Overview of High Throughput Sequencing Technologies to Elucidate Molecular Pathways in Cardiovascular Diseases

Abstract

High throughput sequencing technologies have become essential in studies on genomics, epigenomics, and transcriptomics. While sequencing information has traditionally been elucidated using a low throughput technique called Sanger sequencing, high throughput sequencing (HTS) technologies are capable of sequencing multiple DNA molecules in parallel, enabling hundreds of millions of DNA molecules to be sequenced at a time. This advantage allows HTS to be used to create large data sets, generating more comprehensive insights into the cellular genomic and transcriptomic signatures of various diseases and developmental stages. Within HTS technologies, whole exome sequencing can be used to identify novel variants and other mutations that may underlie many genetic cardiac disorders, whereas RNA sequencing (RNA-seq) can be used to analyze how the transcriptome changes. Chromatin immunoprecipitation sequencing (ChIP-seq) and methylation sequencing (Methyl-seq) can be used to identify epigenetic changes whereas ribosome sequencing (Ribo-seq) can be used to determine which mRNA transcripts are actively being translated. In this review, we will outline the differences in various sequencing modalities and examine the main sequencing platforms on the market in terms of their relative read depths, speeds, and costs. Lastly, we will discuss the development of future sequencing platforms and how these new technologies may improve upon current sequencing platforms. Ultimately, these sequencing technologies will be instrumental in further delineating how the cardiovascular system develops and how perturbations in DNA and RNA can lead to cardiovascular disease.

Introduction

Until the discovery of retroviruses, the central dogma of molecular biology stated that genes are transcribed to make RNA and in turn RNA is translated into protein ^{1, 2}. This dogma outlines how the variable expression of genes can dynamically control cellular functionality and identity from a single genome. Gene expression is dynamically controlled and variations in transcription and translation can result in major functional changes within the cell. If the underlying DNA sequence is mutated, or if the downstream message is changed during transcription and translation, cellular function may be compromised, leading to various disease pathologies. While our environment affects the manifestation of disease, many diseases also have a strong underlying genetic component. Diseases that have a stronger genetic component rather than environmental component may include those that surface at birth (congenital diseases) and those which run in families (familial inheritance diseases). To determine how one’s genetic background contributes to disease, a large collection of genomic and transcriptomic data sets will be required. By sequencing multiple genomes, it is therefore possible to evaluate human genomic diversity, as demonstrated by the 1000 Genomes Project ^{3, 4}. In addition, the ENCODE ⁵ and HapMap project ⁶ have employed many high throughput sequencing (HTS) applications outlined below to understand the functional attributes of each region of the genome. With advancements in HTS technologies, sequencing costs have now dramatically decreased and it may soon be possible to sequence the entire human genome for $1,000 or less ⁷. As the price of sequencing decreases, sequencing may become commonplace, which will vastly contribute our understanding on genomic variability and how this variability may increase one’s susceptibility to develop cardiovascular diseases. Ultimately, by lowering sequencing cost and in turn, making sequencing technologies mainstream, implementation of HTS technologies will be invaluable in determining the molecular pathways involved in cardiovascular development and disease.

First Generation Sanger Sequencing

DNA sequencing information has traditionally been elucidated using Sanger sequencing ⁸. This technique was developed by Dr. Frederick Sanger, who was subsequently awarded the 1980 Nobel prize in Chemistry ⁹. In this method, a complementary strand of DNA is made from the input template DNA from a mixture of 2′-deoxynucleotides (dNTPs) including 2′,3′-dideoxynucleotides (ddNTPs), which are labeled with fluorescent dyes ⁸. ddNTPs are nucleotides that lack a 3'-OH group required for cDNA elongation and when a ddNTP is incorporated into the elongating DNA strand, elongation is terminated, resulting in the generation of multiple DNA fragment sizes. The sizes of these fragments are separated using single base-pair resolution capillary electrophoresis, resulting in an electropheragram that appears to be a direct read-out of the nucleotide sequence from the original template molecule ¹⁰. Sanger sequencing can have an average read length of 800 base pairs, but it is limited by the amount of DNA that could be processed at a given time. To address the low throughput, newer sequencing technologies were developed that could read the sequence of multiple DNA molecules in parallel. Parallel capillary systems greatly increased the throughput of the number of DNA strands that could be analyzed ^{11, 12}, because 1–6 megabases of DNA sequence could be acquired per day in a standard 96-capillary instrument ¹³. However, parallel capillary based systems are still limited by the amount of capillary columns which could be processed at a given time. Because the human genome consists of approximately 3 gigabases, containing approximately 20,000 genes that span 45 megabases (1.5% of the whole genome) ^14–16, it took the Human Genome Project over a decade and billions of dollars to complete using Sanger sequencing ^{17, 18}.

