Microarray hybridization has been a robust and reliable method that has been used for decades, but it has disadvantages. First, a significant amount of input RNA is required and so it is not suited to applications where the starting material is limited, notably in single-cell analyses. Second, it has a poor ability to discriminate between different very weak expression signals and between different very strong expression signals. The amounts of different transcripts in a cell can vary over five orders of magnitude but although microarray hybridization can reliably discriminate between weak and strong expression, its capacity for quantifying transcription is limited. Microarray hybridization analyses have also been limited by prior knowledge of genes (probes are traditionally designed using the sequences of known genes are so are limited to tracking known genes only).
As an alternative to microarray hybridization, more quantitative, digital profiling is possible by sequencing cDNA copies of transcripts. In the past, methods were devised to retrieve sequence tags from individual cDNAs that could be sequenced quickly, such as the ingenious SAGE (serial analysis of gene expression) method where short sequence tags, up to 25 bp long, from multiple individual cDNAs were concatenated into artificial long constructs that were then sequenced. More recently, high-throughput DNA sequencing has largely supplanted microarray hybridization as the preferred approach for high-throughput transcription profiling.
Transcript profiling by RNA-Seq
For whole-transcriptome profiling, RNA-Seq is the method of choice. An RNA-Seq experiment often involves fragmenting some starting RNA, converting it into cDNA, and then ligating adaptor oligonucleotide sequences to the ends of the fragments; following amplification of the individual sequences using adaptor-specific primers (adaptor- ligation PCR), the ends of millions of fragments are sequenced (Figure1). The resulting sequence reads can be individually mapped to a reference sequence, the source genome (or the reference transcriptome, when finding novel transcripts is not a high priority).
Fig1. RNA-Seq. The method begins by converting input RNA into cDNA fragments, and alternative methods are available. The RNA can first be copied, using a reverse transcriptase and an oligo(dT) or random hexamer primer, and the original RNA strand then destroyed with the enzyme RNaseH, leaving a single DNA strand. Thereafter, the single DNA strand is copied using a DNA polymerase to make a second DNA strand, and the resulting double stranded DNA is fragmented. Alternatively, it has become common to fragment the input RNA first (typically by RNA hydrolysis or nebulization). The RNA fragments are converted into single-stranded DNA using a reverse transcriptase, and the resulting single-stranded DNA is converted to double-stranded cDNA. Therafter, adaptor oligonucleotides are ligated to the ends of the cDNA fragments. Primers specific for the adaptors can allow amplification of the DNA fragments whose ends can then be sequenced by a high-throughput sequencing method using adaptor-specific primers (adaptor-ligation PCR). In order to retain information on the “strandedness” of the RNA (to distinguish between sense and antisense transcripts), different adaptor oligonucleotides may be attached to the two ends of the fragments (shown here by green and orange coloring); that can be achieved in different ways, such as by using forked (Y-shaped) adaptor oligonucleotides.
RNA-Seq is sensitive and offers a way of profiling transcripts of single cells. It allows quantification of transcripts over five orders of magnitude, and because it does not rely, like microarray hybridization, on prior knowledge of genes, it can be used to identify new transcripts and alternative isoforms, extending genome and gene annotation. Like microarray hybridization, RNA-Seq is often used in differential gene expression analysis, comparing gene expression profiles of different cell sources, and different clustering programs can be used to identify shared groups of genes that show similar expression profiles in the different cell sources, and ones that show dramatically different profiles.
Because of possible amplification bias—some sequences may amplify more readily than others—different controls are used. It is common to have external controls: to the input RNA is added a collection of known, external “spike-in” RNAs whose concentrations have been pre-determined. Because long transcripts are fragmented into many smaller pieces, there are often additional internal controls (a single transcript can be represented by several different starting sequences, which are then amplified; consistency in the estimated number of sequence reads ultimately copied from the same transcript provides reassurance that quantification is accurate). The review by Hrdlickova et al. (2017) (PMID 27198714) appraises current practice.
After sequence reads from RNA-Seq have been mapped back to the transcription unit/exon of origin, traditional quantification has used the statistic RPKM (reads per kilobase per million mapped reads). That is, RPKM = C/LN, where C = number of mappable reads on a feature (a transcript or exon); L = length of the feature; and N = total number of mappable reads (in millions).
The RPKM statistic is useful for comparing expression profiles from two sources, but more recently, molecular barcoding has provided absolute quantification, counting individual RNA molecules. It requires that a primer used for amplification be partially degenerate: the oligonucleotide synthesis is designed so that at a consecutive number of internal nucleotide positions, each of the four nucleotides is made available for synthesis. In that case, the relevant primer is not a single sequence but a heterogeneous collection of very many related sequences. According to which of the many alternative primer sequences is used, cDNA copies of transcripts receive one specific “barcode” sequence that enables direct counting of individual molecules. We show the general principle of molecular barcoding in Figure2.
Fig2. Principle of molecular barcoding in RNA-Seq: incorporating random sequence tags from an amplification primer with a partially degenerate sequence. (A) Oligonucleotides are synthesized from the 3′ end, and here we imagine standard synthesis for the first 17 nucleotides in the primer on the left. Thereafter, at each position in the eight-nucleotide degenerate region (positions 18–25), all four possible bases are represented in the population of primer molecules (the primer molecules will have positions 1–17 in common but have variable sequences at positions 18–25). Amplification of cDNA sequences using a partially degenerate primer (plus a normal second primer) can allow absolute quantification of the expression of a gene because amplified cDNAs corresponding to individual transcript sequences 1, 2, 3, 4, 5, 6, and so on from the same gene will have incorporated a primer sequence with a randomly selected, variable, eight-nucleotide sequence tag, a “molecular barcode” often described as a unique molecular identifier (UMI). The eight degenerate nucleotide positions in this example would allow a total of 48 (65,536) different UMIs. Specific examples of using molecular barcoding are described for whole-transcriptome analysis of single cells in Section 7.4. (B) Splitting-and-pooling strategy to create degenerate nucleotide positions in an oligonucleotide. Here, after the seventeenth nucleotide has been incorporated into the growing primer oligonucleotide, the primer population is split into four equal parts that are then exposed to the four different dNTPs: at position 18, one-quarter of the primer population receives an A, one-quarter receives a C, one-quarter receives a G, and one-quarter receives a T. Thereafter, pooling of the four populations creates a mix of primers where the eighteenth nucleotide position is represented by A, C, G, or T. Further splitting-and-pooling contines until the last degenerate position.
