Fig 1.
An overview of the central dogma of molecular biology.
The flow of genetic information from double-stranded genomic DNA template to post-translationally modified proteins is depicted with molecular features critical to each stage enumerated. RNA-seq typically targets the mature mRNA molecules. Abbreviations: donor splice site (D); acceptor splice site (A); polyadenylation (poly (A)); untranslated region (UTR); splice site (SS); exonic splicing enhancer (ESE), exonic splicing silencer (ESS), intronic splicing enhancer (ISE); intronic splicing silencer (ISS).
Fig 2.
A typical RNA-seq experimental workflow involves the isolation of RNA from samples of interest, generation of sequencing libraries, use of a high-throughput sequencer to produce hundreds of millions of short paired-end reads, alignment of reads against a reference genome or transcriptome, and downstream analysis for expression estimation, differential expression, transcript isoform discovery, and other applications. Refer to S1 Table, S3 Table, and S7 Table for more details on the concepts depicted in this figure.
Fig 3.
RNA-seq library fragmentation and size selection strategies that influence interpretation and analysis.
RNA-seq library construction may involve both fragmentation and size selection. These procedures may be modified according to the integrity and amount of starting total RNA. The distributions of RNA molecule sizes are depicted for input total RNA and at various stages during the process of RNA/cDNA fragmentation and size selection. Commonly used methods for fragmentation and size selection are depicted along with the expected output of a quality-control assay at each stage (in the form of a capillary electrophoresis trace). Note that in the final library, it is typical that the majority of RNAs below a certain size (typically <150–200 bp) are underrepresented. Refer to S3 Table and S7 Table for more details on many of the concepts depicted in this figure.
Fig 4.
RNA-seq library enrichment strategies that influence interpretation and analysis.
RNA-seq library construction protocols differ widely, and these differences have significant consequences for data interpretation and analysis. The figure above illustrates representative alignment results for either total RNA or one of three commonly used enrichment strategies at a hypothetical genomic locus with very highly expressed ribosomal RNA (pink), highly expressed protein coding (green), lowly expressed protein coding (brown) and lowly expressed noncoding RNA (blue) genes. (A) If total RNA is sequenced without enrichment, the vast majority of reads correspond to a small number of very highly expressed RNA species such as ribosomal RNAs (rRNAs). In humans, ~95%–98% of all RNA molecules may be rRNAs. A significant amount of genomic DNA (gDNA) and unprocessed heteronuclear RNA (hnRNA, also known as pre-mRNA) contamination may also remain after typical RNA isolation procedures. As a result, most reads will align to intronic, intergenic, and especially to ribosomal gene regions. Since analysis of these molecules is rarely the target of RNA-seq, various enrichment strategies are commonly employed. The amount of gDNA contamination in total RNA can be reduced, but not entirely eliminated, by use of a deoxyribonuclease (DNase) treatment. The amount of unprocessed RNA can be reduced, but not entirely eliminated, by employing an RNA isolation method that attempts to keep nuclei intact and removing these to enrich for mature mRNAs present in the cytoplasmic compartment. Additional strategies are discussed in S3 Table. * When sequencing total RNA, a complete representation of the transcriptome is theoretically present, but in practical terms, insufficient sequence reads are obtained to sufficiently sample all transcripts of all types, and some enrichment strategy is required to reduce extremely abundant rRNA species. (B) Selective rRNA reduction kits use oligonucleotides complementary to ribosomal sequences to specifically reduce the abundance of rRNAs while maintaining a broad representation of transcript species. Since the oligonucleotide probes used in these kits are only designed to bind to and deplete rRNA sequences, a significant amount of unprocessed RNA and gDNA contamination may remain. (C) Poly(A) selection and (D) cDNA capture methods specifically enrich for (primarily) mature polyadenylated RNA species or specific targets (e.g., all known transcript exons), respectively. Since poly(A) selection specifically targets RNAs that have been polyadenylated—a modification that happens at the end of the transcription process—poly(A) selection results in an enrichment for mature, completely processed RNAs. Poly(A) selection and cDNA capture methods sacrifice some transcriptome representation for increased signal to noise for transcripts of greater interest. Poly(A) methods will fail to represent most noncoding and other nonpolyadenylated RNAs. Capture methods on the other hand will under-represent any loci not specifically included in the capture design. For example, in this case the brown gene was not included in the design, and therefore, expression of this gene would be underestimated. Each of the methods depicted here has advantages and disadvantages (S3 Table and S7 Table). Furthermore, the relative amounts of each class of RNA depicted in each panel are hypothetical examples meant to demonstrate the goals and principles of each enrichment strategy and should not be interpreted quantitatively. Refer to S4 Table for additional information on the effect of each enrichment strategy.