Key HTS Platforms

Commercially available sequencing platforms are expanding the potential of sequencing by exponentially increasing the throughput of their technologies. While many sequencing platforms are available, Illumina’s platforms (http://www.illumina.com/) have dominated much of the sequencing industry (Figure 1) ¹⁹. Illumina’s bridge amplification method allows for generation of small “clusters” with an identical sequence to be analyzed. Clusters formed on an Illumina flow cell create multiple primer hybridization steps allowing multiple sequencing start points. This allows the sequencing of both ends of the original template molecule known as paired-end sequencing. Sequencing information from paired-end reads play an important role in Illumina’s technology by increasing the output from a sequencing run, identifying splice variants in RNA-seq, and to deduplicate (remove duplicate copy) reads originating from the same original template molecule. Paired end reads are also important for identifying large structural variants such as inversions from whole genome sequencing, which are unnoticed with short sequencing techniques. A third read may also be used to separate out samples as long as each sample in the sequencing library had a unique barcode read engineered into the adapter construction as a third separate read. In Illumina sequencing however, all four nucleotides are available during incorporation which can lead to an overall substitution error rate of 0.11% ²⁰.

Figure 1. Overview of DNA-sequencing using the Illumina platform.

In next generation DNA sequencing, DNA is first fragmented into smaller input-sized fragments by enzymes or by sonication. The ends of these fragments are repaired and specific adapters are ligated to the ends of the fragments, allowing hybridization to a flow cell to occur. A bridge amplification step is performed to create a “cluster” of fragments with the same sequence. One stand of DNA is removed and fluorescently labeled nucleotides are passed by each cluster. An image of the flow cells is recorded for the first cycle and a computer processes which nucleotide was incorporated at each cluster’s co-ordinates. The fluorescent label is cleaved and a second round of fluorescently labeled nucleotides is passed by each cluster. Again (cycle 2) the nucleotide is recorded and each cycle leads to the sequence of each fragment (a “read”). These reads are then aligned to a reference genome. By assembling reads (merging short reads together), it is therefore possible to reconstruct the unfragmented original sequence.

Both Ion Torrent (http://www.iontorrent.com/) and 454 (http://www.454.com/) employ the use of polymerase chain reactions to amplify DNA within an emulsified droplet (Figure 2). Sequencing information is correlated with either light (in 454) or hydrogen ions (in Ion Torrent) detection during each nucleotide incorporation event. If multiple nucleotide incorporation events occur, this is interpreted as stretches in the sequence of a particular nucleotide (homopolymers). All high throughput sequencing technologies have difficulties sequencing homopolymers, but sequencing homopolymers in Ion Torrent and 454 are more problematic because the nucleotides used lack a blocking moiety, resulting in entirely incorporating homopolymers during one cycle. Difficulties in interpreting homopolymers arise because these homopolymer signals are non-linear and have a Poisson distribution. This results in homopolymers of just two or three nucleotides to sometimes be contracted or expanded.

Figure 2. Ion Torrent and 454 sequencing.

Both Ion Torrent and 454 immobilize DNA fragments onto beads. In both platforms, template molecules are first immobilized on a bead which is emulsified so that subsequent amplification can occur clonally within the droplet. After clonal amplification, enrichment for DNA positive beads is performed using additional beads that can bind and isolate the available end of the library molecule, thus removing DNA negative beads. Enriched beads are deposited at the bottom of a well and sequencing is performed by flowing one base at a time over the templates. In Ion Torrent, an incorporation event is measured by a pH change from the release of protons resulting from the incorporation, whereas 454 uses a cascade of reactions resulting from pyrophosphate being released from each incorporation reaction. This leads to a photon being released by the enzyme luciferase. Therefore, the detection of light (in 454) or hydrogen ions (in Ion Torrent) when adenine is passed over each chamber is interpreted as thymine being the next nucleotide in the DNA sequence. Amplification of DNA fragments occurs in an emulsion and each bead is placed into a well large enough for each bead. Nucleotides are sequentially passed by each well where nucleotide incorporation occurs. If nucleotide incorporation occurs, a series of enzymatic reactions occurs that results in light being detected in the 454 platform. In Ion Torrent platforms, nucleotide incorporation results in the release of hydrogen ions and these ions are detected by each well. In Ion Torrent, if homopolymer repeats of the same nucleotide are present (GGG), multiple hydrogen ions will be released, generating a higher electrical signal. This is subsequently interpreted as multiple identical nucleotides being present in the sequence.

The Pacific Biosciences Real Time Sequencer (PacBio RS) (www.pacificbiosciences.com/) requires that each circular library molecule be bound to a polymerase enzyme as the input for sequencing on their single-molecule real-time sequencing (SMRT) cells. The library/polymerase complex is diffused over the SMRT cell’s Zero Mode Waveguides (ZMW’s), allowing the template to occupy the lumen of the ZMW sometimes. The RS uses video imaging of fluorescent nucleotides pausing at the bottom of the ZMW to record an incorporation event ²¹. Because the “pausing” can range from 1–3 seconds per incorporation event and nucleotides can diffuse freely into the ZMW, insertions are more common. It is possible to get multiple passes of sequencing around the same circular library molecule to generate what is called circular consensus sequencing (CCS) because of the long read length ²². The RS has seemingly random errors, whereas the other sequencing technologies tend to be less random with mistakes. These random errors, combined with multiple passes over the same circular fragment, generate a relatively low number of high quality reads, allowing the RS to be used as a cheaper variant validation tool over Sanger sequencing. For true single molecule sequencing, no amplification should occur on the sample to avoid amplification bias. Native DNA contains modified bases like 5-methyl cytosine that can be measured directly based on signature pausing signals with PacBio sequencing. For now, the lack of amplification may be the biggest drawback of single molecule sequencing, because some samples are just too low in material. This highlights the importance of creating new tools that can manipulate smaller reactions and use less input material.

