If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Healthy Longevity Translational Research Program, Department of Physiology, Yong Loo Lin School of Medicine, National University of Singapore, 117597 Singapore, Singapore
Biologists have long desired to understand multi-cellular processes at the resolution of the single cell. Tremendous efforts have been made over more than a century to decipher biology at the single cell level from the advent of immunohistochemistry to high-plex multi-parametric cytometry. More recently, technological developments in extracting and labelling nucleic acids from single cells have boosted single-cell information acquisition to include the genome, transcriptome, epigenome, proteome, and more, even simultaneously collecting data from multiple modalities. Here we will review some of the original motivations that have driven the development of new single cell tools, providing perspective on why these new tools were created and which tools we hope to see developed in the future.
Single cell sequencing technologies are unbiased bottom-up approaches
Ultimately, the questions being answered with single cell sequencing technologies are some form of: How do the individual parts add up to make the whole? This overarching question can be addressed by solving many smaller problems including: how to classify different cell types, how much variation exists in one cell type, how many routes could one cell possibly take during a cell state transition, how individual cells interact with each other to create new functions and how to use all of that information to improve those functions.
Single cell sequencing technologies have multiple advantages compared to earlier single cell analysis methods such as immunohistochemistry or flow cytometry. First, they do not require much prior knowledge. Although known markers of certain cell types could help to confirm the bioinformatic classification, single cell methods do not necessarily require this information to separate distinct cell clusters. This is especially meaningful when dealing with cell types that have only been loosely studied or are even unknown. Many recent studies have identified new cell types or subtypes using single cell RNA-seq (scRNA-seq) [
]. This one modality alone has already reshaped our understanding of the diversity of cells that compose humans and other animals.
Second, single cell sequencing technologies have the ability to assess the cell state, that is to measure heterogeneity inside one cell type or subtype. Understanding the possible cell states of one cell type is not only important for setting a standard transcriptional cutoff to distinguish specific cell types, but also essential for predicting the range of responses that a cell could make under physiological or pharmacological conditions. Many factors contribute to the cell state including cell cycle [
]. An example of the importance of understanding the cell state within a cell type can be found in human embryonic stem cells (hESCs). Through scRNA-seq, hESCs have been shown to exist in different states, mediated by cyclins at different cell cycle stages. These different cell states display differences in signaling pathway activity which further leads to a difference in the differentiation potential into multiple lineages [
Third, single-cell sequencing technologies can be multi-layered. As technologies have developed, researchers can now look at multiple molecular modalities across cells or even within the same cells. Examples include chromatin accessibility (scATAC-seq [
]). Traditional single cell analysis can also be combined with single cell sequencing, for example, the combination of flow cytometry with scRNA-seq (MARS-seq [
]). These combinations will help to dissect the regulatory network at single cell level in the future as they reach to better precision and coverage.
Finally, these single cell sequencing methods can be combined with CRISPR screens to be able to measure the molecular phenotype of thousands of individual cells across multiple modalities while perturbing the function of one or more genes within those cells [
With the explosion of new single cell measurement techniques, we soon hope to be able to convert any bulk sequencing assay into a single cell version. Below we will describe some of the single cell sequencing technologies available and discuss ways in which these techniques are being put to use to answer fundamental biological questions. We also summarize this information in Table 1 to help the reader choose a suitable single cell technology for their biological question.
Table 1Biological questions and their suitable single cell solutions.
C1 HT provides full transcript information for thousands of genes in single cells, but is more expensive. Targeted sequencing is accurate and cost-effective for cancer types with known mutation hotspots.
homing CRISPR could trace a longer period of time. CARLIN is an inducible system that can be tissue specific and the detection of mutation is directly from single cell transcriptome.
MARS-seq uses traditional flow cytometry and has limitation in multiplexing. The other 3 methods could include more surface markers. ESCAPETM is a commercialized product.
antigen specific T/B cell responses
targeted VDJ sequencing
Diagnosis, Prognosis
scRNA-seq(Chromium), targeted sequencing
scRNA-seq provides more comprehensive information. Targeted sequencing requires more prior knowledge but is more cost-effective and clinically applicable.
Visium is a commercialized product that does not require special instruments, and is suitable for studying changes in areas or zones rather than in individual cells.
