Multiplexed Primer Extension sequencing: a targeted RNA-seq method that enables high-precision quantitation of mRNA splicing isoforms and rare pre-mRNA splicing intermediates

Abstract

The study of pre-mRNA splicing has been greatly aided by the advent of RNA sequencing (RNA-seq), which enables the genome-wide detection of discrete splice isoforms. Quantification of these splice isoforms requires analysis of splicing informative sequencing reads, those that unambiguously map to a single splice isoform, including exon-intron spanning alignments corresponding to retained introns, as well as exon-exon junction spanning alignments corresponding to either canonically- or alternatively-spliced isoforms. Because most RNA-seq experiments are designed to produce sequencing alignments that uniformly cover the entirety of transcripts, only a comparatively small number of splicing informative alignments are generated for any given splice site, leading to a decreased ability to detect and/or robustly quantify many splice isoforms. To address this problem, we have recently described a method termed Multiplexed Primer Extension sequencing, or MPE-seq, which uses pools of reverse transcription primers to target sequencing to user selected loci. By targeting reverse transcription to pre-mRNA splice junctions, this approach enables a dramatic enrichment in the fraction of splicing informative alignments generated per splicing event, yielding an increase in both the precision with which splicing efficiency can be measured, and in the detection of splice isoforms including rare splicing intermediates. Here we provide a brief review of the shortcomings associated with RNA-seq that drove our development of MPE-seq, as well as a detailed protocol for implementation of MPE-seq.

1. Introduction

The past decade has witnessed a growing appreciation for the role that pre-mRNA splicing plays in regulating eukaryotic gene expression. Contradicting the original concept that ‘one gene equals one protein’, it is now clear that alternative splicing plays a critical role in expanding the proteomes of eukaryotic organisms, enabling production of wide varieties of protein isoforms from single genetic loci [1]. Indeed, ever-increasing numbers of alternatively spliced variants are being identified, often associated with changing developmental or cellular states [2,3]. Consistent with its widespread role in gene regulation, the last several years have similarly seen an explosion of connections between human disease and mutations in the splicing pathway [4-8]. Included among these are mutations in the splice site sequences and regulatory cis-elements of individual transcripts that influence their expression, as well as mutations in components of the core machinery that catalyzes pre-mRNA splicing, which potentially impact the splicing of many transcripts [9]. Nevertheless, despite its significance, our field currently lacks answers to many critically important questions about pre-mRNA splicing, ranging from aspects of its basic mechanisms to understanding the mechanistic and physiological implications of disease-related mutations.

The advent of short-read sequencing technologies has dramatically impacted our understanding of the role of pre-mRNA splicing in eukaryotic biology [10]. By identifying the subset of alignments that traverse a splicing boundary, RNA sequencing (RNA-seq) has been used to identify countless novel splice variants, expanding our understanding of the expressed proteome and the role that splicing plays in generating this diversity [1]. Yet while this approach has been unquestionably successful in identification of novel transcripts and splice isoforms, most standard experiments lack sufficient power to quantitatively assess changes in many splicing events. Because of the uniform alignment coverage across transcripts that traditional RNA-seq generates, only a small number of splicing informative alignments – those which cross splice junctions or align within an intron – are generated for any given splicing event. This low sampling results in poor precision in quantification of many splice isoforms, confounding studies aimed at identifying the subsets of splicing events that are impacted by a particular treatment. We have recently overcome this deficiency by developing an approach we termed Multiplexed Primer Extension sequencing (MPE-seq) that targets reads to user selected splice junctions, greatly increasing alignment coverage at selected loci [11]. Here we present a brief review of the challenges associated with quantifying splice isoform abundance from traditional RNA-seq approaches as well as provide a detailed protocol for MPE-seq.

2. Splicing status is often poorly sampled by RNA-seq experiments.

Over the past decade, RNA-seq has become the tool of choice for genome-wide analyses of the transcriptome and has led to the identification of scores of previously unidentified transcripts and splice isoforms in many organisms [10]. The identification of sequencing alignments that unambiguously support different splice isoforms or pre-mRNA intermediates can be used to monitor changes in splicing status under changing conditions (Fig. 1A). Yet most standard RNA-seq experiments, having historically been optimized to monitor transcript levels, undersample many if not most splicing events. Importantly, alignments that unambiguously distinguish between mature and premature mRNAs, or between canonically and alternatively spliced isoforms, are present at only a small fraction of the total alignments per transcript (Fig. 1B). As such, the coefficient of variation associated with counting these splicing informative alignments can be very high, particularly for lowly expressed transcripts (Fig. 1C).

Fig. 1.

Quantitation of splicing isoforms from RNA-seq data: (A) Only a subset of alignments are splicing informative: those which can be unambiguously mapped to premature (including completely unspliced and lariat intermediate) or mature (fully spliced) molecules. Black arrows denote the conversion between these forms in the 1^st and 2^nd chemical steps of the pre-mRNA splicing reaction. (B) The fraction of alignments from a standard RNA-seq experiment that are completely exonic (E), mature (J), and premature (EI + IE) is shown for all single intron containing genes in S. pombe. (C) The coefficient of variation of replicate measurements of each event in (B) is plotted as a function of the (geometric) mean count depth for the isoform. Count data for (B) and (C) can be found in Table A.1. RNA-seq data are from Xu, etal. (2019) [11].

The high, but non-uniform variance associated with most splicing informative alignments complicates statistical analyses of their abundance (Fig. 1C). Data sets such as these are characterized by high false discovery rates: false positive discoveries derived from high (and unbalanced) variance between samples; and false negatives derived from events that are too poorly sampled to properly reveal the underlying difference between samples. Whereas high confidence events that are changed between two samples can often be identified from these data, these events are much more likely to be drawn from highly expressed transcripts because of the lower variance associated with their measurements, introducing selection biases which can dramatically influence the results of the study [12]. Moreover, many investigators seek to compare ‘affected’ versus ‘unaffected’ pools of events to better understand the properties of the affected events (e.g. the presence of cis-regulatory sequences, the gene ontology of the parent transcript, etc.). But such an analysis requires high confidence knowledge of both categories of events, and far too often in this field the absence of evidence of a change in isoform levels from an RNA-seq experiment is incorrectly inferred to mean that there is evidence of absence of a change. While statistical approaches have been published that work to reduce the challenges associated with this problem in isoform detection [13-16], none of these approaches can fully solve the problems associated with low read-count events. For these reasons, we set out to develop an approach that would increase the sampling of these otherwise poorly sampled regions of the transcriptome.

3. MPE-Seq

3.1. MPE-seq Method Overview

We recently described MPE-seq as a targeted RNA sequencing method based on primer extension that focuses sequencing reads at up to thousands of user selected loci [11]. By targeting reads to splice junctions (Fig. 2A), MPE-seq greatly increases the number of splicing informative alignments generated at a given site for a given experiment, allowing for the detection of rare splicing isoforms including pre-mRNA intermediates (Fig. 2B), and increasing the precision with which splicing efficiency can be measured (Fig. 2C). While we have recently applied this method to the budding yeast s.cerevisiae and the fission yeast S. pombe, yielding increases in the accuracy and precision of splice isoform measurements genome-wide, we anticipate that it will be readily translatable to other organisms, and here include suggestions for steps that might benefit from modification.