Each sequencing instrument is limited as to the number of rounds it can record as well as the accuracy of the recording. In addition, there is a limit on the number of template molecules that can be read on each sequencing cell. Table 1 summarizes some of the main sequence platforms on the market today in terms of their relative costs, sequencing yields, quality scores, and sequencing times.

Table 1. Comparison of the current next generation sequencing platforms.

A cross comparison of Illumnia (Hiseq2500, Miseq), Ion Torrent (PGM 318, Proton I), PacBio (RS), and Roche 454 (FS FLX+, GS Junior) next generation platforms is presented. Each instrument was compared to show the specifications provided by the vendors including the cost, speed, accuracy, primary error type, and size of the data set that can be expected from each instrument. From those specifications, a cost per MB index was calculated. As each instrument needs to be maintained over time, the instrument maintenance cost was also provided by the vendors.

Platform	Cost/run	Unique library molecules sequenced	Benchmark Accuracy	Read Length at Benchmark Accuracy	Total Output at Specified Accuracy (GB)	Cost ($/Benchmark MB)	Sequencing time (not including library prep)	Single Molecule (Yes/No)	Primary Error Type	Annual cost of maintenance contract
Hiseq2500 2×100bp Rapid Mode (2 flow cells)	$5,830	1.2 billion	99.90%	80% @ 100bp PE, 80×80 (160+bp total)	100 GB	$0.06	27 hours	No	Substitution	$56,925
Hiseq2500 2×100bp High Output Run Mode (2 flow cells)	$5,830	1.5 billion	99.90%	80% @ 100bp PE, 80×80 (160+bp total)	540 GB	$0.01	11 days	No	Substitution	$56,925
Miseq 2×250bp	$995	15–17 million	99.90%	70% @ 250bp PE 200×150 (350+bp total)	5.6GB	$0.18	39 hours	No	Substitution	$11,250
Ion Torrent PGM 318 chip	$749 Including 318 chip OneTouch 2 template kit Sequencing Kit	3–6 million	99% raw	Up to 400bp	2Gb	$0.38	3.5 hours	No	Insertion Deletion	$8000/PGM System (includes torrent server and onetouch)
Ion Torrent Proton I	$1,000 including Ion PI Chip PI template kit for OneTouch 2	60–80 million	99% raw	Up to 200bp	10Gb	$0.10	2–4 hours	No	Insertion Deletion	$24,400 for Ion Proton System (includes OneTouch and TorrentServer)
PacBio RS 1, 120min video XL Binding kit and C2 chemistry	$110	23 thousand	87% raw	Up to 21,000 bp (4,500 average)	100MB	$1.10	12 hours	Yes	Insertion	$85,000
PacBio RS 2, 55min videos, XL Binding kit and C2 chemistry	$110	62 thousand	87% raw	Up to 13,000 (4,200 average)	260MB	$0.42	12 hours	Yes	Insertion	$85,000
454 GS FLX +	$2,500	Over 1 million	99.90%	Up to 1000 (700 average)	700 MB	$3.57	6 hours	No	Insertion Deletion	$50,000
454 GS Junior	$800	70–100 thousand	99.90%	Up to 600 (400 average)	40 MB	$20.00	3.5 hours	No	Insertion Deletion	$12,600