Technologies with more applications than indicated above were labeled with bold fonts.
Multiple methods of scRNA-Seq have been developed ranging from low to high throughput. We will touch on a few of them here. scRNA-Seq experiments originated by placing individual cells into tubes or wells of a 96 well plate. Cells were collected by flow cytometry or limiting dilution and then reverse transcription and PCR amplification were carried out inside each individual well adding unique well-based barcodes to distinguish between data originating from different wells in the resulting sequencing (Fig. 1A). The SMART-Seq method, developed by Ramsköld, et. al. [
] and commercialized by Clontech (now Takara), was the first efficient method for obtaining single cell gene expression libraries and has become the gold standard method for single cell gene expression data collection. While relatively low throughput, single tube or plate-based methods provide full length gene level data and have relatively high gene counts compared to other methods. For in-depth analysis of small cell populations, plate-based sequencing is the best option. However, in addition to the low throughput nature of the assays, they are also very expensive, costing tens of dollars per cell in reagents.
Fig. 1Single cell isolation and barcoding strategies. A. Plate-based methods. FACS or limiting dilutions isolate single cells into individual wells (pipette picture by Freepik). Each well contains one unique barcode to label each single cell RNA-seq libraries. B. Droplet-based methods. Water-in-oil droplets are formed in microfluidics. Each droplet contains one hydrogel bead with uniquely barcoded primers and one single cell. C. Nanowell-based methods. Each nanowell is designed to be just enough for one single cell to sit in, and each nanowell contains one unique barcode. D. Combinatorial indexing methods. Cells go through cycles of splitting into individual tubes adding a tube specific barcode and pooling for random distribution to the next splitting. After a few cycles, most cells would receive different combinatorial barcodes indicated by different colors.
Droplet-based scRNA-Seq, especially commercial products from 10x Genomics, are currently the most commonly used methods for measuring gene expression in single cells. With droplet technology, single cells are encapsulated into water-in-oil droplets using microfluidic devices. A working droplet contains one hydrogel bead with unique barcoded primers, one single cell, lysis and reaction buffer (Fig. 1B). The RNA in each droplet is barcoded during the reverse transcription step so that all subsequent downstream steps can be performed in bulk. Using droplet methods, one can easily sample tens thousands of single cells with acceptable transcript coverage at a cost as low as $0.01 per cell. However, in any droplet-based system, more than one cell can be captured in a droplet and these ‘multiplets’ can confound the downstream data analysis.
In contrast to droplet-based methods, multiple academic and commercial groups have developed nanowell-based systems for collecting single cell gene expression including Cyto-seq [
] and commercial systems by BD Biosciences and Singlron Biotechnologies. In general, these methods work by creating small nanometer diameter structures, each containing a bead coated by uniquely barcoded RT primers and that are only large enough to hold one cell (Fig. 1C). These methods reduce the problem of obtaining multiplets, but suffer from lower gene coverage and are restricted to capturing cells of a limited size range.
96-well plate-based methods have also been developed using a technique called split and pool barcoding or combinatorial indexing. These methods achieve higher throughput with reduced cost compared to standard plate-based or droplet-based methods. The methods, which include Sci-seq [
], are capable of simultaneously interrogating millions of cells. In these assays, cells are pooled into each well of a plate and different barcodes are added to all of the RNA or cDNA in each cell within a given well. The cells from all wells are then pooled together and then split back out into different wells for another round of barcoding. By performing multiple rounds of splitting, pooling and barcoding the cells, the RNA from each cell eventually receives a unique set of barcodes, as if they had been separated into droplets or nanowells (Fig. 1D). Like droplet methods, multiplets can be identified, but compared to nanowells, these methods have the advantage of being able to utilize any cell type or even nuclei as input. While significantly less expensive than droplet systems and not requiring any special hardware such as nanowell systems, the cells must be chemically fixed, resulting in a reduced gene coverage compared to both droplets and nanowells, although higher gene coverage could be achieved with more saturated sequencing depth.
