA modifications in mouse embryonic stem cells, which showed massive global demethylation during embryonic development (Guo et al., 2013). Collectively, these studies have begun to dissect the complex transcriptional regulation and epigenomic reprogramming that occurs during the earliest stages of embryogenesis.

Organogenesis

In most tissues, traditional classification of cell types has previously been limited to a few dozen markers that have been used for decades. RNA SCS methods provide a powerful new unbiased approach to perform transcriptional profiling and identify groups of cells that share common expression programs, representing distinct cell types. In the first study to apply this approach RNA SCS was used to analyze lung epithelium development (Treutlein et al., 2014). From these data, the development of lung progenitor cells were traced as they formed alveolar air sacs that regulate gas exchange. In this study the authors identified hundreds of novel markers for distinguishing the four major cell types and used them to reconstruct cell lineages during alveolar sac differentiation. In another study, RNA SCS was used to analyze gene expression patterns of single cells during kidney development in mice at E11.5, E12.5 and P4 (Brunskill et al., 2014). These data revealed a multi-lineage priming model in which many genes and pathways were repressed during nephrogenesis, rather than being activated from a ‘blank slate’. These initial studies demonstrate the utility of applying unbiased RNA SCS methods to classify cell types and identify novel markers of cell lineages during organ development.

Immunology

The immune system is broadly classified into the adaptive and innate components, which comprise a large variety of cell types that work together in a concerted fashion to recognize and clear antigens. Although the major immune cell types have been known for decades, there is little known about the transcriptional heterogeneity within cell types in responses to antigens. In one study, RNA SCS was used to analyze mouse bone-marrow derived dendritic cells that were stimulated under different conditions in vitro and found that individual cells show variable responses that are mediated by interferon paracrine signaling (Shalek et al., 2014). In another study, RNA SCS was used to identify bimodal gene expression patterns in bone-marrow-derived dendritic cells stimulated by lipopolysaccharide that was modulated through an interferon feedback circuit (Shalek et al., 2013). Another study performed unbiased RNA SCS to profile 4000 single cells from mouse spleen in response to antigen activation with LPS which revealed seven classes of immune cells and identified 1575 variable gene responses after antigen activation (Jaitin et al., 2014). These studies show that unbiased RNA SCS methods can be used to investigate heterogeneous transcriptional responses in immune cells after antigen activation.

Cancer Research

Tumors evolve from single normal cells. During this process the cancer cells accumulate mutations and diversify to form distinct lineages and subpopulations. This intratumor heterogeneity confounds the clinical diagnosis and therapeutic treatment of patients. Clonal diversity is likely to play a key role in tumor progression during processes such as invasion, clonal evolution and metastasis by providing fuel for evolution to select upon. Genomic diversity also enables tumor cell populations to survive selective pressures in the tumor microenvironment, including hypoxia, chemotherapy, immune surveillance and geographic barriers. However, to date studying clonal diversity has been difficult in bulk populations of tumor cells using standard sequencing methods. DNA and RNA SCS methods provide powerful new tools for delineating clonal diversity and understanding the role of rare cells during cancer progression.

To date, most SCS studies of cancer have focused on intratumor heterogeneity and clonal evolution in primary tumors. The first study used SNS to investigate aneuploidy evolution in single cells from patients with triple-negative (ER-/PR-/HER2-) breast cancers (Navin et al., 2011). These data revealed that copy number aberrations evolved in punctuated bursts of evolution, followed by stable clonal expansions to form the tumor mass. In another study, single-cell exome sequencing (NUC-SEQ) showed that point mutations evolved gradually over time generating extensive clonal diversity and many rare (<1%) mutations in the tumor mass (Wang et al., 2014). Single-cell exome sequencing has also been applied to study clonal diversity in renal carcinoma (Xu et al., 2012) and a JAK2-positive myeloproliferative neoplasm (Hou et al., 2012), which identified a monoclonal population of cells that shared a common genetic lineage. Similarly, single cell exome sequencing was applied to study a muscle-invasive bladder cancer (Li et al., 2012), and a colon cancer patient (Yu et al., 2014), which identified two distinct subpopulations of cells in each of which tumor diverged, but also shared a common set of founder mutations. Another study used DNA SCS to delineate clonal diversity in glioblastoma, which revealed convergent evolution of EGFR mutations in different subclones from the same primary tumors (Francis et al., 2014).

DNA SCS has also been used to study clonal evolution in hematopoietic cancers. In one study, single cancer cells were sequenced from three patients diagnosed with MDS-derived secondary AML to reconstruct mutational chronology (Hughes et al., 2014). In another study 1,479 single cells were sequenced from six acute lymphoblastic leukemia (ALL) patients using targeted panels, which identified the presence of multiple clonal subpopulations in many AML patients. Clonal dynamics have also been investigated in xenografts, which showed extensive selection in the first transplantation passages, followed by clonal dominance (Eirew et al., 2014). Collectively, these studies provide strong evidence for clonal evolution in many human tumors (Campbell and Polyak, 2007; Greaves and Maley, 2012; Navin and Hicks, 2010) by showing that single cells can continue to acquire new mutations and evolve to form the primary tumor mass.