Common Applications in Next Generation Sequencing

DNA Sequencing

The order of DNA sequence and its variation dictates human developmental processes, uniquely identify each person, and encodes our susceptibility to diseases ^23–25. By using high throughput DNA sequencing (DNA-seq) technologies, it is possible to identify genetic variants that play a role in human health. In whole genome sequencing, sequencing information of both the exons and introns is obtained ^{15, 26, 27}, which may provide critical information on enhancer regions, promoters, and cis/trans regulatory elements that reside in the intronic regions along with structural variants such as copy number variants (CNV), inversions, and translocations affecting the exonic regions. To give a confidence level of how accurate the sequencing information is, the term “depth” is used to define the average number of times each nucleotide in the genome is observed ²⁸. For example, if each nucleotide of the genome is reported an average of 10 times during sequencing, a depth level of 10× is obtained. Read depth is important to interpret structural variations, because for a given interval, an increase in the amount of reads at a given read depth may indicate an increase in copy number, whereas a decrease in the amount of reads at a given read depth may indicate deletions ²⁹ (Figure 3). In general, as the read depth level increases, the sequencing information becomes more confident. It is recommended that an average read depth of 30× produces an adequate coverage level (the amount of times a nucleotide is reported within an assembled sequence) for whole-genome analysis and at a 50× read depth, approximately 94.9% of single nucleotide variants can be observed ^30–32. Read depth is important in considering the experimental goals. To observe small changes (i.e., point mutations) associated with complex diseases, a high read depth and sequencing multiple individuals are required. However, observing large structural changes in comparison to a reference genome could be achieved with a very low read depth. In most cases, increasing read depth is more costly and alternative, targeted, and more cost-effective methods may be preferred over HTS (i.e., customized hybridization chips). In addition, given that in each sequencer run only a limited number of sequence fragments can be read, it may be more cost effective to only analyze the sequence of the genetic material that is transcribed into mRNA (exons), depending on the experimental needs. In this regard, whole exome sequencing uses sequence capture methods to enrich a subset of genomic DNA ^33–35 using commercially available capture arrays (Roche NimbleGen, Agilent SureSelect, and Illumina 62MB). These arrays work with a set of binding oligonucleotides complementary to the human exome bound to magnetic beads. Further magnetic bead isolation enriches exome DNA, but may also introduce sequencing bias since some capture methods may not uniformly capture target DNA. However, isolating the exome provides relatively inexpensive 100× coverage of the genome coding regions and may be particularly useful for identifying rare genomic variants from a population of cells.

Figure 3. Detection of SNVs and CNVs.

Mapping HTS reads to control annotations (unaffected family member, reference annotation) is used to identify single nucleotide changes up to large structural DNA changes. Discrepancies between the reference annotation and the mapped sequence can be called to annotate SNV. More reads mapped (i.e., higher read depth) results in a higher confidence level to the called SNV in question. If a disproportionate amount of reads are mapped to a gene for a given read depth, this region of DNA may be interpreted as a CNV.

Bioinformatics analysis of DNA sequencing has come a long way since the original chromosome walking technique employed at the beginning of the Human Genome Project. Shotgun sequencing was developed to utilize high throughput short read technologies to assemble large genomes de novo. Large contiguous regions (contigs) are assembled from shorter ones using overlapping regions to link contigs, or reads, and the overhangs are used to extend the contig. Once a reference genome has been assembled for a species (or sometimes an individual), alignment within the reference is possible. For example, one could align reads to the reference genome (using programs such as BWA ³⁶, SOAP3 ³⁷, and BFAST ³⁸) or call single nucleotide variants (using GATK ³⁹, MAQ ⁴⁰ or SAMtools ⁴¹) and compare whole genomes through consensus with some flexibility allowed for variants/differences from a reference. Difficulties in DNA sequencing analysis lie in obtaining coverage in regions of extreme GC/AT content, discerning sequencing and amplification errors from actual variants (especially in the case of heterozygous variants, and rare variants from a population of cells) and being able to utilize short reads to assemble large repeats and/or large structural variants such as inversions. In addition, errors in mapping short reads can also occur given the ambiguous highly repetitive genomic regions and highly homologous gene families. To overcome these difficulties, newer software algorithms, longer read lengths, and/or higher coverage will need to be developed.

HTS has the potential to play an important role in cardiovascular research because many diseases have an unknown underlying genetic component. For example, arrhythmogenic right ventricular dysplasia/cardiomyopathy (ARVD/C) is caused by mutations associated with desmosomes. However, non-desmosomal mutations in TGFβ3 RYR2 and TMEM43 have also been implicated in the disease phenotype of ARVD ⁴². While mutations in multiple genes have been identified to cause ARVD/C, only in 50–60% of ARVD/C patients could an underlying genetic mutation be found (reviewed in ⁴³). In addition, some clinical presentations of ARVD/C are very similar to Brugada syndrome (predominantly exhibited by males, associated with familial inheritance, and exhibits idiopathic ventricular fibrillation) ⁴⁴. Histopathological or advanced imaging modalities are required to distinguish between these two diseases ⁴³. Whole genome and exome sequencing will lead to the discovery of previously unknown mutations that cause cardiovascular diseases as well as aid in the distinction between diseases that share very similar clinical presentations.

High throughput DNA sequencing will be instrumental in the screening and diagnostics of heart diseases related to larger structural genomic changes such as Down syndrome ⁴⁵, DiGeorge syndrome ⁴⁶, 4q- Syndrome ⁴⁷, and 8p- Syndrome ⁴⁸, as well as complex diseases related to copy number variants ⁴⁹ and single nucleotide changes (single-nucleotide polymorphisms (SNPs), single-nucleotide variants (SNVs), and mutations). SNVs are variable regions of the DNA in which single nucleotide differences have been identified in the genetic code, whereas a SNP is a variant that appears with a >1% minor allele frequency in the population ^{15, 50}. These observed polymorphisms may help predict the susceptibility of a patient cohort to develop heart disease. This is exemplified in the study by Matkovich et al.,⁵¹ where pooled sequencing data from four cardiac signaling genes identified a greater representation of specific SNPs within the cardiovascular heat shock protein gene HSPB7 from patients with heart failure. In addition, while one SNP was found to be within an intron of HSPB7, no differences were observed in the splicing or mRNA levels of this gene. Sequencing the adjacent renal chloride channel CLCNKA gene, however, identified a SNP in an exon of this gene, which demonstrated linkage disequilibrium with the intronic SNP in HSPB7. Further functional characterization of the renal chloride demonstrated an approximate 50% loss-of-function of the variant channel ⁵². In summary, HTS performed in these studies led to the identification of a common genetic risk factor for heart failure.