Fig. 2.

MPE-seq targets reads to splicing informative loci, increasing the precision of measurements of splicing efficiency compared to standard RNA-seq. (A) Steps in MPE-seq library preparation. Asterisk indicates the presence of aminoallyl-deoxyuridine, incorporated during reverse transcription. (B) Categories of MPE-seq alignments for quantifying the abundance of molecules at each chemical step of splicing for a single splicing event. (C) The correlation between measurements of splice index (SI, calculated as total unspliced reads divided by total spliced reads) for each splicing event is presented for replicate MPE-seq and RNA-seq S. cerevisiae libraries. Libraries were down sampled to five million reads each. The number of splicing events for which at least one premature and mature alignment exists in both replicates is represented by ‘n’. R² values were calculated using linear regression. MPE-seq count data can be found in Table A.2, and RNA-seq count data can be found in Table A.3.

Three key features distinguish MPE-seq from other approaches and enable the enrichment of sequencing alignments at splice junctions. First, reads are targeted to regions of interest by designing unique reverse transcription (RT) primers to each desired location (Fig. 2A). To effectively monitor splicing efficiency, we designed primers to regions just downstream of each targeted intron such that the cDNA resulting from reverse transcription would contain information about the corresponding transcript’s splicing status: those cDNA molecules that cross the upstream intron-exon junction are derived from unspliced mRNA, while those that cross an upstream exon-exon junction are derived from either a canonically or alternatively spliced mRNA. In addition to the gene specific sequence, each reverse transcription primer includes a portion of a next-generation sequencing adapter and a unique molecular adapter (UMI), the latter allowing for compression of PCR duplicates (Fig. 2A) [17]. Second, a modified nucleotide, aminoallyl-dUTP, is included in the reverse transcription reaction, which allows for purification of extended cDNAs and removal of excess unextended primer molecules (Fig. 2A): NHS-Biotin is coupled to the aminoallyl-uracil followed by streptavidin bead purification. Finally, addition of a portion of the 2^nd sequencing adapter is accomplished via a strand extension step analogous to template switching, wherein the adapter is appended to the 3’ end of the original cDNA molecule (Fig. 2A). Combining this approach with paired end sequencing allows for querying of both the 5’ and 3’ ends of cDNAs, enabling identification of pre-mRNA splicing intermediates.

3.2. MPE-seq Assay Design Considerations

3.2.1. RNA Input

Whereas traditional RNA-seq approaches require an initial step to enrich for mRNAs, either by positively selecting for poly(A)⁺ RNA or by negatively selecting against rRNAs, one advantage of MPE-seq is its ability to be used with unfractionated RNA. We have successfully generated high quality MPE-seq libraries using either total, unfractionated RNA as input, or poly(A)⁺ RNA [11]. Poly(A)⁺ RNA was used to reduce the frequency of off-target priming on rRNA, a variable that will depend upon the quality of the primers and the reaction conditions that are used during reverse transcription (see additional discussion below). It should be noted that selection of poly(A)⁺ RNA will remove nascent RNAs that have yet to receive a poly(A) tail, potentially introducing a bias in the detection and/or quantitation of certain isoforms.

While MPE-seq can be successfully applied using a wide variety of starting quantities of RNA, we have typically used 10μg of total cellular RNA as our initial input. We have generally started with higher total quantities (50μg) when using poly(A)⁺ selection to account for inefficiency in recovery of target RNAs in this process. The appropriate level of starting RNA may also vary based on the number of targets chosen and their relative expression within the organism/cell type of interest.

Fragmentation of RNA input is also encouraged to prevent previously described sequencing artefacts that generate a bias against long DNA molecules [18]. In our initial work in S. cerevisiae, we did not fragment the RNA due to the strong positional bias of introns: because the majority of introns in this organism are located near to the transcription start site, nearly all of the cDNAs generated in our experiments were expected to be short. By contrast, for most other organisms where introns are more equally distributed throughout gene bodies, fragmentation of the RNA prior to reverse transcription should be performed.

3.2.2. Target Selection

MPE-seq provides the unique opportunity to select the specific subset of transcripts for study. In selecting our target locations, we considered several design parameters. First, with increasing number of targets, each target will make up a smaller portion of the overall library. As such, adequate sampling of some targets may require greater sequencing depth. Secondly, the relative expression levels of different targets can have a large impact on the quality of the data generated. If highly and lowly expressed targets are selected in the same experiment, the highly expressed targets are likely to dominate the resulting library, potentially leaving lowly expressed transcripts undersampled. In this instance, it is possible to group targets based on similar relative expression levels, and then generate separate MPE-seq libraries for different expression quanta to ensure adequate sampling of all targets. The budding and fission yeast genomes contain roughly 300 and 5000 annotated introns, respectively, and we found that targeting all junctions in both species yielded adequate sampling of at least 90% of targeted unspliced and spliced isoform. By contrast, because the human genome contains ~200,000 annotated introns, a complete global analysis of each intron via MPE-seq would require read depth approaching standard RNA-seq to yield quantitative data on most introns. In this case binning targets based on expression level or features of interest (e.g. function) would allow for enrichment of splicing information of desired targets.

3.2.3. RT Primer Design & Reverse Transcription

It is important to emphasize that the design of reverse transcription primers is one of the most crucial steps in this method. Poorly designed primers can easily cross-hybridize with undesired RNAs, leading to a decreased fraction of on-target alignments. In designing primers that target a particular region, three primary design constraints should be considered. First, primers should be designed within short windows immediately downstream of the exon boundary of interest such that short-read sequencing is sufficient to cross the exon-exon or intron-exon boundary of interest. In our work, we have designed primers between 24-26 nucleotides in length that are complementary to regions within a 50 nucleotide window downstream of an intron-exon boundary, enabling analysis with 75 nucleotide reads using an Illumina NextSeq sequencer. The locations and lengths of these primer sequences can of course be changed to better address different questions. Second, the melting temperatures of the target regions of the primers should be as homogenous as possible, and as high as possible given the properties of the reverse transcriptase being used. In our experience, the use of thermostable enzymes such as Superscript IV in conjunction with the highest reaction temperature that retained efficient on-target annealing yielded the highest number of on-target alignments with the lowest abundance of off-target alignments in the sequencing data. Third, robust bioinformatic approaches should be employed to ensure that primers are as specific to the target as possible. Many tools exist for batch primer design that allow the user to select multiple regions in which to design primers with parameters for melting temperature and specificity. Particular attention should be paid to rRNA sequences as these are large sources for potential non-specific primer hybridization. Oligowiz, a program originally designed for microarray primer design, was used to design the primers for our initial MPE-seq study, using the basic constraints described above [19].