Given the high dimensionality and the different nature of single cell gene expression data compared to bulk gene expression, data analysis continues to be a major bottleneck in converting the raw data into discoveries. One major challenge in scRNA-seq data analysis comes from the uncertainty of the measurement due to stochastic dropout of reads given the low starting input material in a single cell. Therefore, the observed zero read counts for a given gene could mean that the gene is either not expressed or for technical reasons the gene was not detected. Statistical models and imputation are two main ways to tackle this problem by different algorithms. Data-smoothing, feature selection and dimension reduction can further improve the accuracy and remove technical noises. We recommend recent review articles [
] that provide more comprehensive demonstration of the bioinformatic principles and the respective tools. In general, by looking at correlated expression of multiple genes and/or selected features, researchers have achieved improved accuracies in measuring single cell variabilities.
When single cell RNA-seq technologies move forward to increase their coverage and reduce their cost, the precision may gradually improve to achieve accurate measurement of individual genes. We have seen improved quality of data as 10x Genomics has optimized their chemistry and released new versions [
] and there is also room for improvement for other methods mentioned above, e.g., the choice of fixatives in combinatorial indexing methods, potential bioengineering of a reverse transcriptase to derive higher efficiency in all methods, improving surface treatment of tubes and tips to reduce sample loss and even optimizing the concentration of key components in different buffers. We look forward to continuous upgrading of single cell RNA-seq technologies to achieve more accurate measurement of single cell gene expression and a better balance of cost and throughput.
Human cell atlas
The systematic study of an organism, an organ or a tissue requires generating a reference of all cells that make up that organism, organ or tissue. The Human Cell Atlas Project (HCAP) [
], which is generating a single cell reference database of human cells, is the single-cell equivalent to the Human Genome Project for genetics. The current focus of the HCAP is to develop a transcriptome reference of single cells throughout the body using scRNA-seq technologies.
The HCAP coordinates global efforts in establishing a human cell atlas through intensive international collaborations. The benefit of distributing responsibilities is that scientists could afford to study the organ or tissue of their particular interest. This has led to numerous discoveries [
], but it also brings challenges, in particular in combining datasets collected and processed from different sources and with different technologies, each producing unique technical noise. The HCA has created standard operating procedures for sample processing in order to minimize technical variation [
]) have been developed to correct batch effects in order to improve data integration, but batch variation remains a major technical hurdle. Another strategy to minimize batch effect could be by developing higher throughput scRNA-seq methods so that one group could afford to profile entire organisms. Indeed, using Microwell-Seq, Guo's group was able to complete the first mouse single cell atlas in 2018 [
], respectively with the combinatorial indexing method. While these serve as useful draft atlases, future revision will be required as technologies with a better balance of cost and throughput become available.
Using a mix of different technologies, human cell atlas scientists have collectively profiled more than 39 million cells in different organs of the human body [
] aimed to profile 30-100 million human cells from all major organs in ethnically diverse males and females. We also look forward to new technologies developed to either improve the accuracy or add more modalities to the HCA.
Lineage tracing and cellular trajectory
Understanding how cells relate to each other is a key feature to understanding cell function. Not only is cell lineage important for understanding normal cellular function, but understanding how cancer cells evolve is also important for identifying new therapeutics. Manual lineage tracing is accurate but, labor intensive and only manageable for relatively a small number of cells. The best example is the generation of the Caenorhabditis elegans cell lineage map [
], for which a Nobel prize was awarded in 2002. When studying a larger number of cells or when there is only a short time window for sampling, other approaches are required.
For larger scale studies, researchers can make use of the naturally occurring random mutations accumulating in each round of the genome replication [
]. However, amplifying the whole genome uniformly in single cells is challenging, because there are only two copies to start with, not to mention that the mutations could also be allelic. These problems initially led to high dropout rates where one allele was lost during the amplification and generally low genomic coverage where regions of genome do not get sequenced. Improved single cell DNA amplification methods have been developed including multiple displacement amplification (MDA [
], but even with these updated technologies, the coverage of single cell genome sequencing is still stochastic and low. While further development in the technology may improve the coverage, the main difficulty in single cell lineage tracing is the low frequency and randomness of de novo genomic mutations.
