Application of third-generation sequencing in cancer research

Abstract

In the past several years, nanopore sequencing technology from Oxford Nanopore Technologies (ONT) and single-molecule real-time (SMRT) sequencing technology from Pacific BioSciences (PacBio) have become available to researchers and are currently being tested for cancer research. These methods offer many advantages over most widely used high-throughput short-read sequencing approaches and allow the comprehensive analysis of transcriptomes by identifying full-length splice isoforms and several other posttranscriptional events. In addition, these platforms enable structural variation characterization at a previously unparalleled resolution and direct detection of epigenetic marks in native DNA and RNA. Here, we present a comprehensive summary of important applications of these technologies in cancer research, including the identification of complex structure variants, alternatively spliced isoforms, fusion transcript events, and exogenous RNA. Furthermore, we discuss the impact of the newly developed nanopore direct RNA sequencing (RNA-Seq) approach in advancing epitranscriptome research in cancer. Although the unique challenges still present for these new single-molecule long-read methods, they will unravel many aspects of cancer genome complexity in unprecedented ways and present an encouraging outlook for continued application in an increasing number of different cancer research settings.

Introduction

Genome sequencing has become increasingly available over the past few decades. The second generation of sequencing technologies, known as next-generation sequencing (NGS), “massive-parallel,” or “high-throughput” sequencing, made DNA sequencing dramatically simpler and faster by employing microscopic, spatially separated DNA templates to massively parallelize the capture of data [1]. The sequencing process uses various platforms, such as DNA sequencing, RNA sequencing (RNA-Seq), single-cell RNA and DNA sequencing. DNA sequencing, such as whole-exome sequencing initially allows the detection of single nucleotide variants (SNVs) and copy number variations (CNVs) [2]. Similarly, RNA-Seq data analysis can produce information about gene expression levels, alternative splicing (AS), allelic silencing or differential allelic expression, gene fusions, and RNA editing [3, 4]. Single cell sequencing is emerging as a powerful tool for profiling cell-to-cell variability on a genomic scale [5]. Intratumor heterogeneity is a confirmed major cause of treatment failure and drug resistance in cancer. Single-cell sequencing of tumor cells addresses this issue by identifying subpopulations of cancer cells and immune cells within a single patient. Furthermore, spatial transcriptomics research is expected to generate highly detailed maps of single-cell gene expression at any tissue coordinate in cancer [6]. To date, NGS has become a standard tool for many applications in basic biology as well as for clinical and agronomical research. However, the read length from NGS data remains a bottleneck for biological studies.

With the development of sequencing technology, DNA sequencing has inched the era of single-molecule sequencing (SMS) or third-generation sequencing (TGS) [7]. When DNA sequencing, PCR amplification is not needed to obtain the individual sequence of each DNA molecule and produces long-read in a real-time sequencing process [8]. The two major available TGS technologies are Pacific Biosciences (PacBio) single molecule real-time (SMRT) sequencing and the Oxford Nanopore Technologies (ONT) sequencing platform [9], [10], [11]. These long-read technologies permit sequencing/assembly through repetitive and complex elements, direct variant phasing, and even direct detection of epigenetic modifications that provide preference for certain applications [12], [13], [14]. The methodologies for these technologies (Figure 1) and comparison of the performance of NGS short-read and TGS long-read methods in terms of read accuracy, throughput and cost were comprehensively described in recent reviews [8, 15]; herein, we will focus on their potential use in advancing our understanding of cancer genomics, and applications in cancer research. In this review, we first provide a brief description of NGS in cancer research and outline the limitations of NGS and then describe the developments in TGS technologies. Next, we look at the bioinformatics for TGS data and describe the challenges in the development of bioinformatics tools. We also provide examples of how TGS has been adapted to investigate key aspects of cancer genomics for applications including the analysis of complicated cancer genomes, AS, fusion genes, exogenous RNA, and epigenetic marks. We highlight the unique challenges still present for TGS technologies in cancer research and provide some suggestions to overcome these challenges. We finish by briefly discussing the comparison of advantages between TGS and NGS and provide certain applications suited for each platform.

Figure 1:

Overview of NGS short-read and TGS long-read methods.

(A). In NGS by Illumina technology, DNA is fragmented into manageable sizes, and these fragments are ligated to adapters. After library preparation, individual DNA molecules are sequencing for short reads. Following a sequencing run, raw sequence reads were aligned to a reference genome. (B). In PacBio SMRT sequencing, DNA fragment is ligated to hairpin adapters to form a topologically circular molecule, known as SMRTbell. It is loaded onto a SMRT Cell and bound by a DNA polymerase for sequencing. In ONT sequencing, DNA is tagged with sequencing adapters preloaded with a motor protein on one or both ends. The DNA is combined with tethering proteins and loaded onto the flow cell for sequencing. Following a sequencing run, raw sequence reads were aligned to a reference genome. NGS, next generation sequencing; TGS, third generation sequencing.

A brief description of NGS in oncology

NGS has been widely used in new mutation screening, gene expression profiling, molecular classification, neoantigen prediction, and liquid biopsy in cancers [16], [17], [18]. These capabilities could impact therapy selection by offering insights regarding therapeutic sensitivity or resistance, or factors affecting diagnosis or prognosis [19]. For example, SNVs, insertion/deletion (indels) mutations, structural alterations (CNVs, translocations, inversions), loss of heterozygosity and aneuploidy can all be detected in a whole-genome data set comparing tumor DNA and normal DNA [20, 21]. In addition to genome characterization, NGS has also been deployed to characterize the cancer transcriptome using RNA-seq. RNA-seq can be able to provide information on RNA expression, as well as to detect alternative splicing and fusions [22]. Furthermore, single-cell sequencing through NGS has the advantages of assessing of tumor heterogeneity and separate cell types, such as immune cells [23]. The Cancer Genome Atlas (TCGA), funded jointly by the National Cancer Institute and National Human Genome Research Institute of the National Institutes of Health and the International Cancer Genome Consortium (ICGC), contains data for a series of large-scale studies in different countries across the world that were funded by the governments of each country and have helped establish the importance of cancer genomics and transformed our understanding of cancer [24]. Comprehensive analyses of these cancer genomic data provide unique opportunities for understanding cell-of-origin patterns, oncogenic processes, and signaling pathways [25, 26].

Although the availability of whole-genome (WGS), -exome (WES), or -transcriptome sequencing (RNA-seq) has been increasing, targeted gene sequencing is the method of choice for cancer diagnosis in clinical laboratories [27, 28]. With increasing numbers of cancer driver mutations genes now known, NGS panels are commonly used in hospitals for a wide range of cancers to ensure optimal sequencing quality (read depth and coverage, variant characterization, reporting), cost-effectiveness, and turnaround time [29]. In addition to NGS panels, identified biomarkers are also being used for cancer diagnosis, prognosis, and therapeutic applications in clinical laboratories [30], [31], [32]. In our Precision Cancer Medical Center in Fudan University Shanghai Cancer Center (FUSCC), large custom-designed NGS panels have been developed for breast cancer and other solid tumors for the detection of SNVs, indels, and CNVs [33]. The FUSCC-BC panel is used to detect somatic and germline mutations in breast cancer-specific genes in clinical settings, and identified the mutation characteristics, and potential molecular targets for breast cancer in China.