We add a note here specifically regarding primer design for groups of paralogous genes. As with all RNA-seq methods, differentiating between paralogous genes with MPE-seq can be difficult or impossible as unambiguous assignment requires alignments that reveal sequence differences between the paralogs. When designing MPE-seq reverse transcription primers to paralogous genes it is more important for differences between the paralogs to be contained within the extension region rather than the primer annealing region to maximize confidence in alignment allocation. Small differences in the primer annealing region could be sensitive to mis-priming during reverse transcription, allowing for extension from the incorrect paralog and obfuscating the assignment of the resulting alignment. By contrast, a single primer that hybridizes to an identical region of paralogs, but whose extension reveals even single nucleotide differences between the paralogs will enable higher quality differentiation between paralogous transcripts. Nevertheless, for paralogous genes that lack such differences but rather have identical sequences surrounding the junctions of interest, the inability to unambiguously determine the parental locus has led us to exclude these from our studies, either by removing targeting primers during experimental design, or by excluding unambiguous assignments during data analysis.

As noted above, reverse transcription is performed using gene specific primers that contain a UMI and a portion of the Nextera i5 sequencing adapter appended onto their 5’ ends (Fig. 2A). The UMI is a stretch of random nucleotides that is later used to identify reads resulting from PCR duplicates [17]. The number of random nucleotides included on these primers can be increased or decreased based on the user’s needs. In selecting this number, it is important to consider the number of expected molecules versus the number of possible unique molecules potentially quantified by the UMI.

Primers for use in MPE-seq can either be synthesized individually and then pooled, or synthesized in a pooled setting using a variety of commercial sources. In our original study, primers targeting S. cerevisiae splice junctions (309 in total) were synthesized both individually (by IDT) and in pooled format (by LC Sciences), while primers targeting S. pombe splice junctions (~4000 in total) were synthesized only in pooled format. Because the scale of pooled synthesis is small, larger quantities of the primers need to be generated prior to their use. This is made possible by 2 additional sequence features appended onto the 3’ ends of the synthesized primers: (1) a common sequence that enables PCR amplification and (2) a restriction site (SapI) that enables generation of an appropriate 3’ end on the final products [11]. The complex pool of oligos is first amplified via PCR with a forward primer containing a chemically blocked 5’ end and a reverse primer which is biotinylated at the 5’ end. The double stranded amplicon generated in this reaction is then subjected to digestion with SapI, which generates the mature 3’ end of the desired primers. To generate single stranded DNA, a digestion using Lamda exonuclease is performed. The presence of the 5’ block on the forward primer during amplification precludes degradation of the desired forward strand. Finally, a streptavidin purification is performed to specifically remove the undesirable cleaved DNA which contain a 5’ biotin from the reverse primer used during amplification. A protocol for amplification of bulk-synthesized primers is detailed below. See Table 12 for example sequences of an individually synthesized and a bulk-synthesized reverse transcription primer.

In designing the reverse transcription reaction, an important consideration is the concentration of each primer to include. For our original S. cerevisiae libraries, a total of 1 μg of pooled primer was used. This pool contained 309 individual primers at equimolar concentration, yielding roughly 160 fmol of each primer in the 1^st strand synthesis reaction. For S. pombe, 200 ng of the bulk-synthesized primer pool was included in each 1^st strand synthesis reaction. This pool contained 3918 primers mixed at equimolar concentration, yielding roughly 2.5 fmol of each primer in the 1^st strand synthesis reaction. It is important to note that while the primer concentrations in both of these instances is significantly higher than the concentrations of their target RNAs, the primers are nevertheless present at concentrations that are likely subsaturating for complete binding of the RNAs. Moreover, the level of saturation achieved for different RNA/primer combinations is likely to vary as a function of the thermodynamic properties of these different complexes. These differences should be meaningless when investigators are comparing the levels of two different isoforms, each of which is targeted by a common primer. That is to say, the efficiency of hybridization of a single primer should be identical between its matched spliced and unspliced (or alternatively spliced) targets. As such, splicing efficiency measurements should be indifferent to the absolute efficiency of any primer. By contrast, if investigators wish to directly compare the counts derived from two different primers, additional experiments would be necessary to establish the absolute efficiency of each primer pair.

The reaction conditions during reverse transcription are crucial for minimizing the abundance of off-target alignments and for maximizing cDNA yield. While the RNA template, buffer, and primer pool are mixed at room temperature, the primers are annealed to the RNA template by raising the temperature to 70°C for 1 minute followed by a 65°C incubation for 5 minutes. The reaction is then cooled to the optimal reaction temperature, ideally between 50-55°C, depending upon the enzyme being used and the design of the primers. To reduce non-specific annealing, it is essential that the temperature of this mixture is never allowed to go lower than the temperature at which reverse transcription takes place. In parallel, the reverse transcription enzyme along with the other reaction components are pre-heated to the reaction temperature, after which they are added to the primer-annealed RNA mixture. Here we re-emphasize the importance of maintaining all the reaction components at the reverse transcription temperature prior to mixing so that primers aren’t afforded the opportunity to mis-hybridize even for short times at lower temperatures. The choice of enzyme for use in reverse transcription is important for determining the optimal reaction temperature. Many commercial enzymes have different optimal synthesis temperatures and buffer components which will affect the melting temperature of the reverse transcription primers. This step may need to be optimized by the user by generating libraries after using a gradient of temperatures during reverse transcription. The temperature that yields the best library yield with lowest off-target and primer dimer fraction should be selected. For our studies we have used both Superscript III and IV reverse transcriptases. Reverse transcription was performed at 55°C and 50°C for S. cerevisiae and S. pombe libraries, respectively.

3.2.4. Sequencing Considerations

Depending on the targets selected, the sequencing parameters desired for MPE-seq may change (e.g. read depth, single vs paired end, read length, etc.). The read depth required for MPE-seq experiments is much lower than for standard RNA-seq. For example, S. cerevisiae libraries made by targeting every intron in the genome require only ~5 million reads to adequately sample the unspliced and spliced isoforms of most transcripts, whereas we estimate nearly 500 million reads of RNA-seq data would be necessary to yield splicing information of similar quality [11]. Nevertheless, as the number of MPE-seq targets increases so may the read depth required to adequately sample the desired isoforms. The isoforms of interest will also dictate the use of single versus paired-end reads. If the user is considering ‘simple’ measurements of unspliced versus spliced (or canonically versus alternatively spliced), single end reads that align across the splice junction will be sufficient. When considering read length, reads should be long enough to cross junctions with enough mapped bases on each side to ensure confidence in junction mapping. In our experience, reads that align with >3 nucleotides on each side of a junction can be confidently mapped. However, this may be organism and target dependent. By contrast, paired-end reads enable differentiation between pre-first step pre-mRNAs and lariat intermediate pre-mRNAs (Fig. 2C). The beginning position at which the paired read aligns is the position at which reverse transcription stopped and is informative as to which step of splicing has occurred for a pre-mRNA [11].