To work around this problem, multiple groups have developed novel systems using single cell sequencing to be able to trace the lineage of thousands of cells. Perli, et. al. and Kalhor, et. al. independently developed self-targeting CRIPSR guide RNA systems to trace lineages, where indels can be continuously created in the transgenic guide RNA sequences (Fig. 2) [
]. Church's group further applied their homing CRISPR system to generate transgenic mice and established the most comprehensive cell lineage tree of the mouse so far [
] took advantage of the multiple copies of the mitochondrial genome in each cell, analyzing the mitochondrial sequences in single cell ATAC-seq or RNA-seq data and achieving 1000-fold greater scale of clonal tracking than from the nuclear genome sequencing. Studies on clonal evolution of cancer cells have also become possible through the use of high throughput single cell targeted DNA sequencing and, because many of the known oncogenic mutations are in exons, even scRNA-seq has been used to trace mutation patterns [
Fig. 2Transgenic homing CRISPR system continuously generates mutations for tracing cell lineages. The guide RNA constructs are designed to include a PAM sequence and inserted into the genome by lentiviral infection for self-targeting. After the Cas9, guide RNAs complex edits its own genomic sequence, an indel (1) is generated on the guide RNA sequence. The mutated guide RNA could be expressed to further target the mutated genomic sequence and generate more indels until the PAM is mutated or the spacer is no longer active in editing. Genomic DNA could then be harvested and targeted sequencing could be performed to trace the indels generated.
In a related concept that aims to infer lineage based on gene expression, researchers have developed algorithms to trace cellular trajectory by pseudotime analysis. These algorithms, including frequently used ones such as Monocle [
] typically reduce the complexity of the data by dimensionality reduction, build trajectory to describe the dynamic cell state transition and position individual cells across the trajectory. The different approaches in dimensionality reduction and trajectory building taken by different algorithms could lead to varying performances. We recommend reading a comprehensive comparison by Saelens, et al. [
] was later developed based on an observation that the relative abundance of spliced and unspliced transcripts could be used to infer the dynamics of transcriptional changes. This method brings not only more experimental confidence in predicting where individual cells are moving towards in transcription, but also better estimates the speed of these changes. In general, these methods perform well in reconstructing the cell state dynamics during a continuous developmental process, however, do not reflect true developmental time nor the cell's clonal relationship. True lineage tracing can provide direct evidence of the lineage route and confirm the predicted cellular trajectory by pseudotime analysis.
Looking forward, a combination of homing CRISPR following targeted DNA sequencing and single cell RNA-seq could better demonstrate the correlation between a cell lineage tree and cellular trajectory. It would also be interesting to see if additional mitochondrial sequences in the single cell RNA-seq data could add to improve the performance of lineage tracing by homing CRISPR. In non-transgenic situations, increasing the coverage of whole genome sequencing or the depth of mitochondrial sequencing could potentially allow more accurate lineage tracing. Efforts could be made, for example, in engineering DNA polymerases with higher fidelity and more uniform amplification.
Finding cells responsible for organism level phenotypes
Identifying and studying cells that are responsible for a phenotype or disease condition is another fundamental biological question that can be approached with single cell sequencing technologies. When the cell population is rare or the phenotype is not obvious, single cell sequencing technologies can greatly help with the identification of the cell types involved. Recent studies on SARS-Cov2 have shown how single cell technologies can identify possible infection routes, antigen specific T cell responses and even identify therapeutic antibodies. Muus, et. al. analyzed the expression pattern of SARS-Cov2 entry genes across different tissues to explain the viral tropism [
], however, could not be made in large batches. Cao, et. al. performed single cell VDJ sequencing of B cell receptors from convalescent patients and identified 14 potent neutralizing monoclonal antibodies [
]. Ren, et. al. generated a comprehensive single cell transcriptome atlas that describe key immune features of COVID-19 patients with different severity, stage, age and sex [
]. In each case, single cell sequencing technologies made it possible to study the emerging infections and generate clinically useful information in near real time.
Single cell sequencing is also being used in cancer research where sample heterogeneity is a defining factor in tumor growth and response to therapy. For example, minimal residual disease (MRD) is considered as a high-risk factor for the relapse of many types of blood cancers including acute myeloid leukemia (AML). Targeted single cell DNA sequencing on hotspot mutations is currently being used to improve the assessment of persistent leukemia cells and guide therapeutic decisions [
Single cell technologies, especially scRNA-seq and targeted single cell DNA sequencing methods, have gradually become a routine tool in finding cell types that cause a phenotype or disease. As single cell methods advance into the clinic, they will greatly improve the diagnosis of diseases and provide valuable suggestion improving precision medicine.