Another application of NGS in oncology is the identification and enrollment of patients for appropriate clinical treatments, such as liquid biopsy [34]. Blood contains many types of biological materials like circulating cells, extracellular vesicles (EVs), non-coding RNAs (ncRNAs), and cell-free DNA (cfDNA). The minimally invasive procedure for sample acquisition for this type of DNA or RNA assessment has been coined a “liquid biopsy”. Tumor cells in the body can release cfDNA through apoptosis, necrosis, or activate release [35]. Furthermore, cancers infrequently shed cells into the circulation, known as circulating tumor cells (CTCs) [36]. Therefore, DNA can be obtained from cancer patient blood samples or other bodily fluids (cerebral spinal fluid, cervical mucus, and urine). In addition, EVs are found in various body fluids and serve for intercellular communication by delivering their cargo molecules to other cells [37]. The emergence of EVs analytics in combination with liquid biopsy sampling opened a plethora of new possibilities for the detection of tumors [38]. RNA-seq, which can provide accurate and comprehensive gene expression profiling of liquid biopsies including genomic components of extracellular vesicles (EVs) and ncRNAs from blood, dramatically enhancer the probability of ncRNAs as biomarkers [39]. Therefore, the detection of ncRNAs using RNA-seq is a noninvasive, innovative approach for diagnosis [40]. The tumor-specific signatures in these samples can act as a new type of cancer biomarker and help to identify cancer patients from a group of healthy individuals [18]. Facilitated by the rapid development of NGS technologies, liquid biopsy can achieve much higher sensitivity than tissue biopsy and can be designed for different purposes [41]. However, there are several challenges due to its low concentration of DNA and high fragmentation (ranging from 100 to 10,000 bp fragments) in these samples [42], as well as the hard to build accurate classification model, using liquid biopsy for cancer screening and early detection remained to be solved.

Drawbacks of NGS methods

NGS is advantageous in many aspects, such as low cost, high speed and high yield. However, NGS methods also have some limitations. One of the most obvious limitations of NGS is the short-read length. This limits the precision of many biological studies, especially large genome assembly studies and precise specific isoforms analysis [43], [44], [45]. The read length of 100–200 bases is too short in the context of a vast genome, which makes it extremely difficult to accurately assemble the genome sequences from billions of short-reads [1, 46]. In addition, larger structural variations (SVs) are more challenging to detect and characterize using short-reads. Despite advances in sequencing technologies and bioinformatics, de novo assembly of large genomes remains challenging [3].

NGS platforms rely on clonal amplification to create multiple molecules, and do not have the sensitivity to detect nucleotides at a single molecule level. Transcriptomes based on the NGS platforms and RNA samples are typically subject to RNA fragmentation to a certain size range. Then, the cDNA fragments are sequenced in a high-throughput manner to obtain millions of short sequences. During sample preparation section, reverse transcription, PCR and size selection add base incorporation errors in individual molecules within a cluster [47]. The amplification process also creates an underrepresentation of bases in areas of high or low GC contents [48]. Therefore, during both the sequence processes and computational analysis phases, imperfections and biases may be introduced. This process greatly reduces the accuracy in the analysis of AS, gene fusion, and paralogous regions. Although generating high-read output and developing new computational approaches would be particularly beneficial for the quantitative analysis of transcript isoforms, other biases and limitations will result from the myriad of computational methods [3, 15].

NGS is also limited by its inability to directly sequence RNA. Short-read platforms rely on converting mRNA to cDNA before sequencing; as such, they are typically blind to nucleotide modifications. Consequently, by using NGS, indirect methods are required to identify transcriptome-wide RNA modifications. However, these methods cannot provide quantitative estimates of the modification at a given site and are often unable to identify the underlying RNA molecule that is modified [49].

TGS/long-read sequencing

PacBio and ONT technologies provide alternative long-read technologies that enable SMS of complete individual RNA molecules after conversion to cDNA. These two platforms both permit sequencing of non-amplified DNA of exceptionally long linear read lengths, high throughput and fast sequencing times (Table 1). PacBio utilizes a sequencing-by-synthesis approach by which polymerases incorporate distinguishable fluorescently labeled nucleotides to single DNA molecules and the fluorescent signals are recorded in real time in the zero-mode waveguides (ZMWs) [50]. Nanopore sequencing uses the minor changes in ionic current when nucleotide bases of single-stranded DNA/RNA molecules pass through protein nanopores to identify different nucleotides [51, 52]. The development and general features of PacBio and ONT sequencers are listed in Table 1 and Figure 2A.

Table 1:

Throughput of different third-generation sequencing platforms.

Platform	Sequel	Sequel II	Flongle	MinION	GridION	PromethION
Company	PacBio	PacBio	ONT	ONT	ONT	ONT
Throughput	15 Gba	96 Gba	1.8 Gb	42 Gb	210 Gbb	5.88–11.76 Tbc
Run time	Up to 20 h	Up to 30 h	Up to 16 h	Up to 72 h	Up to 72 h	Up to 72 h

Numbers are based on company website documentation (https://nanoporetech.com and https://www.pacb.com, both accessed April 29, 2021). ^aOne SMRT cell. ^bFive flow cells. ^cTwenty-four or forty-eight flow cells.

Figure 2:

Overview of cancer applications by PacBio and ONT.

(A) The development platforms of third generation sequencing (TGS); (B) Sequencing a region with two nearly identical repeats (blue) separated by a unique sequence (green) will generate reads corresponding to the upstream region (yellow) in short-read sequencing. Assembly programs for NGS will assemble these reads into a single contig; (C) Haplotype phasing. SNPs (single nucleotide: A or C) between maternal and paternal alleles; (D) Illustration of the use of direct mapping of long-read to a reference genome to resolve complex structural variations, including insertions, deletions, duplications, and complex variation; (E) Detecting of fusion genes; (F) Long-read single cell sequencing; (G) Base modification in a DNA and RNA molecule (represented by a red cycle) give rise to specific signals in TGS data that can be identified using computational methods; (H) Detecting of ncRNAs, including lncRNA, miRNA, and circRNA; (I) A multitude of mRNA transcript isoforms can be generated from a single gene through alternative intron splicing and alternative polyadenylation. Long-read sequencing covers the entire transcript and detects the whole alternative splicing events; (J) TGS-based DNA/RNA analysis for liquid biopsy; (K) Detecting of exogenous RNAs, including viral RNA and bacterial RNA. NGS: next generation sequencing; PacBio: Pacific Bioscience; Oxford Nanopore Technologies (ONT).

Long-read are an advantage of TGS technology. This technology overcomes some of the issues associated with short-read approaches and greatly improves the quality of genome assembly [8]. The high rate of false-positive splice junction detection by NGS is reduced with TGS, and the computational methods for de novo transcriptome analysis are also simplified, which leads to a more complete capture of isoform diversity, alternative polyadenylation (APA), fusion transcripts, structural variations, and paralogous regions [53]. Furthermore, another advantage presented by these platforms is that they have the ability to detect base modifications in native DNA. For instance, PacBio SMRT sequencing can directly detect and differentiate between base modifications such as 5-methylcytosine (5mC) and 5-hydroxymethylcytosine (5hmC) [54]. The ONT method determines the sequence of nucleic acids in a molecule directly without the need for amplification, sequencing by synthesis, or modification. This approach, termed dRNA-Seq, removes the biases generated by these processes, and permits the identification of RNA modifications and determination of poly-A tail length [55, 56]. Consequently, ONT technology can be used not only to sequence DNA but also to sequence RNA, including miRNA directly [50, 57].