3.2.5. Optimization and Troubleshooting

When optimization and troubleshooting are required, the efficiency and progress of most steps of the MPE-seq library prep process can be assayed. In this section, we suggest several methods for assaying library prep progress and we note the steps during the protocol that are amenable to these methods. In our development and optimization of MPE-seq, quantitative PCR (qPCR) was the most useful method to assay each step of the protocol. At virtually every point in the protocol qPCR can be performed to assess the abundance of specific cDNA molecules. There are two primary qPCR methods that we utilized. The first is SYBR based qPCR, which has the advantage of being widely available and straight forward to use. One drawback of SYBR based qPCR is the inability to directly quantify the absolute abundance of DNA molecules in a sample. For this reason, we also used digital droplet PCR (dPCR), which allows for absolute quantification of target molecules in a sample with superior accuracy and precision compared to SYBR based qPCR [20]. Both methods require primers to be designed against specific targets. A typical MPE-seq experiment will contain many targets that may behave differently at each step of the protocol and for that reason we suggest designing PCR primer sets against multiple targets. For example, it may be useful to design PCR primers against targets across a range of expression levels. The position of the amplicons in relation to the reverse transcription priming sites should also be considered. For example, if using RNA that has been fragmented to a mean length of 200 nt, a quantitative PCR amplicon greater than 200 nt from the reverse transcription priming site, would underestimate the abundance of that target cDNA.

One of the most crucial steps in the MPE-seq protocol is the 1^st strand synthesis reaction. We recommend that this be the starting point in any optimization/troubleshooting. As previously noted, an appropriate reaction temperature for reverse transcription is crucial for maximizing cDNA yield as well as minimizing the abundance of off-target priming events. cDNA synthesis efficiency can be assayed immediately after the 1^st strand synthesis reaction. From this point, the efficiency of cDNA synthesis can be determined, and a starting number of cDNA molecules can be discerned using qPCR. Quantitative PCR can then be performed at any point downstream and compared to the initial cDNA synthesis values to assess the efficiency of each step in the library prep process. Off target priming events are best analyzed through sequencing test libraries and counting off target alignments. For a typical MPE-seq library from S. cerevisiae RNA, we expect to see ~10-20% of reads mapping to off target sites.

Before performing the final library PCR amplification, SYBR based qPCR is performed to determine the appropriate number of amplification cycles. This step is important for minimizing PCR duplicates but also gives an indication of the quality of the library prep. The threshold cycle (C_t) is proportional to the abundance of sequenceable molecules present in the library. Any erroneous step in the protocol could lead to a high C_t value and would be indicative of a poor library. For example, poor cDNA synthesis efficiency would lead to a high C_t value. In our experience, S. cerevisiae libraries with a Ct of less than 18 will result in quality datasets, although this value may differ depending on the qPCR machine/software and assay design. Libraries that require more amplification than this are likely to contain a large proportion of primer dimers made from unextended reverse transcription primers which were not removed prior to the 1^st strand extension step of library prep. Another similar indicator of overall library quality is the number of unique molecules sequenced, which is discerned from the number of unique UMIs for each target. A low number of UMIs indicates a low number of sequenceable molecules and/or PCR over amplification.

3.3. MPE-seq Method

3.3.1. Required Equipment

Table 1.

Equipment required for MPE-seq

Equipment	Description	Cat #
Thermo-cycler
Zymo-5 columns	Fiberglass columns for nucleic acid purification from Zymo Research	D4013
Magnetic Stand	Preferably 2 stands: one that can handle both 1.5 mL and 2.0 mL Eppendorf tubes and one that can handle 0.2 mL PCR tubes.
Rotator	For bead binding incubations
RT-PCR Machine	For determining library number of amplification cycles
Qubit	For library quantification and pooling
Acrylamide gel apparatus	For library size selection
Eppendorf tubes	2.0 mL, 1.5 mL, 0.5 mL
PCR tubes	Strips or plates for thermocycler
Centrifuge	For Eppendorf and PCR tubes
Bioanalyzer or Fragment analyzer	Library size distribution and pooling
Bioinformatics capable machine or server access	Access to appropriate software (e.g. Quality assessment, trimming, alignment, feature counting, etc.)

3.3.2. Materials and Reagents

Table 2.

Materials & reagents required for streptavidin bead purification

Reagent	Stock Concentration	Vendor	Cat #
Dyna Beads MyOne Streptavidin C1	10 mg/mL	ThermoFisher	65001
2x Bind and Wash Buffer	10 mM Tris-HCl pH 7.5, 1 mM EDTA, 2 M NaCl
1x Bind and Wash Buffer	5 mM Tris-HCl pH 7.5, 0.5 mM EDTA, 1 M NaCl
1x SSC	150 mM NaCl, 15 mM Sodium Citrate
Denaturing Solution	0.1 M NaOH
TE Buffer	10 mM Tris pH 7.5, 1 mM EDTA
Bead Elution Buffer	95% Formamide, 10 mM EDTA, pH 8.2

Table 3.

Materials & reagents required for nucleic acid column purification

Reagent	Stock Concentration	Vendor	Cat #
Binding Buffer	2 M Guanidinium-HCl, 75% Isopropanol
Washing Buffer	10 mM TRIS pH 8.0, 80% Ethanol

Table 4.

Material & reagents required for amplification and processing of array synthesized 1^st strand synthesis primers

Reagent	Stock Concentration	Vendor	Cat #
PCR Primers (oHX093 & oHX094)*	100 μM & 10 μM	IDT
Phusion polymerase	100x	NEB	M0530S
Phusion buffer	5x	NEB	M0530S
Array synthesized oligos		LC Sciences
DMSO	100%
dNTP mix	10 mM each (dATP, dGTP, dCTP, dTTP)
Sap1 restriction enzyme		NEB	R0569S
Cutsmart buffer	10x	NEB	R0569S
Lambda exonuclease		NEB	M0262S
Lambda exonuclease buffer	10x	NEB	M0262S
Isopropanol	100%
Ethanol	70%
Sodium Acetate	3 M, pH 5.3
Zymo DNA binding buffer		Zymo Research	D4003-1-L

^*Primer sequences can be found in Table 12

Table 5.

Materials & reagents required for RNA fragmentation

Reagent	Stock Concentration	Vendor	Cat #
10x Fragmentation Buffer	100 mM ZnCl2, 100 mM Tris-HCl pH 7.0
Fragmentation stop buffer	0.5 M EGTA

Table 6.

Materials & reagents required for 1^st strand synthesis

Reagent	Stock Concentration	Vendor	Cat #
SSIV RT Buffer	5x	ThermoFisher	18090050
Superscript IV		ThermoFisher	18090050
1st strand primer pool	Each primer in S. cerevisiae pool at ~80 nM (309 primers). For S. pombe each primer at ~2.5 nM (3918 primers)	IDT
DTT	0.1 M
Individual dNTPs	100 mM	Sigma-Aldrich
5-(3-Aminoallyl)-dUTP	50 mM	ThermoFisher (Ambion)	AM8439
Aminoallyl-dNTP mix	10 mM dATP, dGTP, dCTP, 6 mM dTTP, 4mM Aminoallyl-dUTP	Mixed from above components
DEPC Water

Table 7.