Given that SNV analysis can determine which genetic regions may influence a patient’s susceptibility to develop heart disease, SNV analysis could also be used to determine which drug therapy will be best suited for a particular patient. For instance, warfarin is an anticoagulant drug often prescribed to prevent thrombosis, and patients with common SNPs in the CYP2C9 and VKORC1 genes have been suggested to successfully predict a patient’s response to the anticoagulant effects of warfarin ^{53, 54}. Further clinical studies will be required to warrant the use of SNP data to predict warfarin treatment. In addition, SNP analysis is being used to identify which SNPs can be either cardioprotective versus cardiotoxic to the effects of doxorubicin ^55–57. Future SNP analysis studies will be important for optimizing patient specific treatment to existing cardiovascular drugs and for determining the effectiveness and safety of drugs under development ⁵⁸.

Chromatin Immunoprecipitation Sequencing (ChIP-seq)

Gene expression can be influenced by epigenetic modifications which can be assessed by ChIP-seq. DNA in the nucleus is divided into actively transcribed regions called euchromatin, or transcriptionally silenced regions called heterochromatin ⁵⁹. These regions represent loosely or tightly compact DNA regions and these different states are influenced by histone protein modifications ^{60, 61}. Histone acetylation and methlyation are two modifications for histones, and depending on the histone modification, genes may be actively transcribed or repressed during these processes. For example, H3K27Me3 modification represses gene expression ⁶¹ whereas H3K4Me3 modification enhances gene activity ⁶². By performing chromatin immunoprecipitations with antibodies towards various histone modification states and sequencing the resulting immunopreciptated DNA, it is possible to assess different regions of DNA that may be actively transcribed or are transcriptionally silent. In ChIP-seq, formaldehyde is first used to covalently bond DNA to proteins with which they are interacting (Figure 4). The DNA-protein complex is fragmented and immunoglobulins specific for the protein of interest are used to pull down the fragment of DNA to which they are attached ⁶³. From here the target DNA is isolated and a sequencing library is made using a standard library preparation method. Sequencing of a ChIP-seq library generates reads that align near the genomic regions associated with the target protein. Controls include a negative control antibody library and an input DNA library. Given that some antibodies are better at pulling down the target protein-DNA interactions, the use of a ChIP certified antibody greatly improves the signal-to-noise ratio in downstream analysis. In addition, under- or over-paraformaldehyde crosslinking as well as under/over DNA sheering can affect downstream ChIP-seq analysis. Bioinformatic analysis of ChIP-seq data involves mapping reads to a reference genome and using peak detection software (currently over 31 open source programs)¹¹ to identify regions that have enriched mapping frequency. Therefore, the peak detection of immunoprecipitation from H3K27Me3 and H3K4Me3 histone modification states can provide information on which regions of the DNA are in an active or open chromosome state versus transcriptionally silent.

Figure 4. Chromatin immunoprecipitation for ChIP-seq.

To determine sequences of DNA that may be “open” for transcription, ChIP-seq uses the immunoprecipitation of chromatin bound to different histone modification states [transcriptionally silent (H3K27me3) and transcriptionally active (H3K4Me3) markers]. Crosslinking chromatin to the histones is first performed using formaldehyde, and chromatin is sheared into smaller fragments. Immunoglobulins specific for each histone modification state are incubated in separate reaction tubes, and magnetic or Sepherose beads known to bind to immunoglobulins are used to isolate bound immunoglobulin/chromatin complexes. Chromatin is separated from the histones by reverse crosslinking (high salt conditions) and protein digestion (proteinase K). DNA libraries are then made from the isolated DNA and sequencing of these DNA fragments is processed using one of the next generation sequencing platforms.

An additional important application of ChIP-seq is to determine which genes are bound by transcription factors and to determine which enhancer regions are active in the heart. In a study by Blow et al.,⁶⁴ ChIP-seq of the transcriptional co-activator protein p300 on embryonic day 11.5 heart tissue was used to elucidate which enhancer regions are active in the developing mouse heart. ChIP-seq can also be used to determine how aberrations in transcription factor binding can disrupt gene expression and lead to cardiovascular diseases. Given that NKX2-5 mutations have been shown to cause hypoplastic left heart syndrome, atrial septal defects, and patent foramen ovale ⁶⁵, immunoprecipitation of control and mutant NKX2-5 could be used to identify how NKX2-5 mutations affect transcription.