Fig 5.
An example RNA-seq analysis workflow is depicted for a typical gene expression and differential expression analysis. Such workflows have several common themes across different tool sets and RNA-seq analysis goals. RNA-seq analysis typically relies on inputs such as reference genome sequences, gene annotations, and raw sequence data. Working with these inputs requires familiarity with several standardized file formats such as FASTA (.fa), FASTQ, and gene transfer format (GTF). Typical RNA-seq analysis workflows start with raw data quality control (QC), then perform read trimming, alignment or assembly of reads, apply customized algorithms for a particular analysis goal (e.g., Cufflinks and Cuffdiff for gene expression analysis), and end with summarization and visualization of the results. For each step, alternative and representative tools and strategies are shown. There are many others. Each of the workflow steps depicted here and additional analysis vignettes are implemented in the Supplementary Tutorials accompanying this work and available online at www.rnaseq.wiki. Refer to S1–S3 Tables and S7 Table for more details on many of the concepts depicted in this figure as well as alternative tools for each step.
Fig 6.
Comparison of stranded and unstranded RNA-seq library methods and their influence on interpretation and analysis.
(A) Many RNA-seq library construction protocols do not maintain the strand identity of RNA transcripts in the sequence data (S1 Table). In these “unstranded” strategies, double-stranded cDNA is sequenced, and knowledge of the transcription strand of the RNA molecule is lost. This results in an even mix of reads from both strands. In panel A, a gene transcribed on the positive strand is shown in green, a second gene transcribed on the negative strand is shown in brown, and a third gene transcribed on the positive strand (partially overlapping the second gene) is shown in yellow. The first two genes are protein coding with the open reading frame (ORF) portion depicted as thick rectangles and the UTRs depicted as thin rectangles. The third gene is a noncoding RNA gene. Aligned paired-end read sequences (read 1 and read 2) are depicted as short colored bars connected by thin lines. The thin connecting line in each read pair depicts the portion of the cDNA fragment that remains unsequenced when the cDNA fragment is larger than two times the read length. Each read is colored according to the strand sequenced, blue for the positive (forward/sense) strand and red for the negative (reverse/antisense) strand. Using known annotations, the mapped position of each read, and knowledge of exon splicing patterns, the likely transcription strand of some reads can be inferred. However, for many aligned reads the transcription strand cannot be inferred and sense-antisense expression analysis is not possible. Note that for each gene, an approximately equal proportion of reads corresponding to each strand are observed. Also note that read pairing information can sometimes be used to infer which gene a read was likely derived from. These reads are referred to as “encompassing” read pairs, in which one read of a pair aligns within one exon and the second read of a pair aligns within another exon. However, reads that align within a region corresponding to overlapping genes cannot be unambiguously assigned to either gene (e.g., the portion of the brown and yellow genes that overlap). Note that in this figure we are not depicting any reads in which a single read of a read pair spans across an intron. These exon–exon “spanning” reads can usually be matched unambiguously to a transcript, even in an unstranded library, because the exon–exon junction alignments line up with known splice sites and exon boundaries. (B) More recent “stranded” RNA-seq library strategies allow the strand information to be retained. In the resulting alignments, depicted in panel B, the strand of the alignment corresponds in a predictable way to the transcription strand of the sequenced RNA molecule. Now we see that reads aligning within a gene are indicated as being derived from the expected transcription strand for that gene. Furthermore, in regions where two genes overlap on opposite strands, we can now unambiguously assign reads to each gene. (C) When strand information is maintained by the RNA-seq protocol, it can be visualized in genome browsers such as IGV [62]. For example, to make IGV color read alignments according to strand, use the “Color alignments” by “First-of-pair strand” setting (refer to S5 Table for more strand-related software settings).