However, PacBio and ONT share a common disadvantage of a high error rate of ∼5–20% randomly distributed errors before correction [12, 58]. Thus, it is necessary to reduce the error rate before subsequent utilization. The technology behind these systems is considerably different. For higher accuracy, these two platforms developed their own methods to achieve the accurate sequence. ONT developed a method to sequence both strands of a double-stranded DNA molecule, which was called “two-directional” (2D) sequencing. For PacBio SMRT sequencing, the library preparation on this platform creates a circular input molecule. The molecular can be sequenced many times, and the random errors can be mitigated by increasing the read depth using circular consensus sequence (CCS) for high-fidelity (HiFi) sequence reads. Furthermore, self-correction software, such as long-read multiple aligner (LoRMA), which needs high coverage to obtain accurate correction, can be used for error correction [59]. Alternatively, the hybrid error correction strategy uses short-reads from NGS to correct long-read [60]. As such, a hybrid sequencing strategy combining second- and third-generation sequencing technologies could reduce the error rate and quantify transcript isoforms or fusion genes [61, 62].

Bioinformatics for TGS data analysis

As mentioned above, TGS offers a number of advantages over short-read sequencing. However, the data from these platforms is qualitatively different from NGS, thus necessitating tailored analysis tools. In addition, raw reads from PacBio SMRT sequencing and Oxford nanopore sequencing have high error rates with most errors occurring due to false insertions or deletions [50,63], [64], [65]. The error mode in SMRT sequencing is remarkably stochastic in nature [50]. For nanopore sequencing on the other hand, the error profile has been reported to be biased. For example, A-to-T and T-to-A errors were estimated to be very low [66, 67]. Therefore, they require new bioinformatics approaches to overcome their complex errors and modalities. The development of bioinformatics approaches that could take full advantage of long-read sequencing has become one of the most important issues in bioinformatics. We now have various tools for base calling, error correction and polishing, de novo genome assembly, mapping, and phasing using long-read data. Here we present an overview of the analysis pipelines for PacBio and ONT data and highlight popular tools in cancer research (Table 2). There are also many existing tools that detect variants, gene isoform resolution and epigenetic modifications from long-read. Recent reviews focusing on long-read bioinformatics tools can be found in the literature [68], [69], [70]. For example, Amarasinghe et al. [70] presented an overview of the analysis pipelines and popular tools for TGS data. They also introduced a complementary open-source catalogue of long-read analysis tools: long-read-tools.org, which allows users to search and filter tools based on various parameters.

Table 2:

Typical long-read analysis pipelines for PacBio and ONT data.

Application	PacBio	ONT
Basecalling	Generate CCS [71]	Guppy [92]
Quality control	Isoseq3 [71]	NanoQCa, NanoStata
Read error correction
Non-hybrid methods	LoRMA [59]	Canu [198], LoRMA
Hybrid methods	LORDEC [199], proovread [200], LSC [201], FMLRC [202]	Nanocorr [203], FMLRC
Polishing	Arrow [71], Raconb	Nanopolish [110], Racon
Alignment	Pbmm2 [71], BLASR [71], minimap2c	Pbmm2
Structural variation analysis	CORGi [81], PBHoney [82], pbsv [71], Sniffles [83], SMRT- SV [84], SVIM-asm [85]	NanoSV [86], Picky [87], Sniffles, SVIM-asm
Isoform detection	IsoSeq3, Cupcaked, IsoCon [88]	FLAIR [90], Pinfish [92]
Isoform quantification	Salmon [91], FLAIR, featureCounts [93], Wub [92]
Differential analysis of isoform	DEseq2 [204], limma [205], edgeR [206]
APA events detection	TAPIS [97], PRIPI [98]
Fusion transcript detection	IDP-fusion [151]
Base modifications detection	SMRT LINK, Tombo [112]	NanoMod [111], Tombo, Nanopolish, signalAlign [113], D-Nascent [114], DeepSignal [115], mCaller [116], DeepMod [117]
circRNA transcripts detection	PRAPI	isoCirc [192], CIRI-long [193]

^aNanopack, https://github.com/wdecoster/nanopack, accessed June 9, 2021. ^b https://github.com/isovic/racon, accessed June 9, 2021. ^c https://github.com/lh3/minimap2, accessed June 9, 2021. ^d https://github.com/Magdoll/cDNA_Cupcake, accessed June 9, 2021.

The challenges in the development of bioinformatics tools

Basecalling

The first step in any long-read analysis is basecalling, the computational process of translating raw data to nucleotide sequences. This step is more standardized and usually performed using proprietary software. SMRT sequences detect fluorescence events that correspond to the addition of one specific nucleotide by a polymerase. The template of SMRT sequencing is circular, and the polymerase go over the DNA fragment multiple times. During basecalling process, a continuous long-read is converted from the pulses, then split into subreads. These subreads are aligned together, and derive a circular consensus sequence using workflow CCS [71]. ONT sequencers measure the ionic current fluctuations when single-stranded nucleotide passes through the nanopores. Basecalling for ONT is more complicated than PacBio basecalling, and more options are available. Wick et al. [72] examined the performance of different basecalling tools, and Guppy basecaller performs well overall, with good accuracy and fast performance.

Error correction

Despite increasing accuracy of both TGS platforms, error correction remains an important step in long-read analysis pipelines. In the case of SMRT sequencing, the CCS quality is heavily dependent on the number of times the fragment is read. As mentioned above, the random errors can be mitigated by increasing the read depth using CCS. Long HiFi reads with an average length of 13.5 kb generated using the CCS mode on the PacBio Sequel Systems, can provide base-level resolution with >99.8% single-molecule read accuracy [73, 74]. CCS reads enable structural variant detection and de novo assembly at similar contiguity. However, CCS reads retain a major residual error and exhibit a bias for indels in homopolymers [74]. Furthermore, the current tools, such GATK, which was designed for short-reads, does not properly model the CCS error profile. For ONT on the other hand, indels and substitutions are frequent in nanopore data. The error profile has been reported to be biased, which tend to occur in homopolymer regions [66]. These error characteristics pose a challenge for long-read data analyses, and error correction algorithms are needed to be developed for fixing sequencing error.

Two groups of methods to error correct long-reads can be employed: methods that only use long-reads (non-hybrid methods) and methods that leverage the accuracy of additional short-read data (hybrid methods). A few studies have showcased comparisons among rapidly evolving error correction algorithms to evaluate the quality and computational resource requirements of these tools [75], [76], [77]. Non-hybrid methods perform self-correction with long-reads alone. All reads are first aligned to each other and generate consensus sequences using overlap information. This consensus sequences are used to correct individual reads, which can be taken forward for assembly or other applications. However, the performance of non-hybrid methods deteriorated significantly when sequencing depth was decreased. Effective error correction requires high sequencing coverage, which needs both cost and time consumed during sequencing and analysis. Hybrid error correction strategy, which uses short-reads is still outperforming long-read-only correction. The same biological sample must be sequenced using both technologies in certain applications require high base-level accuracy. Hybrid error correction methods can be further divided into two categories according to how short-reads are used, short-read-alignment-based methods and short-read-assembly-based methods, respectively. In short-read-alignment-based methods, the short-reads are directly aligned to the long-read using a variety of aligners, to generate corrected long-read. In short-read-assembly-based methods, the short-reads are first used to generate contigs using an existent assembler, or only build the de Bruijn graph (DBG). Then the long-read are corrected by aligning to the assembly or by traversing the DBG.