Materials & reagents required for RNA hydrolysis

Reagent	Stock Concentration	Vendor	Cat #
RNA Hydrolysis Solution	0.3 M NaOH, 0.03 M EDTA
Neutralization Solution	0.3 M HCl

Table 8.

Materials & reagents required for biotin coupling

Reagent	Stock Concentration	Vendor	Cat #
Sodium Bicarbonate	1.0 M pH 9.0
EZ-link NHS-Biotin	Diluted to 300 nmol/μL in DMSO	ThermoFisher	20217

Table 9.

Materials & reagents required for 1^st strand extension

Reagent	Stock Concentration	Vendor	Cat #
NEB Buffer 2	10x	NEB	M0212L
Klenow exo-	5000 units/mL	NEB	M0212L
dNTPs	10 mM each (dATP, dGTP, dCTP, dTTP)
1st Strand Extension Primer	100 μM	IDT

Table 10.

Materials & reagents required for library qPCR and amplification

Reagent	Stock Concentration	Vendor	Cat #
Phusion Buffer	5x	NEB	M0530S
Phusion Polymerase	100x	NEB	M0530S
dNTPs	10 mM each (dATP, dGTP, dCTP, dTTP)
DMSO	100%
Nextera i5 primer*	10 μM	IDT
Nextera i7 primer*	10 μM	IDT
SyBr Green	100x

^*Nextera i5 and i7 library amplification and barcoding sequences can be found in Illumina documentation

Table 11.

Materials & reagents required for gel purification

Reagent	Stock Concentration	Vendor	Cat #
Acrylamide/Bis	29:1
10X TBE	1 M Tris, 1 M Boric acid, 0.02 M EDTA
1x TBE	100 mM Tris, 100 mM Boric acid, 2 mM EDTA
Ammonium persulfate (APS)	10%
TEMED
SyBr Gold	10,000x
100 bp Ladder		NEB	N3231S
Gel Extraction Solution	0.3 M Sodium Acetate
Glycogen	20 mg/mL
Isopropanol	100%
AmpureXP beads		Beckman	A63880
Ethanol	70%

Table 12.

Oligonucleotide sequences

Oligo Name	Description	Sequence 5’ – 3’
YBL092W	Example array-based synthesized RT primer	TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGNNNNNNNGACAATCTTTGGGTGAGGTAAGGACCTCGAAGAGCATTACGGCTCCTCGCTGCAG*
OHX093	Array-based oligo pool amplification primer forward	/5SpC3/ TCGTCGGCAGCGTCAGATGTGTATAAGA
OHX094	Array-based oligo pool amplification primer reverse	/5Bioag/ CTGCAGCGAGGAGCCGTAATGC
OJP788	1st strand extension oligo (dN9)	GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGNNNNNNNNN /3c6/
YBL092W	Example individually synthesized RT primer	TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGNNNNNNNGACAATCTTTGGGTGAGGTAAGGA*

^*Gene specific regions of RT oligonucleotides are bolded

3.3.3. General Protocols:

Below are 2 general protocols that are repeated several times throughout the library preparation. They are listed here for reference.

3.3.3.1. Column clean-up

1. Add 7 volumes of binding buffer to each sample and mix well

2. Transfer each sample to a zymo-5 column placed inside a collection tube

3. Spin samples at 14K RPM for 1 min

4. Discard flow through and place zymo column back in collection tube

5. Wash columns by adding 700 μL washing buffer to each column

6. Spin for 1 min at 14k RPM

7. Discard flow through

8. Repeat washing for a total of 2 washes

9. Dry column by spinning for an additional 1 min at 14k RPM

10. Add appropriate elution buffer volume to each column and incubate for ~1-2 min at room temperature

11. Place columns in labeled collection tubes and spin at 14K RPM for 1 min

12. You now have purified nucleic acid

3.3.3.2. Bead purification

1. Vortex beads vigorously to resuspend

2. Transfer 20 μL of beads per sample to be purified into a 1.5 mL Eppendorf tube

3. Place tube on magnetic stand and incubate for 1 min

4. While leaving tube on magnetic stand, aspirate buffer

5. Remove tube from magnetic stand

6. Wash beads by adding 500 μL of 1x bind and wash buffer to tube containing the beads and mix well with a pipette

7. Place tube on magnetic stand and incubate for 1 min

8. Aspirate buffer

9. Repeat wash for a total of 2 washes

10. Resuspend beads in 50 μL of 2x bind and wash buffer per sample to be purified

11. Mix beads well with a pipette

12. Aliquot 50 μL of beads into a new tube. 1 for each sample to be purified

13. Add each 50 μL sample to an aliquot of washed beads

14. Place on rotator and incubate at room temperature for 30 min

15. Place tubes on magnetic stand and incubate for 1 min

16. Aspirate buffer

17. Remove samples from magnetic stand

18. Add 500 μL of 1x bind and wash buffer and mix well with pipette

19. Place on magnetic stand and incubate for 1 min

20. aspirate buffer

21. repeat washing for a total of 2 washes

22. Wash each sample with 100 μL 1x SSC, mixing well by pipetting

23. Place on magnetic stand for 1 min

24. Aspirate buffer

25. Remove samples from magnetic stand

26. Add 100 μL of denaturing solution to each sample

27. Incubate each sample for 10 min at room temperature

28. Place samples on magnetic stand for 1 min and aspirate buffer

29. Wash by adding 100 μL of denaturation solution to each sample. Mix well by pipetting

30. Place samples on magnetic stand for 1 min and aspirate buffer

31. Add 100 μL of TE buffer and mix well with pipette

32. Place on magnetic stand and incubate for 1 min

33. aspirate buffer

34. Repeat TE buffer wash 2 times for a total of 3 washes

35. Add 100 μL of bead elution buffer to each sample

36. Incubate samples at 90°C for 2 min

37. Immediately place samples on magnetic stand and incubate for 1

38. Aspirate each sample and transfer to a new tube

3.3.4. Complex oligo array amplification

3.3.4.1. 1^st PCR Amplification

1. Perform a SYBR based qPCR experiment to establish the number of PCR cycles to perform to prevent the reaction from plateauing.

2. Mix the qPCR mix detailed in Table 13

   Table 13.

   qPCR PCR amplification reaction mix

   Reagent	Volume (μL)
   Phusion buffer	10
   dNTP mix	1
   OHX093 (10 μM)	1.25
   OHX094 (10 μM)	1.25
   LC oligo mix	0.125% total mass
   DMSO	0.5
   SyBr Green	5
   Phusion	0.5
   H20	Up to 50

3. Aliquot 15μL of qPCR reaction mix into 3 wells of a qPCR plate

4. Run the following cycle program:

   • Initial Denaturation

     1. 98°C 30 seconds
   • Amplification (40x)

     1. 98°C 10 seconds
     2. 60°C 20 seconds
     3. 72°C 30 seconds
   • Final extension

     1. 72°C 3 minutes

5. Chose the cycle number closest (but prior) to the plateau of the amplification curve for all 3 samples. Use this cycle number for the 1^st PCR amplification.

6. Mix the PCR mix detailed in Table 14

   Table 14.