Spatial control
Many biological questions exist in the context of spatial control, e.g. cell fate determination in early blastocyst or area patterning in brain. The single cell sequencing technologies described so far all use dissociated cells as input, stripping away the information about cell location. Spatial genomics, which performs sequencing while maintaining spatial information, is a growing field that is driving toward single cell resolution. When the number of cells to interrogate is small, microdissection, rather than single-cell dissociation can be performed to harvest samples while recording positional information. Peng, et. al. used this method to molecularly annotate the mid-gastrula mouse embryo [
] and has the potential to be developed into a single-cell assay as single cell isolation has been achieved by others with laser capture microdissection [
To address single cell spatial information in more systematic ways, groups initially turned to multiplex RNA in situ hybridization based approached to obtain gene expression data from high numbers of genes in the context of a tissue. Chen, et. al. developed a single molecule imaging approach, MERFISH, using multiplexed fluorescent in situ hybridization and could image 100 to 1000 RNA species in individual cells [
]. Around a similar time, Lee, et. al. reported a method of fluorescent in situ RNA sequencing (FISSEQ) that measured 4171 genes in human primary fibroblast [
]. This method was later commercialized by Spatial Transcriptomics, later acquired by 10x Genomics and now known as VisiumTM. However, due to the size limitation of printing positional barcodes and the thickness of tissue sections, the resolution does not reach single cell level, but represents data from a small number of nearby cells [
]. With a similar strategy, Rodriques, et. al. established Slide-seq with improved resolution and found that around 65% of their unique barcode could be matched with a single cell type [
]. These two new approaches greatly bring down the difficulties for many labs to study the spatial control of certain biological processes at near single cell resolution. Many recent studies have successfully applied spatial transcriptomics techniques, ranging from cancer, Alzheimer's disease, brain development [
] investigated the transcriptional changes in the brain in response to the presence of amyloid plaques using spatial transcriptomics and demonstrated early alterations of gene networks by correlating gene expression changes to the distance from the plaques [
]. Whenever a physical structure plays a functional role, spatial transcriptomics may be a useful tool to solve the problem. The combination of scRNA-Seq and spatial transcriptomics has also been used [
] to improve the quality of the spatial sequencing data. We believe that the greatest understanding of the interplay between cells will involve a combination of approaches. For example, collecting transcriptomic data from single cells obtained through dissociation and then using it to select the key genes for imaging-based methods of multiplex in situ hybridization.
Conclusions
We have seen explosive advances of single cell technologies since Tang, et. al. developed the first single cell RNA-seq method in 2009 [
]. Biological and technological research have always worked hand-in-hand, but never more so than in the field of single cell sequencing. Technology driven research focuses on developing new methods and improving existing ones for the needs of the biologists who then can solve ever more complicated questions. Biology driven research, on the other hand, accumulates background knowledge and raises the necessary demand for the desired technology development. In this review, we summarized the broad biological questions requiring single cell resolution and provided our views of suitable technology solutions for them.
As the field continues to develop, new technologies for generating single cell gene expression will emerge, lowering costs and increasing throughput. We expect to see further optimization of the chemistry in single cell RNA-seq technologies and the bioengineering of new enzymes leading to better precision and accuracy. As we learn more about human biology through the use of single-cell technologies, we anticipate some of the greatest gains in the field of precision medicine, based not only on new drug development, but also by directly using single cell technologies in a diagnostic clinical setting.
What we have only mentioned a bit here but are also important is the equally immense growth in both the need for and the number of bioinformatics tools. Each technology described requires its own set of data analysis tools and each technology has its own quirks that require specific knowledge to address. While single cell sequencing technologies are enabling a greater understanding of biology and disease, the data can only be interpreted correctly with a close collaboration of the wet and dry lab scientists. When working together, there is no limit to the questions these new technologies can answer about the fundamentals of cell biology.
Declaration of Competing Interest
Tao Yu declares that he has no conflict of interest. Jonathan Scolnick is a shareholder in and consultant for Proteona Pte Ltd.
References
Zeisel A.
M͡oz-Manchado A.B.
Codeluppi S.
et al.
Cell types in the mouse cortex and hippocampus revealed by single-cell RNA-Seq.
00013-3&id=gr1.jpg){kind=link}
00013-3&id=gr2.jpg){kind=link}