Compared with non-hybrid methods, hybrid methods aided by short accurate reads can achieve better correction quality, especially when handling low coverage-depth long-read. The relative cost per base pair using TGS is still several folds higher when compared to the NGS. Users are recommended to choose hybrid methods in certain applications. Despite continuous improvements in the accuracy of long-read, error correction remains indispensable in many applications. Complexities resulting from GC-rich regions, tandem repeats and highly variable gene complexes cannot be accurately sequenced using short-reads sequencing [78]. Therefore, repeats, or complex regions may not be correctly handled using hybrid methods. For future work, removing the need for short-reads, better and more efficient self-correction algorithms are expected to reduce the cost and complexity of genomic projects.

Structural variation analysis

Long-read help to increase the detection of SVs as they considerably ease de novo genome assembly and mapping. Two recent reviews described the algorithms of structural variant calling from long-read data in detail [79, 80]. For instance, algorithms detect SVs from SMRT data by leveraging intra-read and inter-read signatures. CORGi [81], PBHoney [82], PBSV, Sniffles [83], SMRT-SV [84] and SVIM-asm [85] detect SVs through combinations of these two signatures. Due to higher operational costs and a large input DNA requirement, long-read have been mostly applied to single-genome assemblies. In the case of ONT, the signatures to detect SVs are similar to those used in PacBio data. Callers that detect SVs from nanopore data include NanoSV [86], Picky [87], Sniffles and SVIM-asm. Compared with PacBio sequencing, ONT provides improved read lengths, lower adaptation costs and higher throughput. It is more effective to detect many SVs. However, indels are frequent in nanopore data, which make it less suitable for smaller SVs. Furthermore, evaluating the performance of long-read SV callers is complicated by the fact that benchmark data sets may be missing SVs in their annotation.

Long-read transcriptome analysis

The goal of long-read transcriptome analysis mainly consists of the following three parts, which are isoform detection, quantification, and differential analysis. PacBio supplies the IsoSeq3 analysis pipeline for the analysis of their cDNA CCS reads, allowing the assembly of full-length transcripts. Cupcake provides scripts for collapsing redundant isoforms and merging Iso-Seq runs from different batches. Iso-Con [88] and SQANTI [89] pipelines attempt to mitigate the erroneous merging of similar transcripts of the Iso-Seq pipeline. For assaying the sequences of highly-similar gene families, IsoCon detects isoforms from genes with significant alternate splicing and minor shifts in the splice junctions. In the case of ONT, both Pinfish and FLAIR are intended for nanopore data. However, full-length alternative isoform analysis of RNA (FLAIR) needs to use short-read reads to improve junction annotation [90]. In addition, their accuracy has not yet been extensively verified. Several methods, such as Salmon [91], Wub [92], featureCounts [93], FLAIR [90], etc., can be used to quantify the abundance of transcripts. However, these methods rely on a complete and accurate isoform annotation. Therefore, quantifying and performing differential expression analysis of transcript levels on the isoform instead of the gene level remains significant challenges. It is hard to decide at which point a known and a newly identified isoform, and systematically differentiate allele-specific isoform expression.

AS events, APA and alternative transcription initiation (ATI) are major processes that contribute to transcriptome diversity. ONT directed RNA sequencing was used to measure the length of poly(A) and identify a range of novel transcript isoforms including those with AS, ATI, and APA [94], [95], [96]. For the analysis of post-transcriptional regulation based on long-read sequencing, transcriptome analysis pipeline for isoform sequencing (TAPIS) pipeline [97] and Post-transcriptional Regulation Analysis Pipeline for Iso-Seq (PRAPI) [98] are two main bioinformatics tools that use PacBio reads to identify AS and APA. For instance, the survey of APA events using TAPIS is becoming a landmark for annotation in sorghum (Sorghum bicolor L. Moench) [97], moso bamboo (Phyllostachys edulis) [99], wild apple (Malus sieversii) [100], Chinese cabbage (Brassica rapa L. ssp. pekinensis) [101], cotton (Gossypium spp.) [102], Ricinus communis [103], silkworm (Bombyx mori) [104], Lateolabrax maculatus [105], red clover (Trifolium pratense L.) [106], Gnetum luofuense [107], and perennial ryegrass (Lolium perenne) [108]. In addition, PRAPI can also identify several other events, such as production of circular RNAs (circRNAs). However, at present, quantification analysis of AS or APA still depends on NGS due to the low coverage of PacBio and ONT sequencing. In the future, it is expected that both TGS can be used for quantification analysis once the throughput increases.

Base modifications detection

In SMRT sequencing, base modifications in DNA are inferred from the interpulse duration (IPD) between fluorescence pulses [54]. mA, mC and hmC in a DNA template alter the kinetic characteristics, such as the IPD between two successive base incorporations [54]. These changes of kinetic signatures can be analyzed directly via the SMRT Portal for base modification detection. However, reliable calling of these base modifications requires high sequence coverage per strand. For instance, reliable calling of 4 mA and 6 mC requires 25× coverage per strand, whereas 250× coverage is required for 5 mC and 5 mhC [109]. Such high coverage is not realistic for large genomes and does not allow single molecule epigenetic analysis. ONT sequencing detect base modifications owing to the signal shifts caused by the modified RNA or DNA bases as they pass through the nanopore [49, 110]. Several computational tools have been developed to detect base modifications on the basis of these characteristic disruptions: NanoMod [111], Tombo [112], Nanopolish [110], signalAlign [113], D-NAscent [114], DeepSignal [115], mCaller [116], and DeepMod [117]. However, these methods may suffer from a high false discovery rate, since no experimental exist systematically validate the detection of the large variety of modifications present in RNA. In addition, many modifications do not influence the TGS’s dynamics sufficiently to be detected at a useful sensitivity. Therefore, continued efforts in developing and benchmarking tools are required to obtain accurate, and complete (including base modifications) genomes and transcriptomes.

Applications of TGS technologies in cancer

Detecting complicated cancer genomes

Human genome is considered to be one of the most complete mammalian reference assemblies. However, it contains many regions of high and low complexity that have relevance to human’s disease, such as low-complexity tandem repeats, pseudogenes, high GC content, and extremely copy number variable regions [118]. Sequencing those DNA elements is difficult with short-read sequencing, and TGS is a great tool to resolve these gaps. Cancer studies using long-read information to decipher allele-resolution mutation statuses and the complete structures of complicated cancer genomes have been rapidly increasing and continuously progressing. With the application of long-read sequencing alone or in combination with more accurate conventional short-read sequencing, it is possible to detect cancer mutations occurring at a low frequency. For example, PacBio has been applied to screen for the emergence of actionable mutations in the tumor suppressor TP53 [119]. In acute myeloid leukemia (AML) and myelodysplastic syndrome (MDS), many patients harbor multiple TP53 mutations in their tumors, and these TP53 variants are located in different alleles. Further, long-read are also utilized in the detection of genomic mutations at single-allele resolution. For instance, Suzuki et al. [120] applied long-read MinION and found that 72% of the reads harbored two epidermal growth factor receptor (EGFR) mutations (T790M and L858R), and 22% of the reads harbored neither mutation in H1975 cells (lung adenocarcinoma cell line). Then, they also detected aberrantly spliced RNAs in neurofibromatosis type 1 (NF1) and gene fusion transcripts, such as coiled-coil domain containing 6-rearranged during transfection (CCDC6-RET) and echinoderm microtubule-associated protein-like 4-anaplastic lymphoma kinase (EML4-ALK). Finally, they successfully applied this developed approach to characterize the mutation genotypes of eight clinical samples.