   1^st PCR amplification reaction mix

   Reagent	Volume (μL)
   Phusion buffer	80
   dNTP mix	8
   OHX093 (10 μM)	10
   OHX094 (10 μM)	10
   LC oligo mix	1% mass*
   DMSO	4
   Phusion	4
   H20	Up to 400

   ^*

   We chose to use 1% of the total oligo mass to maximize the amount of RT primer generated from an individual synthesis. This amount can be optimized by the user.
7. Run the following cycle program:

   • Initial Denaturation

     1. 98°C 30 seconds
   • Amplification (Cycle number determined in step 6)

     1. 98°C 10 seconds
     2. 60°C 20 seconds
     3. 72°C 30 seconds
   • Final extension

     1. 72°C 3 minutes

8. Repeat steps 1-6 using 0.5 μL of 1^st PCR reaction as template to determine amplification cycle number for 2^nd PCR

3.3.4.2. 2^nd PCR Amplification

1. Prepare a 2^nd PCR mix using the product from the 1^st PCR as the template. Mix the reaction according to Table 15

   Table 15.

   Reaction mix for 2^nd PCR amplification

   Reagent	Volume (μL)
   Phusion buffer	8000
   dNTP mix	800
   OHX093 (100 μM)	100
   OHX094 (100 μM)	100
   Product from 1st PCR	400
   DMSO	400
   Phusion	400
   H20	29800

2. Aliquot 100 μL volumes of the reaction mix into four 96 well plates and run the following cycling program:

   • Initial Denaturation

     1. 98°C 30 seconds
   • Amplification (Cycle number determined in step 9 of 3.3.4.1)

     1. 98°C 10 seconds
     2. 60°C 20 seconds
     3. 72°C 30 seconds
   • Final extension

     1. 72°C 3 minutes

3.3.4.3. Concentrate DNA

1. Pool the reactions and split equal volumes into 50 mL falcon tubes

2. Isopropanol precipitate each sample by adding a 1/10^th volume of 3 M sodium acetate followed by an equal volume of isopropanol

3. Let sample precipitate at room temperature for ~30 minutes

4. Spin in a capable centrifuge at max speed for 1 hour

5. Wash pellets twice with 70% ethanol centrifuging at max speed for 5 min in between washes

6. Allow pellet to dry

7. Dissolve each pellet in 700 μL H₂O

8. Pool samples together and isopropanol precipitate again

9. Dissolve the pellet in 300 μL H₂O

10. Perform column purification on sample using 1500 μL of Zymo DNA binding buffer Once binding buffer is added split sample equally into 4 Zymo-25 columns and follow column purification protocol from section 3.3.3.1 starting at step 2.

11. Elute from each column with 105 μL H₂O

12. Combine eluents

3.3.4.4. SapI Digestion

1. Add 50 μL 10X smart cut buffer

2. Add 30 μL SapI enzyme.

3. Incubate sample at 37°C for 15 hours

4. Isopropanol precipitate sample

5. Resuspend pellet in 125 μL H₂O

3.3.4.5. Lambda Exonuclease Digestion

1. Add 15 μL 10X lambda exo- buffer to sample

2. Add 10 μL lambda exo-

3. Digest at 37°C for 2 hours

4. Total volume is 150 μL.

5. Perform column purification of sample. Once binding buffer is added, split the sample equally into 2 Zymo-25 columns

6. Elute DNA from each column by adding 25 μL H₂O and combine eluants

3.3.4.6. Streptavidin Bead Purification

1. Vortex beads vigorously to resuspend

2. Transfer 50 μL of beads per sample to be purified into a 1.5 mL Eppendorf tube

3. Place tube on magnetic stand and incubate for 1 min

4. While leaving tube on magnetic stand, aspirate buffer

5. Remove tube from magnetic stand

6. Wash beads by adding 500 μL of 1x bind and wash buffer to tube containing the beads and mix well with a pipette

7. Place tube on magnetic stand and incubate for 1 min

8. Aspirate buffer

9. Repeat wash for a total of 2 washes

10. Resuspend the beads in 50 μL 2x bind and wash buffer

11. Add the sample and mix well with a pipette

12. Place sample on a tube rotator and incubate sample for 15 min at room temperature

13. Heat sample to 65°C and incubate for 2 min

14. Place tube on magnetic stand and allow beads to separate for 1 min

15. Aspirate supernatant and place in a new Eppendorf tube. This is the purified pool of RT primers

16. Isopropanol precipitate samples and dissolve pellet in 30 μL H₂O

17. Check DNA concentration via nano-drop or Qubit

    Note: For S. pombe, the purified array synthesized primers were diluted to roughly 100 ng/μL

3.3.5. 1^st strand synthesis

Note: The 1^st strand synthesis protocol detailed here is for an RT reaction temperature of 55°C. See section 3.2 for design considerations.

1. Make a primer/template mix for each sample as detailed in Table 16

   Table 16.

   Reaction mix for 1^st strand synthesis primer annealing

   Reagent	Volume (μL)
   5X RT buffer	4
   1st strand primer pool	2
   Total RNA	10 μg
   DEPC Water	Up to 20
   total	20

2. Place samples in thermo-cycler and run the following cycle:

   1. 70°C 1 minutes
   2. 65°C 5 minutes
   3. 55°C Hold
3. Make the enzyme dNTP mix for each sample as detailed in Table 17

   Table 17.

   Reaction mix for 1^st strand synthesis

   Reagent	Volume (μL)
   5X RT buffer	4
   DTT	2
   Aminoallyl-dNTP mix	2
   SSIV	2
   DEPC Water	Up to 20
   Total	20

4. Heat dNTP Enzyme mix to 55°C by placing it in the thermo-cycler containing the template/primer mix

5. Directly add 20 μL dNTP/enzyme mix to the template/primer mix, ensuring that both samples are kept at 55°C

6. Allow the 1^st strand synthesis reaction to proceed for 10 minutes at 55°C

7. Incubate sample at 80°C for 10 minutes to inactivate the enzyme

   Note: Sample volume is now 40 μL

3.3.6. RNA Hydrolysis

1. Add 20 μL of RNA hydrolysis solution to each sample

2. Incubate at 65°C for 15 min

3. Add 20 μL of neutralization solution to each sample

   Note: Sample volume is now 80 μL

4. Perform Column clean up as detailed in the protocol above

5. Elute each sample with 16 μL H₂O

3.3.7. Biotin coupling

1. Add 2 μL of sodium bicarbonate to each sample

2. Add 2 μL of NHS-biotin to each sample

3. Incubate at 65°C for 1 hour in the dark

   Note: Perform the next few steps quickly to prevent biotin precipitation

4. Briefly spin tubes in a centrifuge

5. Add 40 μL H₂0

   Note: Sample volume is now 60 μL

6. Perform column cleanup

7. Elute in 50 μL H₂0

8. Perform streptavidin bead purification on each sample

9. Perform column purification and elute in 40 μL H₂O

10. You now have purified 1^st strand cDNA

11. The protocol can be paused here by storing the samples at −20°C

3.3.8. 1^st strand extension

1. Prepare the following reaction mix described in Table 18 for each sample

   Table 18.

