Introduction to single cell RNA sequencing

Single cell vs bulk RNA sequencing

Advances in sequencing technologies have transformed the way we study biological systems. They have enabled shifting the analysis from tissue to single cell level, and allowed a deeper level of understanding of complex tissues. Human tissues are often composed of a wide variety of different cell types, each with unique molecular characteristics such as gene expression. In order to better understand the gene expression and regulation in heterogeneous tissues and for example in clonal tumors, it is useful to take the analysis to the single cell level.

In bulk RNA sequencing, the gene expression is analyzed from a mixture of cells within a tissue sample (Figure 1). This gene expression is a weighted average across all cells. In heterogeneous samples, containing various cell types, the gene expression predominantly reflects the abundant cell types and can mask the effect of more rare cell types. This can potentially obscure a signal critical for the disease in cases where there are cell-type-specific effects such as blood cancers.

Bulk sequencing cannot reveal differences in gene expression between cell types. However, there are few methods developed to estimate cell type specific expression from bulk RNA-seq data such as MuSiC [1].

Single cell RNAseq (scRNA-seq) measures the gene expression at the level of individual cells. Single cell methods enable the detection of different cell types in the tissues as well as their abundance (Figure 1). Studying gene expression on a single cell level can give more insights into the molecular characteristics of different cell types and how the expression varies between the cells in health and disease.

Figure 1. Created with BioRender

Overview of single cell RNA sequencing

Droplet-based sequencing

There are several different types of methods for scRNA-seq but the most popular of them is droplet-based method such as 10X Genomics [2].

The 10X Genomics scRNA-seq method creates Gel beads-in-emulsion (GEMs) (Figure 2). The workflow generates oil droplets, that encapsulates one barcoded cell bead, single cell, and reagents for the reverse transcription and cell lysis. By encapsulating all the items necessary for the sequencing library preparation inside a droplet, it’s possible to mark the RNA transcripts of each cell separately.

Figure 2. Image credit: 10X Genomics

How the cells and transcripts are detected

In the 10X Genomics scRNAseq method, the Gel beans are barcoded ie. they include a short sequence which is sequenced with the transcript that helps to identify the cell of origin. There are also other tags, such as unique molecular identifier (UMI) that gives each transcript its own unique molecular fingerprint. UMIs are useful in the quality control of the sequencing data and can be used to reduce PCR amplification bias. Poly(dT) sequence is used to capture mRNA transcripts from their 3’ or 5’ ends. In addition, adapter sequences required for sequencing are included.

The composition of the capture/sequencing tag and the workflow what happens inside a single GEM can be seen in figure 3. Inside a GEM, there is the gel bead with the capture/adapter tags attached, one cell (ideally), and reagents for creating a sequencing library from the mRNA transcripts inside the cell. The cell barcode is the same for each tag attached to a single GEM.This marks all the transcripts inside this GEM to a single cell. UMIs, however, are unique and they give an unique tag for each single transcript in the cell.

The enzymes inside the GEM lyse (break down) the cell, releasing the mRNA trapped inside the cell. The mRNAs pair to the poly(dT) part of the tag with their poly-A tail and the reverse transcription can start. The original mRNA is used as a template to build complementary DNA that is attached to our capture/sequencing tag. The gel bead dissolves and the tagged cDNA transcripts can be purified from the GEM. All the cDNAs can be now sequenced together, because they all contain a cell specific barcode and a UMI tag.

Figure 3. Created with BioRender. Workflow figure adapted from 10X Genomics.

Challenges and limitations

Despite the advancements in single cell technologies, scRNA-seq is still a complex technique with many challenges and limitations. These all should be taken into careful consideration during the planning of scRNA-seq experiments, and when deciding whether single cell resolution is even needed or would bulk sequencing be enough.

Costs and scalability

For single cell methods, the sample generation, library preparation, and sequencing is more expensive compared to bulk methods.This is a significant barrier for many researchers in using these methods, especially when the number of cells or samples studied is large. In addition to higher costs of single cell sample preparation, it is also more laborious and time consuming.

While droplet-based methods like 10X Genomics allow high throughput, there are limitations in terms of the maximum number of cells or the sensitivity for rare transcripts. With the limited number of maximum cells, it might be difficult to study rare cell types efficiently.

Computational challenges

In short, analysis of single cell data is more complicated and difficult to interpret. ScRNA-seq generates a large volume of data when each sample consists of thousands of cells and for each cell from 50,000 to 150,000 reads are generated [3]. In comparison, a typical bulk RNA-seq analysis is up to 20 million reads per sample. ScRNA-seq data has high dimensionality that requires advanced statistical methods as well as high computational power for the analysis. This means more memory, larger storage, and more running time.

Many current methods are not optimized for scRNA-seq data with a large amount of samples (Seurat, inferCNV). Currently there are few GPU-based solutions for analysis and more should be developed because the datasets and thus the computational requirements keep growing (For example rapids-singlecell within the scverse : [4]).