Particularly for cancer applications, cancer-associated SVs, such as large insertions, deletions, inversions, duplications, and translocations of variable genomic sequences, could be detected by long-read sequencing approaches [121], [122], [123]. Considering the limitations of NGS, long-read sequencing can improve the validation, resolution, and classification of germline SVs [124]. For example, in chromophobe renal cell carcinoma, structural alterations in the telomerase reverse transcriptase (TERT) promoter region identified by WGS analysis can be validated by PacBio sequencing [125]. Although short-reads can also be assessed SVs, large and complex SVs and repetitive regions cannot be detected this way [43, 46]. Norris et al. [122] applied nanopore sequencing to detect a series of well-characterized SVs and successfully identified cancer-related SVs in the cyclin dependent kinase inhibitor 2A (CDKN2A) and Mothers against decapentaplegic homolog 4 (SMAD4) genes at a low level in pancreatic cancer. Interestingly, they demonstrated that nanopore sequencing can detect these SVs at dilutions as low as 1:100, with as few as 500 reads per sample, which indicates that this technology could become an ideal tool for the low-level detection of cancer-associated SVs. In addition, Williams et al. [126] applied targeted nanopore sequencing and identified ABCB1 structural variants in THP-1 AML cells and high-grade serous ovarian cancer cells. In the context of non-small-cell lung cancer, Sakamoto et al. [127] found that long-read sequencing (PromethION) was particularly useful for precisely identifying and characterizing structural aberrations and identified several medium-sized structural aberrations, consisting of complex combinations of local duplications, inversions, and microdeletions, in lung cancer cell lines and clinical samples. In the context of breast cancers, a pioneering study by Schatz’s group involved the sequencing of the SK-BR-3 breast cancer cell line genome using PacBio SMRT long-read sequencing and demonstrated that amplification of the Erb-B2 receptor tyrosine kinase 2 (ERBB2) oncogene (also known as HER2) appeared within complex rearrangements, which can only be precisely identified by long-read sequencing [123]. Further, they sequenced SK-BR-3 cells and patient-derived organoids representing tumor and matched normal cells from two breast cancer patients via ONT, PacBio, and Illumina/10× Genomics for the comprehensive analysis of SVs. Interestingly, they found that long-read sequencing allowed for substantially more accurate and sensitive SV detection and that hundreds of variants within known cancer-related genes were detectable only through long-read sequencing [128]. Recently, Lin et al. [129] identified five genomic regions on 17q as potential hotspots of chromothripsis in breast cancer. Nanopore sequencing further detected translocations between chromosomes 17q23 and 20q13 and confirmed complex rearrangements between them that harbor a dense estrogen receptor α (Erα) hub and their corresponding target loci, respectively. These findings highlight the need for long-read sequencing to enable an in-depth analysis of how SVs disrupt the genome, and they also shed new light on the complex mechanisms involved in cancer genome evolution.

Characterization of AS

AS in particular is known to affect more than half of all human genes [130]. However, assessment of the differences in mRNA isoform expression between tissues and determination of which mRNA splice isoforms are potentially deleterious can be challenging [131]. TGS of long-read has the potential to identify and quantify isoforms simply by sequencing cDNA or mRNA molecules end-to-end from 3′ polyA tail to 5′ cap. For example, de Jong et al. [132] performed MinION nanopore sequencing of long-range PCR amplicons to identify 20 novel breast cancer gene 1 (BRCA1) isoforms, 18 of which contained multiple individual splicing events, and found that these events can co-occur within single transcripts. In addition, nanopore sequencing can be used in single-cell analysis [133], [134], [135], [136], [137]. Singh et al. [133] described a rapid high-throughput method to sequence full-length transcripts using targeted capture and Oxford nanopore sequencing of T-cell receptor and B-cell receptor mRNA transcripts and linked this with short-read transcriptome profiling. They revealed the clonal and transcriptional landscape of lymphocytes at single-cell resolution. This novel method, termed Repertoire and Gene Expression by Sequencing (RAGE-Seq), offers a new genomic toolkit for advanced single-cell analysis.

Particular splicing events are associated with many cellular processes, such as cellular growth, differentiation, tissue development, and oncogenesis [131]. Aberrant splicing events frequently occur in cancer and are associated with the hallmarks of cancer [138]. The importance of studying connections between AS and cancer is underscored by the possibility that some specific splice isoforms drive the oncogenic process and could represent attractive therapeutic targets. Nanopore sequencing has been used to characterize the global transcriptome signatures of mitochondrial and ribosomal gene expression in human cancer stem-like cell populations, which might provide a basis for the application of additional pathway-directed therapies such as those targeting mitochondria and ribosomes [139]. Moreover, unique splice variants or sets of splice variants are strongly associated with particular types of cancers and have diagnostic and prognostic value [138]. For example, RNA immunoprecipitation with a LINE 1-specific antibody followed by nanopore sequencing detected LINE1 transcripts of 90 individual elements in VM-Cub-1 UC cells. Further study showed that the expression of the individual variant long interspersed element 1 (LINE1s) is highly heterogeneous among cancer types [140].

Mutations in the splicing factor spliceosome factor 3b (SF3B1) have been associated with characteristic alterations in splicing. In a recent study, nanopore technology was applied to resequence a subset of tumor samples and normal samples from chronic lymphocytic leukemia (CLL) patient with wild-type SF3B1 or the K700E mutation and to sequence normal B-cell samples [90]. They demonstrated differential 3′ splice site changes associated with SF3B1 mutation, which is consistent with the known effects of SF3B1 mutation. They also observed a strong downregulation of intron retention events associated with SF3B1 mutation. This study of primary CLL samples by nanopore sequencing demonstrates the ability of the nanopore approach to identify and quantify cancer-specific transcript variants.

Long-read can completely cover full-length transcript sequences, so long-read sequencing is a superior approach over short-read RNA-Seq in detecting AS transcripts or transcript isoforms. Kohli et al. [141] applied modified RNA-Seq and found frequent coexpression of androgen receptor variant (AR-V) 9 and AR-V7 in prostate cancer. They further performed SMRT isoform sequencing (Iso-Seq) with a PacBio RSII to determine the full sequence of each AR mRNA transcript. In this study, they identified a common shared 3′ terminal exon as the molecular basis for frequent AR-V7 and AR-V9 coexpression in castration-resistant prostate cancer (CRPC). AR-V7 has been studied as a potential biomarker for drug resistance in prostate cancer. Thus, AR-V9 may also be a predictive biomarker for resistance. They further performed long-read RNA-Seq to analyze the effects on the expression of AR and truncated AR variants and found that AR gene rearrangements correlated with AR overexpression. Tumor-specific overexpression of AR-Vs indicated resistance to endocrine therapies, indicating that AR gene rearrangements are an important mechanism of resistance to endocrine therapies in CRPC [142]. Long-read SMRT sequencing also revealed that alternative isoforms and tumor-specific isoforms that arise from aberrant splicing are common during liver tumorigenesis. More excitingly, unannotated variants of ARHGEF2 (v1 and v3) were found to have biological significance in underscoring two major cancer hallmarks [143].

The splicing pattern of specific isoforms of numerous genes is altered as cells move through the oncogenic process. For instance, BRCA1 associated RING domain 1 (BARD1) interacts with BRCA1 and may act as a potent tumor suppressor. However, BARD1 splice isoforms show effects that are antagonistic to those of BARD1-full-length and have been associated with disease progression and a poor prognosis in multiple cancer types. Walker et al. [144] presented a comprehensive BARD1 mRNA splicing landscape by performing nanopore sequencing and splicing assays for 12 tissue types (normal and cancer tissue). Similarly, a recent study used a combination of nanopore sequencing, RNA-Seq, and RT-qPCR analyses to identify the details of BARD1 AS in melanoma [145]. These studies include the most comprehensive assessments of BARD1 mRNA splicing to date and provide information for the convenient assessment of the roles of these isoforms in human biology.