   Reaction mix for 1^st strand extension

   Reagent	Volume (μL)
   10X NEB Buffer 2	5
   10mM dNTP Mix	1
   cDNAfrom previous	40
   1st Strand Extension Primer	1
   Total	47

2. Incubate each sample at 65° C for 2 min

3. Cool samples to room temperature by placing on bench top for ~5 min

4. Add 3 μL of Klenow exo- enzyme to each sample

5. Incubate samples at room temperature for 5 min

6. Heat samples to 37° C and incubate for 30 min

7. Heat samples to 75° C and incubate for 20 min to inactivate enzyme

8. Perform bead purification on each sample

9. Perform column purification on each sample eluting with 33 μL H₂O

10. You now have a library ready for PCR amplification

11. This is also a safe stopping point. Samples can be stored at −20° C

12. Continue to library amplification

3.3.9. PCR Amplification

3.3.9.1. qPCR

1. Perform triplicate qPCR reactions on each sample by mixing the qPCR reaction mix detailed in Table 19

   Table 19.

   qPCR reaction mix

   Reagent	Volume (μL)
   5x Phusion Buffer	10
   dNTP Mix	1
   1st Strand Extension Product	2
   Nextera i5 Primer	2.5
   Nextera i7 Primer	2.5
   Phusion Polymerase	0.5
   DMSO	0.5
   SyBr Green	0.5
   H2O	30.5

2. Aliquot 15 μL of each qPCR reaction mix into 3 separate wells of a qPCR plate

3. Perform the following qPCR cycling:

   • Denaturation:

     1. 98 °C 30 seconds
   • Amplification (40x):

     1. 98 °C 10 seconds
     2. 62 °C 20 seconds
     3. 72 °C 30 seconds
   • Final elongation:

     1. 72 °C 3 minutes

4. After the qPCR calculate the C_t value for each sample and average the 3 values for each library

5. Calculate the number of amplification cycles to perform by averaging the 3 replicate Ct values for each library, adding 3 and rounding to the nearest whole number

   Note: The number of PCR cycles can be reduced by 2-3 if bead purification is performed instead of gel purification as this results in more efficiency purification

3.3.9.2. PCR Amplification

1. For each sample, prepare the following PCR reaction mix detailed in Table 20

   Table 20.

   Reaction mix for library PCR amplification

   Reagent	Volume (μL)
   5x Phusion Buffer	10
   dNTP Mix	1
   1st Strand Extension Product	10
   Nextera i5 Primer	2.5
   Nextera i7 Primer	2.5
   Phusion Polymerase	0.5
   DMSO	0.5
   H2O	23

2. Run the following cycling conditions on each PCR reaction mix:

   • Denaturation:

     1. 98 °C 30 seconds
   • Amplification (Cycle # determined in step 5 of qPCR):

     1. 98 °C 10 seconds
     2. 62 °C 20 seconds
     3. 72 °C 30 seconds
   • Final elongation:

     1. 72 °C 3 minutes
     2. 4 °C Hold

3.3.10. Size selection, clean-up, & quality check

1. Prepare a 6% native acrylamide gel

2. The entirety of each PCR reaction will be run in a single well so make sure the wells can comfortably hold 60 μL (50 μL sample + 10 μL loading dye)

3. Run a 100 bp ladder for sizing.

4. Once samples are loaded on the gel, run at 250V for 70min.

5. Stain in 1x SyBr Gold for 15 min

6. Image gel for documentation

7. For each sample, cut out the lane from 200 bp to 800 bp and place in a 2 mL tube

8. Add 900 μL of 0.3 M NaOAc to each sample and place tubes on rocker overnight.

9. Added 900 μL isopropanol to precipitate.

10. Add 1 μL glycogen to each sample

11. Allow sample to precipitate.

12. Spin at 14k RPM for 30 min

13. Wash with 500 μL of 70% ethanol twice

14. Dry pellet and resuspend in 40 μL H₂O

15. Assay DNA concentration using Qubit

16. It is also helpful to run the samples on a Bioanalyzer or similar instrument to check the size distribution of the library

17. Pool libraries for sequencing

    Note: The gel running conditions will likely vary between equipment and so careful attention should paid at first. The goal is to get as much separation as possible. We run the gel until the 100 bp ladder band is ~1-2 cm from the bottom of the gel. After purification, library concentration is generally somewhere between 0.5 ng/μL to 5 ng/μL. Figure 3A shows a Bioanalyzer trace of an S. cerevisiae library after size selection with a gel (Fig. 3A).

Fig. 3.

Bioanalyzer traces of replicate S. cerevisiae MPE-seq libraries after size selection by: (A) acrylamide gel, or (B) AmpureXP beads. LM and UM denote the lower and upper markers, respectively. PD denotes the peak resulting from primer dimer molecules.

Alternative Size Selection using AmpureXP Beads

It may be preferable to purify libraries using AmpureXP beads or an equivalent. The purpose of gel sizing is 2-fold: (1) It removes excess PCR primers and, (2) It removes primer dimers that run at ~180 bp. Depending on the user’s design (e.g. number of targets, input RNA, target abundance, etc.) primer dimers may make up a very small portion of the library and not need to be excluded. This can be determined by the user. One downside to gel sizing worth considering is it may introduce a bias against short cDNAs derived from features of interest such as lariat intermediates. Figure 3B shows a Bioanalyzer trace of an S. cerevisiae library with size selection via AmpureXP beads (Fig. 3B).

3.4. Data Analysis

In the following sections we describe a general pipeline for MPE-seq data analysis and suggest publicly available software packages for each step.

3.4.1. Read processing and quality control

The first step in MPE-seq data processing is to de-multiplex reads into their respective samples. This is achieved by parsing reads into fastq files based on their respective barcodes. Depending on the sequencing platform and center, this processing may be provided before the data are returned to the user. Next, we suggest assessing the overall quality of the reads, which can be done with tools like FastQC [21]. The output of this software provides overall quality scores, read lengths, adapter dimer content, and sequence duplication levels. The next step in processing is compressing reads resulting from PCR duplications into a single read. For this step, the entirety of each sequence, which contains the UMI and the rest of the read, is considered. If multiple reads have identical sequences, they are considered PCR duplicates and compressed into a single read for counting purposes [17]. For this processing we use custom python scripts that search for reads with identical sequences within a fastq file and generate a new fastq file containing those unique reads (see Code Availability section). After this step, the reads are trimmed to remove any regions that correspond to the Illumina-based sequencing adapters. This step can be accomplished using publicly available software such as Trimmomatic [21]. In addition, the UMI sequence is also trimmed in this step to prevent mapping artefacts. It may be useful to move the trimmed UMI to the sequence name using a custom script so that it can be accessed in the SAM file after alignment. An alternative approach is to clip the UMI sequence during alignment.