Identifying gene and protein interactions can also be determined using chromosome confirmation capture (3C) sequencing. 3C sequencing is important to identify functional associations among distal chromosomal regions (such as enhancers) ^{66, 67}. These DNA interactions can be determined by first crosslinking DNA/protein complexes using formaldehyde and then using restriction enzymes to digest DNA into smaller fragments, leaving the crosslinked DNA fragments connected. DNA is then intramolecularly ligated and reverse crosslinked with heat. From here adaptors can be ligated on to generate a sequencing library. In a study by Korostowski et al.,⁶⁸ 3C sequencing was used to demonstrate how changes in chromosomal interactions occur with the promoter of the potassium channel Kcnq1 and interactions with the Kcnq1 promoter was demonstrated to influence the transition of a monoallelic to a biallelic expression of this gene during development of the heart. Future studies using 3C sequencing will not only be important to determine which distal chromosomal regions interact but also to study how these interactions occur temporally or in a tissue specific manner. By delineating which regions of DNA interact spatially, a deeper and more complete understanding into the mechanisms causing cardiovascular disease can be achieved.

Methylation Sequencing (Methyl-seq)

Methylation of DNA is another epigenetic modification that can influence gene expression and the methylation status of DNA can be determined by Methyl-seq. 5-methyl cytosine is the most common modified base in humans and generally methylation of cytosine occurs when cytosine neighbors a guanine nucleotide called cytosine guanine dinucleotides (CpGs) ²⁴. Areas of the genome high in CpG concentration have increased methyl transferase activity and may be referred to as CpG islands. Methylation at CpG islands decreases the activity of promoters and generally decreases gene expression. DNA methylation in CpG islands has been shown to strongly suppress promoter activity and seems to occur as a function of age, causing loss-of-function phenotypes, and may be a target in many disorders including heart disease ⁶⁹. One method to sequence methylated regions of DNA (Methyl-seq) involves first isolating DNA, fragmenting the DNA, and then separating this sample into two reactions. One reaction is treated with bisulfite and the other portion is left untreated. Bisulfite treatment changes cytosine nucleotides to uracil nucleotides, leaving methylated-cytosine nucleotides unchanged because they are resistant to bisulfite treatment ⁷⁰.

Methyl-seq data analysis involves sequencing bisulfite-treated DNA and comparing this with the fraction that did not undergo bisulfite treatment, producing differences that can identify the regions that did not convert to uracil; an uracil is “read” as a thymidine during sequencing and is reported as a methylation site. Incomplete bisulfate conversion can be problematic because these regions will be detected as a methylated site. One method to determine if bisulfate conversion went to completion is the addition of spike in control DNA in which the methylation status is known. Methylation rich regions can also be immunoprecipitated using an anti-methylated cytosine antibody. In a study by Movassagh et al.,⁷¹ immunoprecipitation of methylation regions led to genome-wide DNA methylation patterns that were quite similar between control and end-stage cardiomyopathy hearts. Moreover, differences in methylation could be observed when analyzing the methylation pattern at the single gene level. Identification of the gene promoters that are hypermethylated versus hypomethylated may therefore be useful in predicting which gene will become active during various stages of heart disease.

Transcriptome Sequencing (RNA-seq)

RNA-seq is particularly useful in assessing the current state of a cell or tissue as well as the possible effects of disease states or treatment conditions on the transcriptome. RNA-seq also provides information on the differences between the transcriptome and the exome that result from RNA editing. While microarrays have revolutionized the study of transcriptomics and proved useful in determining gene expression profiles, RNA-seq by comparison is more sensitive, provides absolute quantity levels, is not affected by on-chip sequence biases, and gives additional information in gene expression levels and splice junction variants ^{72, 73}.

In RNA-seq, RNA is commonly first converted to a more stable complementary DNA (cDNA) through a combination of reverse transcription and the selection process to isolate the RNA from the abundant rRNA. The input RNA quality is very important in RNA-seq preparation because RNAse enzymes are ubiquitous and extremely stable and fragmentation can also occur simply when a divalent cation is present. Library preparation and sequencing of cDNA follows the same sequencing procedure as DNA-seq. However, numerous variations of RNA-seq library preparations have been developed, each with its benefits and limitations in terms of relative costs and input requirements. The main differences in these various library preparations are the methods of purifying and isolating RNA of interest (mRNA, uRNA, full length transcripts etc.). RNA-seq libraries can be made using polyadenylated tail selection, not-so-random primers (for reverse transcription), and ribosomal depletion ^74–76. Isolating polyadenylated mRNA and then reverse transcribing is the conventional method of preparing an RNA-seq sample, but it favors the 3' end of transcripts and does not work well with low-quality or degraded samples or provide any information about non-coding RNA. A commercially available kit (Clontech SMARTer) is available to generate full-length cDNA from high-quality, low-input RNA samples by using the 3’ poly-A tail as the priming site for first strand cDNA synthesis and by enzymatically adding on a specific primer hybridization site on to the 5’ end after first strand synthesis for the second strand. Two other general methods have been commercially developed to selectively remove rRNA. For example, a method for selectively amplifying non-ribosomal RNA offered in a sequencing preparation kit (NuGEN Ovation) involves the use of a designed set of reverse transcription primers that contain all variants of random oligonucleotides (random primers), excluding the ones that would amplify ribosomal RNA (non-random primers). Ribosomal depletion immobilizes ribosomal RNA to remove it before reverse transcription. The use of one of these methods can recover additional RNA signals that would not be otherwise obtained via a poly-A selection due to the degraded and non-coding RNA. To annotate de novo discovery, strand specific RNA sequencing is used to determine which strand of RNA was the original template in reverse transcription. By pre-processing RNA to select for polyadenylated mRNA, or by selectively removing ribosomal RNA, a greater sequencing depth can be achieved. Depending on the experimental design, a greater sequencing depth may be required when complex genomes are being studied or if information on low abundant transcripts or splice variants is required.