Indeed, in a pioneering study by Oka et al. [146], they performed MinION full-length DNA sequencing to characterize the alternatively spliced isoforms of lung cancer cell lines and then biologically validated the alternatively spliced isoforms. These aberrant transcripts were then found in non-small-cell lung cancer specimens. Interestingly, the authors applied liquid chromatography with tandem mass spectrometry (LC/MS/MS) and demonstrated that at least some alternatively spliced isoforms were truly translated into peptides and could play a role in producing neoantigens in cancer. Most importantly, these peptides derived from splicing isoforms and frameshift mutations could activate the T cell response through interaction with human leukocyte antigens (HLAs). These results clearly show that long-read sequencing can be used to identify novel isoforms and neoantigens that may be overlooked with the current short-read sequencing approaches. To improve the diversity of captured full-length isoforms, Hu et al. [147] developed a normalized single-molecule RNA-Seq method, and identified new cancer-specific transcriptome signatures in human gastric signet-ring cell carcinoma. This method can capture 3.2–6.0-fold more full-length high-quality isoform species for different human samples than the non-normalized capture procedure and provides a new option for specific projects. Taken together, these papers clearly highlight the potential of TGS to identify new isoforms and isoform features, which is essential for the precise identification of aberrant transcript structures in cancer cells.

Identification of fusion genes

Fusion transcripts are the result of a trans-splicing event that joins two separately encoded pre-RNAs into one transcript [148] and are known to be major driver events for carcinogenesis in several types of cancers. Many fusion genes are strong driver mutations in neoplasia and have provided fundamental insights into the disease mechanisms that are involved in tumorigenesis [149]. Many fusion genes are extremely likely to produce unique tumor neoantigens that are recognizable by immune cells; thus, they are an ideal marker for the selection of immune checkpoint inhibitors [150]. Compared with the existing tools, the integration of TGS long-read and NGS short-reads (named Isoform Detection and Prediction [IDP] fusion) to detect fusion genes provides a higher precision and a very low false positive rate [151]. These approaches can also be readily applied to the analysis of cancer-associated fusion transcripts. For instance, using long-read MinION, Suzuki et al. identified cancerous mutations in lung cancer cells and clinical samples and detected the major driver genes, which have diverse patterns, including point mutations and fusions [120]. A recent study identified TTYH1-C19MC fusions leading to the overexpression of microRNAs in embryonal tumor with multilayered rosettes. This effect in turn drives the expression of a brain-specific DNMT3B isoform and promotes tumourigenesis [152]. In another report, the authors applied MinION RNA-Seq to sequence full-length transcripts in lung cancer cell lines and detected a cancer driver fusion transcript of the CCDC6-RET gene from the LC2/ad cell line [153]. In prostate cancer, one study identified a novel fusion transcript comprising the RLN1 and RLN2 genes. The fusion transcript encodes a putative Relaxin 2 (RLN2) with a deleted secretory signal peptide, indicating a potentially biologically important alteration [154].

Transcripts with aberrant structures are extremely likely to produce chimeric proteins and serve as specific targets for treatment [149]. For example, chronic myelogenous leukemia (CML) is a blood cancer that is caused by a translocation between chromosomes 9 and 22, giving rise to BCR-ABL1. Imatinib, a potent inhibitor of the oncogenic tyrosine kinase BCR-ABL, has shown remarkable clinical activity in patients with CML [155]. Several studies have reported that PacBio sequencing and ONT sequencing can be applied to detect BCR-ABL1 fusion events and related tyrosine kinase inhibitor (TKI) resistance mutations in samples from CML patients both at diagnosis and during follow-up [156], [157], [158]. These technologies are also important for identifying other components involved in CML pathogenesis to can be exploited to overcome tyrosine kinase inhibitors (TKI) resistance. In addition, Jeck et al. [159] developed a nanopore-based sequencing assay that can decrease the turnaround time for the detection of BCR-ABL1 fusion transcripts, which may be a valid approach for laboratories with low specimen volumes and for cases in which rapid results are needed.

In the context of AML, FMS-like tyrosine kinase 3 (FLT3) mutation is the most common genetic alteration in patients, and most of these mutations are constitutively activating internal tandem duplication (ITD) mutations. Shah and colleagues applied SMRT sequencing and first confirmed the presence of activating mutations in FLT3 (FLT3-ITD), which are associated with a poor prognosis in AML. Secondary kinase domain (KD) mutations in FLT3-ITD can cause preclinical and acquired clinical resistance to the highly potent type II FLT3 inhibitor quizartinib [160]. Further Shah et al. [161] described the cocrystal structure of FLT3 with the TKI quizartinib and identified a novel FLT3 inhibitor, PLX3397, that retains activity against the F691L mutant. Further, FUsion Detection from Gene Enrichment (FUDGE) was developed to accurately identify fusion genes from low-coverage nanopore sequencing within 2 days. In this assay, Cas9-targeted enrichment of fusion genes is performed, and then the unknown fusion partner and precise breakpoint are identified by nanopore sequencing. FUDGE enables multiplexed enrichment for the simultaneous analysis of several genes in multiple samples in one sequencing run. The application of this assay in the clinic could allow for rapid gene fusion detection [162]. Hence, TGS can now be applied for gene fusion detection as a diagnostic and prognostic tool for therapy initiation and minimal residual disease monitoring following treatment.

Characterization of exogenous RNA

Although cancer is generally considered to be a disease of host genetics and environmental factors, microorganisms are implicated in ∼20% of human malignancies [163]. Microbes and microbiota can contribute to cancer development and progression and the responsiveness to cancer therapeutics [164]. It has been demonstrated that TGS has tremendous potential utility for WGS. The MinION platform was used with a 6-h sequencing run time, and sufficient data were generated to identify bacterial and viral samples down to the species level, which suggests that TGS can accurately identify and differentiate both viral and bacterial species present within biological samples via amplicon sequencing [165]. Hepatitis B virus (HBV) infection is the main cause of hepatocellular carcinoma (HCC) worldwide [166]. HBV can integrate into human DNA and promote carcinogenesis by insertional mutagenesis or by promoting genomic instability [167]. Recently, integrative analysis of HBV genomes based on NGS and TGS of tumor and nontumor liver tissues from HCC patients provided a comprehensive view of the integration process in liver tissues. Interestingly, replicating HBV DNA was more frequently detected in nontumor tissues than in tumor tissues and was associated with a higher number of nonclonal integrations. More importantly, integration of viral enhancers near a cancer driver gene may lead to strong overexpression of oncogenes. HBV integration can drive carcinogenesis by altering cancer driver genes (TERT, TP53, MYC) at a distance, and the number of HBV integration events is an independent prognostic factor in HBV-related HCC [168]. Similarly, Tatkiewicz et al. [169] assessed relative provirus expression in HERV-K (HML-2) via both short- and long-read sequencing in three mantle cell lymphoma cell lines (JVM2, Granta519 and REC1) and observed a strong tissue-specific pattern of provirus expression. Therefore, the development of TGS has enabled more precise characterization of the role of viral and bacterial genetic material in cancer diagnosis and prognosis.