3.4.2. Alignment

Properly trimmed reads can be aligned to a reference genome using a variety of splicing aware aligners such as HISAT [22] or STAR [23]. There are a few important parameters to consider when aligning these reads. First, the minimum alignment length on either side of a splice junction, also known as anchor length, should be specified. We suggest an anchor length of 3, as we have found this best reduces erroneous mappings. Second, we generally include in the output unmapped and multimapping reads so they can be used to quantify alignments resulting from off-target priming events during cDNA synthesis, which is useful for trouble shooting and optimization. Third, reads resulting from primers that were unextended by reverse transcriptase during 1^st strand synthesis and were unsuccessfully removed during library prep purifications can generate erroneous alignment counts and should be removed from the analysis. These reads will contain the gene specific regions used for reverse transcription and will align once adapter sequences have been trimmed. Additionally, a small number of bases can be added during 1^st strand extension due to an overhang of the dN9 region of the 1^st strand extension primer hybridized to the reverse transcription primer. For this reason, in some cases, the aligned region of these reads may overlap with a splice junction and result in erroneous counts. These reads can either be filtered out during or post alignment by requiring an insert size of >30 bases. This can be accomplished with alignment parameters or by post alignment filtering with custom scripts.

3.4.3. Read allocation and isoform quantitation

After reads have been aligned, the abundance of each target isoform can be quantified. Figure 2C shows an example of how paired end sequencing alignments can be partitioned into 6 categories (labeled C1 to C6 on Fig. 2C) which support each isoform for a single splicing event (Fig. 2C). The abundance of Spliced (S), Unspliced (U), Pre-first step (P), and Lariat intermediate (L) isoforms can then be calculated as follows:

$$S=C_1+C_2$$

$$U=C_3+C_4+C_5+C_6$$

$$P=C_3+\left((C_4+C_5)\left(\frac{C_3}{C_6}\right)\right)$$

$$L=C_6+\left((C_4+C_5)\left(\frac{C_6}{C_3}\right)\right)$$

To count the number of s isoforms for a target, the CIGAR string in SAM/BAM alignment files can be utilized. Alignments which contain a gap due to splicing will contain an ‘N’ in the CIGAR string, from which the positions of the spliced intron can be discerned. The abundance of S isoforms can then be counted as the number of alignments that contain an ‘N’ in the CIGAR string corresponding to a gap resulting from the excised intron (C₁, C₂) (Fig. 2C). U isoforms will not contain an ‘N’ in the CIGAR string and will map within the intron (C₃ - C₆) (Fig. 2C). By utilizing paired end reads, MPE-seq affords the ability to differentiate unspliced isoforms further into those that have not undergone the 1^st step of splicing from those that have undergone the 1^st step of splicing but not the 2^nd step of splicing [11]. P and L isoforms can be quantified by considering the mapping location of the first base of the second read of paired-end reads (Fig. 2C). cDNA synthesis will terminate at the branch point adenosine of lariat intermediates due to the inability of reverse transcriptase to process through the branched structure. If the first base of a second read maps within +/− 4 bases from the branch point adenosine, the read pair is unambiguously counted as a lariat intermediate (C₆) (Fig. 2C). A small window around the branch point adenosine is suggested due to our observation that second read ends don’t pile up precisely at annotated branch point adenosines [11]. All read pairs which have a second read mapping location upstream of the branch point window are unambiguously counted as pre-first step (C₃) (Fig. 2C). Read pairs with a second read that maps downstream of the branch point window (C₅) or does not map but has a first read that maps within the intron (C₄) are considered ambiguous and are assigned to lariat intermediate or pre-first step based on the proportion of unambiguous lariat intermediates to pre-first step read pairs. This relationship is shown in the equations above. Each MPE-seq read corresponds to a single reverse transcription priming event on a single transcript, precluding the necessity for length normalization. As a result, counts can be used directly to calculate splicing efficiency as the user sees fit. Data generated in this way can subsequently be compared between experimental conditions using appropriate statistical methods.

The simple quantification method detailed above assumes that the frequency of random reverse transcription termination events and RNA fragmentation that occur within the branch point window is minimal. It also assumes that reverse transcriptase reads through lariat branches at a very low frequency. It is important to note that this quantification also requires a priori knowledge of the position of the branch point adenosine and may not be possible in organisms with degenerate or non-annotated branch points.

As noted earlier, MPE-seq can be used to quantify the frequency of canonical as well as alternative splice isoforms, such as those generated by exon skipping and alternative 3’ and 5’ splice site usage, provided that these events are located within the sequencing window of a targeted region. To quantify these events, the CIGAR string of BAM/SAM files can be utilized. Alignments which contain junctions or gaps denoted by an ‘N’ in the CIGAR string at the coordinates of specific alternative events can be filtered and counted. Some aligners, such as STAR, generate a file containing the genomic positions and counts of all splice junctions identified, removing the need for custom scripts or additional software packages to count these events [23]. These counts can then be compared to quantify the frequency of specific alternative splice isoforms.

A variety of counting programs are available that can be utilized with MPE-seq data such as HTSeq, featureCounts, and Bedtools [24-26]. However, these software packages are designed to count reads that overlap specific features (e.g. genes) and do not differentiate isoforms within a feature (e.g. spliced vs unspliced). For that reason, some combination of data filtering, modification of counting software, and/or use of custom scripts is required for properly counting and analyzing MPE-seq data.

3.5. Conclusions

The study of RNA splicing has been greatly enhanced by the adaptation of next-generation sequencing. RNA-seq has been pivotal in genome wide analysis of the transcriptome, leading to the identification of countless numbers of novel splicing isoforms. While RNA-seq has been crucial for the analysis of transcript abundance, the unbiased nature of cDNA generation in most library generation methods often results in poor sampling of splicing informative loci. By targeting cDNA synthesis to splicing informative loci of choice, MPE-seq provides unique opportunities to measure splicing efficiency with high precision and to detect splice isoforms that are otherwise poorly sampled with standard RNA-seq. It is important to note that the targeted nature of MPE-seq renders it poor at identifying novel splicing events outside of targeted regions. Nevertheless, MPE-seq provides the unique ability to detect lariat intermediates, which allows for the analysis of the efficiencies of the 1^st and 2^nd steps of splicing, a level of resolution that is unavailable with standard RNA-seq. Moreover, the increase in read depth afforded by MPE-seq has the additional advantage of greatly reducing sequencing costs and allows for multiplexing of many samples on a single sequencing lane. Beyond the analysis of steady-state total RNA, MPE-seq can readily be applied to study splicing efficiency in a variety of biological contexts through use of fractionated RNA sources such as, poly(A)⁺ selected RNA, polysome associated RNA, nuclear/cytoplasmic RNA, and nascent RNA purified with metabolic labels.

Code availability:

Details of all data analysis associated with this paper can be found at https://github.com/zdwyer/MPE-seq-methods.

Data Availability:

Raw sequencing data are available through Gene Expression Omnibus (GEO accession: GSE126583)

• RNA-seq has enabled genome-wide studies of splicing mechanism and regulation.

• Standard RNA-seq approaches poorly sample splicing informative loci.

• MPE-Seq targets reads to splicing informative loci, increasing precision.

• MPE-seq enables detection and quantitation of rare pre-mRNA splicing intermediates.