Recent work has begun to use SCS to investigate metastatic dissemination and circulating tumor cells (CTCs) in the blood. In one study, RNA SCS was used to profile CTCs in the blood of melanoma patients (Ramskold et al., 2012). In another study, DNA SCS was used to analyze CTCs from six patients with metastatic colon cancer, showing that many of the driver mutations in the primary tumor could be detected in the CTCs (Heitzer et al., 2013). In another study, MALBAC was used to perform exome sequencing and copy number profiling of single CTCs from 7 metastatic lung adenocarcinoma cancer patients (Ni et al., 2013). In another study, single-cell exome sequencing was applied to a patient with metastatic prostate cancer, which identified 51% of the mutations in the primary and metastatic tumors in the CTC populations (Lohr et al., 2014). RNA SCS was also recently used to investigate CTC clusters in metastatic seeding in breast cancer (Aceto et al., 2014). Another study applied SNS and morphometric imaging to investigate copy number evolution in response to Abiraterone therapy in metastatic prostate cancer (Dago et al., 2014). Collectively, these studies have improved our understanding of CTCs and metastasis in human cancers.

RNA SCS has also been used to study cell plasticity and cancer stem cells in human tumors. An unbiased study of hundreds of single cell transcriptomes in five glioblastoma patients showed that cancer cells displayed a large range of intermediate phenotypes, that do not fall into distinct classes of epithelial or mesenchymal cell types. In summary, SCS methods have already had a large impact on improving our fundamental understanding of intratumor heterogeneity, clonal evolution and metastatic dissemination in human cancers. For a more detailed review on SCS applications in cancer research, please refer to the following review articles (Navin, 2014; Van Loo and Voet, 2014).

CLINICAL APPLICATIONS

SCS methods have direct translational applications in cancer treatment and pre-natal genetic diagnosis (PGD). In cancer research, intratumor heterogeneity presents a major challenge for clinical diagnostics, because single samples may not represent the tumor as a whole. While regional sequencing and deep-sequencing can resolve some clonal substructure, they cannot fully delineate the clonal substructure of a tumor and are inherently unable to determine which combination of mutations occur in each clone. SCS provides a powerful tool for resolving intratumor heterogeneity, and guiding targeted therapy towards the most malignant clones. SCS can also be used to calculate a diversity index for each cancer patient, which may have prognostic utility for predicting poor survival and poor response to chemotherapy. SCS technologies will also have direct applications for non-invasive monitoring, by sequencing single CTCs in the blood to track mutations in the primary and metastatic tumors. Several studies have already shown that over 50% of the mutations in the primary and metastatic tumors can be detected in CTCs of lung cancer (Ni et al., 2013), prostate cancer (Lohr et al., 2014) and colon cancer patients (Heitzer et al., 2013). By sequencing CTCs at multiple time-points over the course of therapy, oncologists can track mutational evolution and make rapid changes to their therapeutic strategies before resistance emerges. SCS methods will also have clinical applications in the early detection of tumor cells in bodily fluids (urine, sputum, blood) and fine-needle-aspirates samples.

Another major area of clinical utility is pre-implantation genetic diagnosis (PGD) and in vitro fertilization (IVF). During this procedure a biopsy of a single cell is collected from a set of blastomeres for DNA SCS to screen for genetic disorders prior to implantation into the uterus. In the past, these methods have traditionally been limited to cytogenetic analysis and single-cell PCR. SCS provides the advantage of being able to profile thousands of mutations and copy number changes associated with diseases that can be screened from one cell using a single assay. In a proof-of-concept study, SCS was used to profile genomic copy number and structural variants in single cells from blastomeres derived from a human zygote after IVF (Voet et al., 2013). In another study, MALBAC was used to sequence polar bodies to identify copy number changes and point mutations prior to implantation (Hou et al., 2013). These preliminary PGD studies demonstrate the technical feasibility of screening oocytes and blastomeres to avoid the genetic transmission of diseases, paving the way for future clinical trials. For a comprehensive review on this topic please refer to the following article (Van der Aa et al., 2013).

COMPUTATIONAL METHODS

While SCS methods are generating torrents of large-scale genomic datasets, the computational methods for analyzing these data are severely lacking. SCS data is distinct from standard NGS data and their analysis tools, due to inherent technical errors and noise, including coverage non-uniformity, sparse data, false-positive errors, amplification biases and allelic dropout events. Some of the first SCS analysis methods focused on quantifying single cell copy number profiles from read count data. To calculate SCS copy number profiles, a variable-binning algorithm was developed that normalizes errors in mappability in the human genome, by adjusting genomic intervals based on the expected number of reads (Baslan et al., 2012; Navin et al., 2011). This processing pipeline was developed into a user-friendly web server platform with impressive visualization tools called Ginkgo (http://gb.cshl.edu/ginkgo). Another copy number method uses SCS read count data generated from DOP-PCR that corrects for GC bias and performs binary segmentation followed by dynamic thresholding (Zhang et al., 2013).