Increasing numbers of studies have highlighted the key role of gut microbiota in mediating tumor responses to chemotherapeutic agents and immunotherapies targeting programmed death-ligand 1 (PD-L1) or cytotoxic T lymphocyte-associated protein 4 (CTLA-4) [170], [171], [172]. Nasal microbiota results at the genus level were compared using Illumina vs. nanopore 16S rRNA gene sequencing, and long-read sequencing had a higher efficiency than short-read sequencing in terms of the taxonomic classification of gut microbiota at the species level [173]. Therefore, this technology can be further adopted for gut microbial community research. The gastric pathogen Helicobacter pylori is the main causative agent for gastric cancer and gastric and duodenal ulcers [174]. In a study by Devi et al. [175], the authors demonstrated that low Bifidobacterium abundance among the lower gut microbiota is associated with H. pylori-related gastric ulcers and gastric cancer, indicating that long-read sequencing may serve as a noninvasive assessment method. In another study, Tetz et al. [176] used a combination of short-read and MinION long-read sequencing technologies to draft a complete genome sequence of Kluyvera intestine sp. nov. isolated from the stomach of a patient with gastric cancer. The identification of antibiotic resistance genes would enable us to understand the possible pathogenicity of these bacteria and their role in cancer. Interestingly, Gaiser et al. [177] performed real-time full-length 16S rRNA gene sequencing (PacBio single-molecule sequencing) on paired cyst fluid and plasma from patients with suspected pancreatic cystic neoplasms to assess the microbial composition and diversity in the cyst fluid. They showed that the intracystic bacterial DNA and interleukin-1β concentrations were significantly elevated and positively correlated with intraductal papillary mucinous neoplasm incidence and neoplastic grade. Recent work has also documented that the gut microbiota is related to the occurrence and development of colorectal cancer (CRC) [178]. Thus, classification of the gut microbial community in CRC should be applied in clinical settings to predict CRC development. MinION platforms for 16S rRNA sequencing have been applied to detect and classify microbial communities, and the MinION sequencing platform coupled with the corresponding algorithm could function as a practicable strategy for classifying the bacterial community down to the species level [179]. The authors further assessed the gut microbiota in clinical subjects, including healthy participants and CRC patients, and found significant differences in gut microbial communities between patients with adenomas and healthy subjects [180]. Taken together, these findings indicate that these technologies can be applied for classifying differential gut microbial communities in distinct clinical specimens and will be useful tools for rapid screening.

Direct identification of epigenetic marks in DNA and RNA

DNA modifications play an essential role in the regulation of a variety of biological processes, and deregulation of the epigenetic machinery has been directly implicated in tumorigenesis [181]. These epigenetic modifications can be interrogated directly by instruments from both PacBio [54, 182] and ONT [113]. For instance, long-read sequencing, including ONT and PacBio sequencing, can be used to concurrently assess the CpG methylation of novel and extant transposable element (TE) insertions in the hippocampus and heart, as well as in paired tumor and nontumor liver samples [183]. One study developed transposons from long DNA reads (TLDR) software to interrogate the methylation patterns of both nonreference and reference TE insertions and found pronounced demethylation of young long interspersed element 1 (LINE-1) retrotransposons in cancer. Finally, the authors recovered the complete sequences of tumor-specific LINE-1 insertion and demonstrated their retrotransposition functions. This approach demonstrated that long-read sequencing can simultaneously survey the epigenome and detect somatic TE mobilization.

In addition, McKelvey et al. [184] applied nanopore Cas9-targeted sequencing (nCATS), to characterize allele-specific methylation in thyroid cancer cell lines heterozygous for the TERT promoter mutation. They found that the mutant TERT promoter allele was significantly less methylated than the wild type allele and that the transcriptional activators GABPA and MYC bind only to the mutant TERT allele. Importantly, epigenetic information may have direct clinical value, as demonstrated in the study by Euskirchen and colleagues [121]. A study was designed to achieve same-day detection of IDH1, IDH2, H3F3A, TP53 and TERT promoters CNVs and methylation profiles using the nanopore MinION approach. A significant correlation was observed in the outcomes of nanopore sequencing and data generated from short-read exome sequencing, Sanger sequencing, SNP array, and/or genome-wide methylation microarray. Overall, the ONT method can be applied for precision medicine development for cancer patients in setting with limited resources within a short period of time and in a cost-effective manner. Moreover, Wongsurawat et al. [185] demonstrated that nCATS can be used to identify IDH1 and IDH2 mutations and simultaneously evaluate MGMT methylation levels not only at the promoter region but also at CpGs across the proximal promoter region within 2 days of surgical resection in fresh biopsies of diffuse glioma at high resolution. The nCATS technique provides a promising tool for enhancing precision cancer medicine with the potential for simultaneously assessing multiple molecular targets.

Furthermore, epigenetic modification affects both DNA and RNA, and these base modifications can have a functional effect on transcription and translation [186]. A study showed that direct RNA sequencing could be applied to detect 6 mA RNA modifications with high accuracy in terms of systematic errors and decreased base-calling qualities [49]. Therefore, direct RNA modification analysis by nanopore sequencing is rapidly developing and improving in reliability and has the potential to provide a complete view of RNA modifications such as N6-methyladenosine (6mA), 7-methylguanosine (7mG), and 5mC. However, extracting RNA modification information from ONT reads is still an unsolved challenge, and this technology has still not reached maturity for routine application in RNA epitranscriptomics.

Identification of non-coding RNAs

Non-coding RNAs (ncRNAs), which include long non-coding RNAs, microRNAs, and circular RNAs, represent functional regulatory molecules that control the development, promotion, and metastasis of cancers. These types of ncRNAs can be captured by long-read sequencing. For instance, nanopore-induced phase-shift sequencing (NIPSS) was developed to directly sequence microRNA, which can demonstrate single molecule sequencing of miRNA, such as the discriminations between different sequences, isoforms, and epigenetic modifications among synthetic miRNA sequences [57]. Since miRNAs are potential therapeutic targets or biomarkers in many human disease, such as Parkinson’s disease [187] and cancer [188]. Direct miRNA sequencing by NIPSS may be directly implemented in clinical applications. Further, Troskie et al. [189] sequenced RNA from normal mixed adult and foetal human tissues on a PacBio platform, and defined a complex tissue-specific pseudogene transcriptome, which can be utilized as a resource for transcriptomic analyses and to design functional screens. This study sets up a foundation for the use of long-read sequencing to comprehensively identify full-length pseudogene transcripts.

Although, circRNAs can be discovered and quantified using short-read RNA-seq data via identifying the occurrence of back-splice junctions (BSJs), the ability to reconstruct circRNAs is still limited to determine the full-length sequences and internal AS events with circRNAs [190, 191]. Long-read RNA-seq is a powerful tool for resolving full-length transcript isoforms. By combining circular reverse transcription and size selection strategy along with nanopore sequencing, isoCirc [192] and circRNA identifier using long-read sequencing data (CIRI-long) [193] are two major methods to effectively characterize the full-length circRNA isoforms. However, the application of these approach is limited at the current sequencing depth, due to the low throughput and high per-sequence cost for long-read sequencing.