In general, bioinformatic analysis consists of aligning the sequence reads to a reference genome, assembling the reads into transcripts, and detecting differences in transcript expression between or among groups. The Tuxedo Suite consisting of Bowtie ⁷⁷, Tophat ⁷⁸, and Cufflinks ⁷⁹ can be used as open-source software packages to perform these operations and multiple updates to these software packages have increased the speed and accuracy in RNA-seq analysis. Additional splice variant detection and alternative exon usage can be identified using software packages such as MISO ⁸⁰ and DEXseq ⁸¹, which can quantify reads to individual exons. While many of these software packages give a probabilistic framework to identify changes in transcript splicing patterns, false discovery cutoffs are necessary to identity true splicing events.

Transcriptome changes assessed by RNA-seq play a valuable role in cardiovascular medicine, because transcriptome changes can identify how cardiovascular diseases change with time. Lee et al. ⁸² used RNA-seq to study how murine hearts change during heart failure, whereas Song et al. ⁸³ employed RNA-seq to decipher the transcriptome differences between physiological hypertrophy and pathological hypertrophy. In addition, Hu et al. ⁸⁴ studied the mRNA and microRNA transcriptome changes that occur during pressure overloading hypertrophy in mice hearts by HTS. By identifying which mRNAs and microRNAs changed during hypertrophy, and which microRNA–mRNA interaction occurred with immunoprecipations of argonaute 2 RNA-Induced Silencing Complexes (termed RISC-seq), they demonstrated that small changes in microRNA expression could lead to global mRNA changes during heart stress. Other results show that the use of RNA-seq with ChIP-seq information has provided significant advances in the study of how transcription factor binding can influence changes in gene expression. To demonstrate the potential of this approach, RNA-seq was performed on the hearts from Tbx20 knockout mice that had rapidly developed heart failure. By combining the transcriptome changes that occur from the loss of Tbx20 with the putative transcription factor binding sites of Tbx20 previously identified with ChIP-seq ⁸⁵, a comprehensive analysis of how the loss of Tbx20 leads to heart failure was achieved ⁸⁶. In addition, combining ChIP-seq along with RNA-seq was also used to successfully identify genes and chromatin marks involved in the progression of cardiomyocyte differentiation from human induced pluripotent stem cells (iPSCs) ⁸⁷.

Ribosomal Sequencing (Ribo-seq)

The RNA content within the cell does not automatically lead to the production of functional proteins. While the analysis of the total RNA can give an overview of the current RNA fragments present in the cell, selecting the RNA that is bound to ribosomes can offer a better indication of which RNA fragments are in the state of translation. Studying the sequence of RNA bound to ribosomes is called ribosomal footprinting and sequencing these short RNA fragments is called “Ribo-seq” ^{88, 89}. Ribo-seq is particularly useful when studying transient transcriptional events that are tightly controlled, including mitosis ⁹⁰. To perform Ribo-seq, cycloheximide treatment is first used to block the elongation phase of eukaryotic translation ⁹¹. Cells are lysed and fragmentation of RNA is performed (RNase I treatment). Ribosomes containing short fragments of RNA are then separated by ultracentrifugation and the short RNA fragments are separated from the ribosomes using proteases (proteinase K). These isolated short RNA fragments are sequenced to indicate which RNA fragments were being actively translated. While this technique can provide valuable information on how the cell’s translational machinery operates, it is also more technically challenging, and consequently only a few studies have used Ribo-seq to study the processes of cardiovascular diseases thus far.

The Future of Sequencing Technology

Some observers have compared the pace of advancements in DNA sequencing to that seen in the computer industry, which has been able to reduce costs and processing time exponentially since its inception. However, genetic sequencing is less mature as a field and faces many more technical challenges ahead. For instance, accuracy is paramount for sequencing to become more widely applicable in the clinic, a goal that may be amenable with better algorithms that can correct for reading errors and advanced molecular biology techniques and applications. One application that may prove useful in cardiovascular medicine is cell free DNA and RNA sequencing. Given that one of the components of circulating whole blood is cell free nucleic acids, sequencing these cell free nucleic acids may indicate the current state of the cardiovascular system. RNA is short lived in the presence of RNAses and this makes RNA present in the blood a good temporal measure of what is occurring in the body at the time of extraction. Therefore, DNA and RNA sequencing from cell free nucleic acids could prove to be a relatively noninvasive measure of cardiovascular health.

