Decision-Making for RNA Sequencing

RNA sequencing is increasingly used by many laboratories for understanding the transcriptomic changes taking place inside the cell. However, it’s important to first understand how to correctly perform these sequencing experiments as they tend to be expensive, labor-intensive, and require some training and skill to be done right. This article includes excerpts from a recent Bench Tips webinar on “Improving and Optimizing Sequencing Workflows.”

Deciding on the type of sequencing

The first question to be asked before embarking on a RNA sequencing experiment is, “What type of sequencing is needed to answer the biological question at hand?” Bulk RNA sequencing measures the average gene expression levels across a tissue sample. With single-cell RNA sequencing (scRNASeq), gene expression can be measured at the cellular level. There are trade-offs in terms of time, cost, effort, and level of detail obtained from each technique. “Bulk sequencing is useful for comparative transcriptomics and biomarker studies, but it cannot capture the heterogeneity of the tissue,” says Mandovi Chatterjee, Ph.D., Research Associate at the Harvard Institute of Therapeutic Sciences and Director of the Single Cell Core at Harvard Medical School. “The constituent cell types and their abundance in a complex tissue sample can only be analyzed using scRNASeq.” Hence, scRNASeq is useful for studying the variability in gene expression in a certain cell population, for time course studies, for developmental studies, or lineage tracing.

If a sample has high cellular heterogeneity, scRNASeq can provide a better understanding of the observed changes. Yonatan Katzenelenbogen, a Graduate Student in the Laboratory of Dr. Ido Amit Lab in the Department of Immunology at the Weizmann Institute of Science in Israel, is using scRNASeq for immunology research. According to Yonatan, cellular framework depends on cell type and cell state. “The cell states within different cell types can be used to answer interesting biological questions and some of this complexity is lost with bulk sequencing.” However, scRNASeq is more expensive and has lower sensitivity than bulk sequencing. If there are cell surface markers to help sort and select the different cell populations in a sample, then scRNASeq may not be necessary.

Selecting the right platform

If using scRNASeq, the question becomes what platform to use. Different platforms are available for scRNASeq where the sequencing can be performed in wells (CEL-Seq, MARS-Seq, SMART-Seq, SCRB-Seq, Quartz-Seq), in nano/microwells (Seq-well), or in droplets (InDrops, DropSeq). Some platforms are suitable for high-throughput analysis, while others are better for low-throughput use. With any platform there is always a trade-off between throughput and sensitivity. Throughput refers to the number of cells that can be profiled, while sensitivity refers to the number of RNA molecules that can be detected. The higher the throughput, the lower the sensitivity. The choice of platform also depends on the sample type and experimental needs. “High-throughput sequencing is good for looking at overall heterogeneity in the tissue or for detecting rare sub-population of cells, which requires profiling a large number of cells/samples,” says Chatterjee. However, it’s not possible to do high-throughput sequencing when the sample size is limited. High-throughput sequencing is also not ideal for studies that require full-length transcripts.

Choosing and optimizing the protocols

Every scRNASeq workflow includes sample preparation, barcoding, library generation, sequencing, and, finally, data analysis. Sample prep involves dissociating the sample tissue to a cell suspension so the cells can be sorted, lysed, and barcoded. “Sample prep is unequivocally the crucial step in getting good-quality data,” says Chatterjee. “High-quality sample is a prerequisite for scRNASeq.” It requires minimal debris and a single-cell suspension with no clumps. The viability of the cells should ideally be more than 90% and the integrity of the cell membrane should be maintained until encapsulated. “The cells should be alive up until the time that they are barcoded,” says Chatterjee. “While we do run samples even if the viability is low, the free-floating RNA from the lysed cells leads to higher background noise, which translates to more wasted reads.”

Thomas Krausgruber, Ph.D., a Senior Postdoctoral Fellow in the Laboratory of Dr. Christoph Bock, at the Research Center for Molecular Medicine of the Austrian Academy of Sciences, uses both bulk and scRNASeq for studying structural immunity. He isolates structural cells (endothelium, epithelium, and fibroblasts) from 12 different organs such as, skin, lungs, liver, and brain and purifies them using flow cytometry-based cell sorting. “We had to optimize our isolation protocols to make sure we got high-quality and high-viability samples from each organ for our experiments,” he says. This enabled him to reveal organ-specific immune adaptation of structural cells and the complex interactions between the structural and immune cells.

Katzenelenbogen agrees that some dissociation protocols tend to enrich or deplete certain cell populations, which are hard to benchmark. “When we start a new project, we compare various dissociation protocols to see which one gives us the best results for maintaining cell heterogeneity.” There are many good online resources that give information on tissue-specific protocols and having good quality control tests done before and after sequencing also helps.

Batch effect can be a big challenge when running scRNASeq experiments, and it is often unavoidable when working with patient samples and when doing time-course or perturbation studies. Batch effects can arise due to many reasons. They can result from the operator, sample prep, different batches of reagents, animals, different times for library prep, different sequencing runs, and more. Some of these can be minimized by using the same operator, the same sets of reagents, prepping libraries and sequencing together, by pooling samples by barcoding (hastagging, MULTI-seq), and by using biological replicates. “Batch effect correction works best when individual samples contain sufficient internal complexity to identify shared sources of transcriptional variation,” says Chatterjee.

One of the other important questions that needs to be addressed when designing an experiment is, “How many cells need to be analyzed and how deep do you need to sequence?” “The rule of thumb is that you need 50–100 cells with a unique transcriptome signature to form a distinct cluster in a t-SNE plot,” says Chatterjee. The rarer the cell population, the higher the number of cells to be analyzed and the deeper the sequencing to be done. The higher the sequencing depth, the higher the cost. “There is a fine balance to strike,” says Chatterjee. “Hence, people sometimes start their experiments with a shallow sequencing depth and then look at the data to decide how much deeper to go.”

Finally, it’s important to take into account funding and resource constraints to figure out which platform to invest in and how many samples to run. “Every experiment requires some personalized attention,” says Chatterjee. “What works for others may not work for you.”

How to Obtain a Good-Quality Sample for Sequencing

• Keep sample preparation time as short as possible
• Maintain a low temperature during the sample prep
• Optimize a dissociation protocol that is best suited for the cell type used
• Keep lysis conditions as gentle as possible
• Use short protocol time and bigger nozzle for cell sorting
• Reduce time for centrifugation and resuspension
• Remove debris by filtration or density medium separation
• Include BSA (up to 1%) or FBS (up to 2%) in the final buffer
• Always run a pilot experiment to make sure everything goes as expected