Technical challenges

Technical variability across cells and samples may cause more variation than the actual biological cell state and steps should be taken to remove them.

Capture efficiency

Characteristics typical for scRNA-seq data that complicate the analysis include the low sequencing depth per cell. Droplet-based methods have shallow sequencing depth (10–50% of the transcriptome detected) which results in zero counts for many genes. This creates an issue because we cannot say if the zero counts are due to the transcript not captured or if the gene is actually not expressed. The capture efficiency may also be cell specific — different cells may get different types and numbers of transcripts captured.

Library quality

The quality of the sequencing library affects the quality of the data and downstream analyses. ScRNA-seq analysis is based on an assumption that each GEM contain a single cell and many of the library quality issues violate this. Library quality issues often indicate problems at sample processing:

• samples were not processed fast enough and RNA degraded or cells started to die
• cell dissociation was not done gently enough which might have led to a release of RNA causing ambient RNA effect
• dissociation also might not have been done ‘hard’ enough and the cells stick together leading to doublets

Figure 4. Created with BioRender.

Amplification bias

Amplification bias is well known effect in PCR reaction for sequencing （Figure 5). Some transcripts are amplified more than the others which results in faulty quantities of the transcripts.This bias is corrected with UMIs — only the unique transcripts are counted.

Figure 5. Image credit: Lexogen [5]

Batch effects

Batch effects are one of the most significant confounding issues in scRNA-seq analysis. Multiple different factors during the processing and sequencing create batches:

• Performing RNA isolations on different days
• Library preparations on different days
• Different persons performing sample preparations
• Different reagents
• Different locations of sample preparation or sequencing

Ideally, all the samples would be processed exactly the same way. However, this is not always possible. In this case, it is important to plan the processing batches in a way that the different groups/phenotypes being studied are not in separate batches. Processing the groups in separate batches makes it difficult to separate the batch effect from the true biological effect between these groups (Figure 6).

Figure 6. Image credit: Hicks et al. [6].

Instead, try to make sure you split the replicates of each group across batches (Figure 7). Make sure to keep the batch information for later analysis stage where you can regress out the variation due to batch effect or integrate across batches.

Figure 7. Image credit: Hicks et al. [6].

Biological variability across cells/samples

Biological variability across cells/samples is both an opportunity and a challenge in scRNA-seq studies. Uninteresting sources of biological variation may obscure the important differences between cell types/states. Some of the (uninteresting) sources of biological variability are transcriptional bursting, varying rates of RNA processing, continuous or discrete cell identities, environmental stimuli, and cell cycle.

Transcriptional bursting and cell cycle

Genes can be transcribed from DNA to RNA in “bursts” or “pulses,” a phenomenon known as transcriptional bursting. Since genes aren’t needed continuously, transcription occurs only when necessary. This bursting pattern explains some of the high variability observed between cells. The process is closely tied to the cell cycle, as cells express different genes based on their current state or transition phase (G1, S, G2/M). For instance, DNA replication genes are upregulated during S phase, while genes controlling mitosis peak during G2/M.

RNA processing

Messenger RNA (mRNA) is not a very stable molecule and it is quickly processed or degraded depending on the needs of the cell. The needs for various mRNA products may differ a lot which leads to different rates of mRNA processing: some transcripts may be degraded faster than the others.

Cell identities

Cell identities (cell types) can be either continuous or discrete, with each type having a unique gene expression profile that reflects its function and state. In discrete cell types, the differences are clear — for example, T-cells and B-cells express distinct markers that identify their type. However, some cell types are more challenging to distinguish. In bone marrow, for instance, hematopoietic stem cells gradually specialize into different blood cell types through multiple intermediate progenitor stages. These transitional states make it difficult to clearly separate and identify the cells along their developmental pathway.

Environmental stimuli

Various sample-specific aspects in the environment may affect the gene expression patterns. These can be things such as stress, disease status, infections, and medications. If you are studying diseases such as cancers, it is important to take into notice what sorts of treatments and medications the patients have been given before sampling. These all can have a serious effect in the analysis.

References

1. Wang, X., Park, J., Susztak, K. et al. Bulk tissue cell type deconvolution with multi-subject single-cell expression reference. Nat Commun 10, 380 (2019). https://doi.org/10.1038/s41467-018-08023-x
2. 10X genomics
3. Single cell discoveries https://www.scdiscoveries.com/blog/knowledge/single-cell-transcriptome-sequencing-costs-considerations/
4. NVIDIA-Genomics: RAPIDS single cell
5. lexogen.com
6. Stephanie C Hicks, F William Townes, Mingxiang Teng, Rafael A Irizarry, Missing data and technical variability in single-cell RNA-sequencing experiments, Biostatistics, Volume 19, Issue 4, October 2018, Pages 562–578, https://doi.org/10.1093/biostatistics/kxx053