Liquid biopsy

Liquid biopsy is a powerful technique that can be used to identify cancer patients to improve early diagnosis and improve intervention. In the area of NGS, it has been applied to sequence cfDNA. Unfortunately, TGS platforms are not designed for cfDNA analysis. Nevertheless, Martignano and colleagues [194] modified nanopore standard protocols to make them compatible with small cfDNA fragments. They sequenced cfDNA from cancer patients and healthy subjects, and successfully obtain a CNV profile from plasma cfDNA of cancer patients. By comparing the performance of nanopore sequencing with a standard NGS approach, nanopore sequencing has the same performance of NGS approaches. In addition, it would be possible to inspect the CNV profile in less than a working day while the run is still ongoing, which is unique to nanopore sequencing. The applications of this approach exploit the full potential of liquid biopsy for both research and clinical purposes. In the context of cervical cancer, human papilloma virus (HPV) is the major cause of the disease, and HPV16 and HPV18 are the two most prevalent high-risk HPV types worldwide [195]. Nanopore sequencing was used to detect HPV integration in cervical cancer samples and cervical liquid-based cytology samples [196, 197]. This approach could potentially be utilized as a diagnostic tool for cervical cancer.

The challenges of long-read sequencing in cancer research

As demonstrated by the multitude of applications mentioned above, TGS has several important advantages to NGS. However, there still remains significant challenges to be overcome.

As mentioned above, the long-read produced suffer a high error rate, which might hamper the accuracy of genome sequencing projects. The expression of short and long transcripts varies for each sample and each sample will only include a fraction of all transcripts in the reference annotation. This length bias is rooted in the way samples are prepared for sequencing. Thus, median-read length approaches are further constrained by challenges in sample preparation and biases in library preparation. The high accuracy rate for short-reads and the longer length of long-read can be combined to achieve better accuracy. Second, in contrast with short-read sequencing which have spent the last decade creating a large number of tools for data analysis, the tools for TGS are still in development. These challenges have necessitated new algorithms for the efficient analysis of longer reads, including isoform identification, quantification, and modification detection, etc.

Third, all current TGS approaches suffer from experimental artefacts caused by degraded DNA/RNA molecules. The integrity of DNA/RNA going into TGS experiments is most important. However, it is not always possible to obtain sufficiently large intact samples of DNA and full-length RNA from clinical samples. Surgical specimens and biopsies are commonly preserved as formalin-fixed paraffin-embedded (FFPE) tissues for histopathological staining and long-term storage. DNA/RNAs from FFPE samples are highly fragmented and damaged. The rapid growth of biobanks has enabled the collection of thousands of fresh frozen tissues. However, it is not yet clear what represents the best extraction and processing method for DNA/RNA. In our studies, we extract RNA from biobanking fresh frozen samples rely on physical disruption and trizol based protocols, followed by precipitations or column-based clean-up. The integrity of RNA was damaged, and the long RNA transcripts were degraded in some samples (the size of cDNA library was less than 1 kb). Long-read RNA-seq depend on long RNA molecules being present as full-length transcripts, so these samples were excluded for further long-read sequencing. As such, we need to carefully control the quality of the samples used for RNA extraction. Forth, TGS is ultimately limited by the number of reads available for analysis. For example, to truly explore the complexity of mammalian transcriptomes, hundreds of millions of reads covering full-length transcripts will be required per tissue or organ. Up to now, PacBio and ONT sequencers routinely generate 30 million of reads per $1,000 of human RNA sequencing. The relatively high costs associated with current single-molecule sequencing will be driven down as throughput increases.

Finally, despite all applications held by long-read sequencing technologies, they still have several barriers withholding their application in clinical sequencing settings. Most of the demonstrated applications mentioned above involved research studies. It takes a very long way to go for a transition into clinical routine. For instance, the evolving TGS platforms and associated primary analysis tools must be validated and confirmed for International Organization for Standardization (ISO) and associated administrations. In addition, the clinical adaptation is also hindered by a knowledge gap between genetic counselors and bioinformatics experts.

Conclusion and future perspectives

TGS platforms, such as those provided by PacBio and ONT, are rapidly advancing the field with improved reference genomes, more comprehensive variant identification and more complete views of transcriptomes and epigenomes (Figure 2). The use of TGS to address fundamental problems in cancer research presents an encouraging outlook for its continued application in an increasing number of different clinical and research settings (Table 3). Despite all advantages held by long-read sequencing technologies, they still have some limitations, as well. In particular, they suffer from much lower throughput and much high error rates than NGS platforms. The limitation of lower throughput makes them difficult to perform differential gene expression analysis. The key advantages of the TGS platforms are the long-read length and the ability to detect base modifications in native DNA and RNA. However, their ability to capture more of full-length transcripts are additionally dependent on having high-quality RNA libraries as input. The key advantages of the NGS platforms are their high accuracy, relative low cost, and high throughput, which make them the most popular platforms in both clinical and research settings currently. Therefore, each platform offers its own advantages that provide preference for certain applications (Table 4). In the near future, with developments in sample preparation protocols, sequencing accuracy, and computational tools, TGS long-read approaches alone or integrated with NGS short-read approaches will expand low-cost diagnostic sequencing to more loci at higher accuracies and pave the way for novel clinical applications.

Table 3:

Applications of TGS in human cancers.

Application	Sequencing technology	Targets	Cancer
Detect mutations at a low frequency	PacBio	TP53	AML and MDS [119]
Detect mutations at single-allele resolution	ONT	EGFR	Lung cancer [120]
Detect Low-level of cancer-associated SVs	ONT	CDKN2A and SMAD4	Pancreatic cancer [122]
Detect complex rearrangements	ONT	Chr 17q23 and 20q13	Breast cancer [129]
Detect full-length transcript sequences	PacBio	AR-V9 and AR-V7	Prostate cancer [141]
Comprehensive assessment of isoforms	ONT	BARD1	Melanoma and others [144, 145]
Detect neoantigens	ONT	NDST1, SENP2 et al.	Lung cancer [146]
Detect fusion genes	ONT	CCDC6-RET	Lung cancer [153]
Detect fusion genes	PacBio	RLN1 and RLN2	Prostate cancer [154]
Detect fusion genes	PacBio/ONT	BCR-ABL1	CML [156], [157], [158]
Virus integrative analysis	PacBio	Cancer driver genes	HCC [168]
Axonomic classification of gut microbiota	PacBio/ONT	NA	Gastric cancer, CRC [174], [175], [176, 178]
Transposable element epigenomic profiling	ONT	TE	HCC [183]
Same-day detection of CNVs and methylation	ONT	Cancer driver genes	Brain cancer [121]
Identification of miRNA and circRNA	ONT	NA	NA
Liquid biopsy	ONT	NA	Lung cancer [194]

NA, not available.

Table 4:

Applications most suited for NGS, PacBio, and/or ONT in cancer research.

Application	Optimal technology	Reason
WES, WGS, and GWAS	NGS	High throughput at low cost
Gene panels in cancer	NGS	High throughput at low cost
Cancer-specific biomarkers, such as MSI, TMB	NGS	High throughput at low cost
Complicated cancer genomes	PacBio or ONT	Read lengths can traverse most repeat structures of the genome
De novo genome assembly	PacBio or ONT	Long-read enable much higher N50s
Haplotype phasing	PacBio or ONT	Long-read permit direct phasing
Gene expression	NGS	RNA-seq with low cost
Identification of ncRNAs	NGS	RNA-seq with low cost
Alternative splicing/transcripts	PacBio or ONT	Full-length RNA transcripts sequencing
Identification of fusion genes	PacBio or ONT	Long-read permit characterizing SVs and full-length RNA transcripts
Epigenetics	PacBio or ONT	Direct detection of DNA modifications
RNA modification detection	ONT	Direct detection of RNA modifications
Microbial genome sequencing	PacBio or ONT	Read lengths can span the majority of the bacterial 16S rRNA gene
Single-cell RNA/DNA sequencing	NGS	High throughput at low cost