Several computational methods have also been developed for analyzing RNA SCS datasets to mitigate technical error. In one method RNA spike-in controls were used to quantify technical noise during WTA (Katayama et al., 2013). In another method unique molecular identifiers (UMI) were used to label RNA before WTA and sequencing, to eliminate amplification bias (Islam et al., 2014). Computational methods have also been developed to model noise in RNA SCS data using a low-magnitude Poisson processes (Brennecke et al., 2013). Another RNA SCS method called Monocle represents each cell as a point in a high-dimension space, and uses dimensionality reduction to extract essential features over time (Trapnell et al., 2014). Another study developed a latent variable model for single cell RNA data to reduce technical noise from over amplification and cell cycle genes (Buettner et al., 2015). Several algorithms have also been developed for assembly of microbial genomes from single cells. One method called E+V-SC uses lower initial coverage cutoff and then progressively increases the cutoff to incorporate more bases (Chitsaz et al., 2011). Another method called IDBA-UD uses similar filtering with progressive coverage thresholds strategy and error correction (Peng et al., 2012). A third method, SPAdes, tackled the uneven coverage problem by constructing paired assembly graphs utilizing read-pairs (Bankevich et al., 2012). In summary, while some initial tools have been developed, new quantitative methods are still urgently needed for analyzing DNA and RNA SCS datasets.

Alternatives to Single Cell Sequencing

SCS is not the appropriate technology to address every question in biology. In many studies alternative approaches will provide more powerful tools for investigating population diversity and identifying rare mutations. Methods such as deep-sequencing (Shah et al., 2012) or multi-region sequencing(Gerlinger et al., 2012) provide a more economical approach for resolving complex population substructure and have the advantage of providing genotyping information on thousands of cells. In cases where living tissue or cells are available, single cells can be subcloned to generate isogenic cell lines or organoids that act as proxies for single cells (Boj et al., 2015; Sachs and Clevers, 2014). These systems have the advantage of providing an unlimited amount of genetic material for analysis and can be used for functional assays. However, a notable limitation is that most cells are not capable of expanding in culture, which can introduce a strong bias in the representation of the final cells that are derived from a population. Furthermore as cells adapt to the cell culture environment they may alter transcriptional or epigenetic programs. In conclusion, alternative methods to SCS may be a better choice when functional studies are required, or when very rare cells must be detected in a population (without prior isolation or enrichment).

CONCLUSIONS & FUTURE DIRECTIONS

SCS methods have provided great insight into our understanding of biological diversity and rare cells that have previously been difficult to resolve in genomic data from bulk tissue samples. These tools have had a broad impact on many diverse fields of biology over the past 5 years, and several common applications have emerged: 1) delineating population diversity, 2) tracing cell lineages, 3) classifying cell types, and 4) genomic profiling of rare cells. While many initial studies have been published, there are still many applications that remain unexplored. In microbiology, SCS methods have yet to be applied to study viruses in single host cells, to understand how they infect and replicate differently in certain cell types. In neurobiology, SCS methods can provide important information on transcriptional programs in response to stimuli, including auditory, sensory and visual stimulation. In development, single-cell RNA sequencing can be used to study cell lineages in many organ systems to identify new markers and cell types. In tissue mosaicism, future studies should be directed at investigating the diversity of point mutations and indels in different tissue types which are likely to show even more diversity than copy number variations. Cancer immunotherapy is another exciting application, where SCS tools have great potential for illuminating phenomenon such as immunoediting and antigenicity in the context of intratumor heterogeneity. In cancer research, SCS can also help to understand the role of clonal diversity in complex biological processes, such as transformation, invasion and the evolution of chemoresistance (Navin, 2014; Van Loo and Voet, 2014).

Future efforts in technology development should focus on in situ SCS methods that can measure genomic data on single cells while preserving their spatial context in tissues (Crosetto et al., 2014). Future technologies should also be directed at linking phenotypes and genotypes in single cells, by combining methods such as live-cell imaging with SCS methods. Forthcoming technologies should also focus on collecting combinations of genomic information from the same single cell in parallel (ex. DNA and RNA, or RNA and epigenomic modifications). Some progress was recently made in this area, by demonstrating the feasibility of measuring both copy number states and RNA expression profiles in the same single cells (Dey et al., 2015). Another important area of technology development is highly-multiplexed single cell DNA and RNA sequencing, to enable the profiling of thousands of single cells in parallel, at a substantially lower cost. A recent technique using microwells and DNA beads with barcodes shows promise for enabling the profiling 10,000 – 100,000 single cells in parallel (Fan et al., 2015). Several companies (Fluidigm, Wafergen, Cellular Research) are also focusing their efforts on developing higher-throughput single cell RNA and DNA sequencing methods, which are expected to come to market soon. While most SCS studies are still cost-prohibitive, we expect that this barrier will largely be dissolved over the next few years, as the costs of NGS technologies (Illumina, Life Technologies) continues to plummet through new technical innovations and fierce industrial competition.

In closing, while the SCS field is still relatively new, it has already made a large impact on many diverse fields of biology and has led to great improvements to our fundamental understanding of human diseases. We expect that the demand and application of SCS tools will continue to grow tremendously in the coming years, as these methods become more refined, high-throughput, inexpensive and easier to use in standard research and clinical laboratories.