Recent advances have led to more precise control of picoliter scale volumes and chemical reactions. The next milestone may be to sequence unamplified, unmodified native nucleic acids. There are single molecule technologies such as the Pacific Bioscience RS, but they do require adaptor ligation to modify the sample before sequencing can take place. The future direction of sequencing technologies will likely involve methods of directly sequencing single molecules of DNA or RNA in native form from low input starting material (e.g., a few cells containing approximately 50 pg of DNA/cell), without sacrificing accuracy or cost ⁹². Automated microfluidic sample prep methods are being developed that can isolate a single cell’s genetic material and process it into a sequencing library all in one closed system.

Nanopore Sequencing Technologies

Oxford Nanopore (http://www.nanoporetech.com/) is developing a nanopore technology that someday may be capable of sequencing unmodified miRNA and mRNA molecules. Nanopore sequencing technologies comprise a relatively newer set of techniques being developed by companies such as Oxford Nanopore and NABsys (http://www.nabsys.com/), which are working on massively parallel sequencers not based on sequencing by synthesis. The concept is to electrophorese molecules through a pore of a membrane and then measure the electrical current through the pore as molecules pass through ⁹³ (Figure 5). By characterizing the current of a pore over time, nanopore technology may be able to determine exactly what has traversed the pore and in which order. These companies hope to develop products that can determine the entire genome, sense and antisense, from small amounts of unmodified input in a short amount of time ⁹⁴. This technology will also be used to measure RNA and proteins directly someday, and a major challenge will be to control the flow of DNA through the pore and to decipher the resulting message ^{95, 96}. To tackle this problem, Oxford Nanopore is using an approach in which an exonuclease releases one nucleotide at a time, whereas NABsys has adopted an approach using a series of oligonucleotide probes hybridized to the denatured sample that subsequently can be "seen" crossing the pore and positioning along the template molecule.

Figure 5. Nanopore technology.

Third generation sequencing is expected to measure the change in ion flow (current) within a membrane as small molecules are passed through a small pore inside the membrane. Different current profiles will therefore indicate which nucleotide passed through and in which order.

Conclusion

In summary, advances in HTS technologies are enabling a more accurate and comprehensive representation of cardiac development and disease processes. While more researchers are using HTS to study cardiovascular medicine, the full potential of current HTS sequencing platforms in cardiovascular medicine has yet to be realized. HTS will be essential in identifying biomarkers of disease, staging disease progression, and linking genotypic to phenotypic outcomes. Given the rapid pace of development in sequencing technology during the last decade, future sequencing technologies promise to further help us understand the roles that the genome, transcriptome, and proteome play in the cell by identifying cellular mechanisms. This may also lead to deeper and more comprehensive insights into disease mechanisms at a subcellular level, possibly connecting causation effects to gene expression levels. As different platforms improve depth by increasing in read length and decreasing in sequencing time, and decrease in cost, improvements in HTS applications will provide a more complete molecular picture into the functionality of biological processes. Ultimately, improved understanding of these biological processes may lead to dramatically safer and more effective therapies for cardiovascular diseases.

Nonstandard Abbreviations and Acronyms

ARVD/C

arrhythmogenic right ventricular dysplasia/cardiomyopathy

CCS

circular consensus sequencing

cDNA

complementary DNA

ChIP-seq

chromatin immunoprecipitation sequencing

CLCNKA

chloride channel, voltage-sensitive Ka

CNV

copy number variants

Contig

large contiguous regions

CpGs

cytosine guanine dinucleotides

CYP2C9

cytochrome P450, family 2, subfamily C, polypeptide 9

ddNTPs

2′,3′-dideoxynucleotides

dNTPs

2′-deoxynucleotides

ENCODE

ENCyclopedia Of DNA Elements

HSPB7

heat shock protein family, member 7

HTS

high throughput sequencing

iPSCs

induced pluripotent stem cells

Kcnq1

potassium voltage-gated channel, KQT-like subfamily, member 1

MB

megabase

Methyl-seq

methylation sequencing

NKX2-5

NK2 homeobox 5

Ribo-seq

ribosome sequencing

RISC-seq

RNA-Induced Silencing Complex sequencing

RNA-seq

RNA sequencing

RYR2

Ryanodine receptor 2

SNP

single-nucleotide polymorphisms

SNV

single-nucleotide variants

SMRT

single-molecule real-time

TGFβ3

Transforming growth factor beta 3

TMEM43

Transmembrane protein 43

KORC1

vitamin K epoxide reductase complex, subunit 1

ZMW

Zero Mode Waveguides