The potential impact of nanopore sequencing on human genetics

Abstract

Nanopore sequencing has been available to researchers for a little over 3 years. Recently, the milestone of sequencing and assembling a human genome on this platform was achieved for the first time. Significant improvements to the platform in yield and accuracy, coupled with higher throughput nanopore sequencers, mean that human genome sequencing at scale is now possible. Here, a brief recent history of the nanopore platform is provided, key papers and innovations are highlighted and some of the challenges for the future are discussed.

Introduction

Since the identification of DNA as being the carrier of hereditary information from one generation to the next, scientists have sought to decode the human genome (1,2). Even though the grand challenge of the $1000 genome has been achieved, arguably the ‘complete’ human genome is still not available. For example, no single human chromosome has yet been completely sequenced from telomere to telomere, although scaffolds for individual chromosomes have been achieved (3). This is a consequence of the many repeat units distributed throughout chromosomes as well as within heterochromatic centromeric and telomeric DNA. The progression towards the $1000 dollar genome has been driven by the rapid emergence of highly parallel short-read sequencing methodologies (4). These well-established methods are highly accurate but are quickly confounded by repeated sequence when the repeats exceed the length of the read (5,6). This leads to genome assemblies within which most assembled contigs are in relatively short fragments compared with the total genome length (7). Several technologies have emerged in recent years addressing this problem including optical mapping (8), chromatin capture approaches such as Hi-C (9) and single molecule long-read sequencing (10–12). Various combinations of these approaches have led to some of the most contiguous mammalian genome assemblies to date (13–15).

Here I focus on the potential impact of just one technology on human genomics, namely nanopore sequencing. In early 2012, Oxford Nanopore Technologies announced a hand-held nanopore-based sequencing device at that years AGBT conference (16). This device promised the ability to directly sequence a DNA molecule by detecting the change in current flow as a molecule of DNA traversed a nanopore embedded within a membrane (12,17,18). This device, called the MinION, would be USB powered, portable and able to sequence DNA with the simplest of library preparations. The excitement at this device masked a second platform, the GridION, which was envisioned as a scalable sequencer that could be placed within conventional computer racking environments and theoretically sequence an entire human genome in 15 min. These announcements triggered speculation, disbelief and excitement in almost equal measure through the community. Subsequent delays in bringing the product to market tempered initial enthusiasm, but in 2014 the MinION emerged into the hands of a range of researchers seeking to apply this technology to their research questions (19). These early applications exploited the portability and ease of use of the device coupled with the constraints inherent in the platform at that time. The yield from a single MinION flow cell was typically in the low hundreds of megabases range. Sequencing runs resulting in more than a gigabase of sequence data were rare. Perhaps reflecting the nature of the ‘early adopter’ many of these early results were disseminated by tweets and blog posts. This atmosphere of rapid data release has continued with many of the key papers around nanopore sequencing appearing first on BioRXiv.

Particular applications, not limited by yield, lent themselves well to the portability of the platform, for example the deployment of MinION sequencers into West Africa by Quick and colleagues during the Ebola epidemic in 2015 (20). This study enabled the real-time tracking of the Ebola virus as it spread and these data were used by the WHO in daily reporting of the outbreak. Subsequently MinION sequencing has been applied to tracing the spread of the Zika virus through Brazil (21,22), and groups are developing applications for tracking a wider range of diseases (23–26). A commonly raised criticism is that the fundamental error rate of base calling, the process by which the underlying nucleotide sequence is determined from the current trace, is too high for meaningful use. Multiple groups have tackled this problem using a variety of post base calling analyses to improve consensus quality (27–29). As long as an appropriate tool chain is used, it is possible to accurately determine SNPs from consensus data (20,30). As a result, MinION sequencing is becoming established as an easily deployable first line diagnostic and epidemiological tool in the field.

The limitation on yield from a single flow cell initially precluded the assembly of large genomes using MinION technology. Early genome datasets from single flow cells were typically bacterial (27). This restriction was dramatically lifted in early 2016 when Oxford Nanopore Technologies announced a switch in the protein nanopore being used in the MinION device (from R7 to R9, a CsgG pore) (31). In combination with updates to the software used for nanopore sequencing (MinKNOW), this resulted in an increase in yield from a single MinION flow cell. At the time of writing, yields in excess of 10 Gbp from a single MinION run are being reported by Oxford Nanopore Customers (see the Oxford Nanopore Technologies Forum). Oxford Nanopore themselves report single MinION runs in the range of 20 Gbp, although to date none have been seen in the field. This increase in yield has led a number of groups to focus on sequencing larger genomes ranging from Caenorhabditis elegans to the tomato and, ultimately, the human genome (32–35). To date, two projects have produced preprints describing their datasets (34,35). These papers provide useful insights into the potential impact of nanopore sequencing on human genetics research in the future (Table 1).

Table 1.

A non-exhaustive list of possible applications for Nanopore sequencing in human genetics in the near future

Application	Advantages/disadvantages	Ideal platform at human scale	References
Whole Genome Sequencing and assembly	Highly contiguous genome assemblies	GridION PromethION	(35,40,41,52,53)
	Ultra-long reads should further increase contiguity and could close existing gaps in assemblies
	Challenges remain for delivery of high molecular weight DNA to the sequencing nanopores
	Direct detection of DNA modifications possible
	DNA amounts required are high (single flow cell requires 1-1.5 ug DNA)
	Assembly tools computationally expensive
Structural variants	Long reads will, by definition, improve the detection of structural variants in genomes	GridION PromethION	(34,37,47)
	Ultra-long read preparations will aid structural variant detection but require high molecular weight DNA
Sequence variation	Variant calling with Nanopore data is possible, but enhanced by the use of post base calling polishing (nanopolish), or short read correction	GridION PromethION	(20,27)
	Homopolymers were not called by early Nanopore base callers. Homopolymer base callers are now available (Albacore—Oxford Nanopore Technologies)
Phasing	Phasing across ever larger regions of the human genome will be facilitated by longer reads	GridION PromethION	(34,35,51,54)
Targeted Sequencing	Targeted methods for DNA capture prior to sequencing utilising Cas9 suggest that individual regions of the genome can be captured for sequencing.In principle, selective sequencing (‘read until’) could enable dynamic selection of target during a sequencing run.	All	(48,55)
Direct Sequencing of RNA	A unique property of nanopore sequencers is the ability to directly sequence RNA, including the detection of modifications.	MinION GridION PromethION	(49,50)
	Currently, direct RNA requires 500 ng polyA-tailed mRNA as input.

Prior to this, a number of groups used the MinION to explore the utility of Nanopore sequencing for human genetics even at low coverage. Erlich and colleagues (36) demonstrated that nanopore sequencing, even at low coverage, can be used to identify a single individual from a previously sequenced cohort. With lower yield approaches Norris et al. (37) employed an amplicon-based approach to identify targeted structural variants and determined that rearrangements could be reliably detected even against a 100-fold greater background of the wild type amplicon. Mitsuhashi et al. (38) sequenced specific BACs covering the D4Z4 array on 4q35, a variant of which causes one of the most common autosomal dominant variants of muscular dystrophy, demonstrating that nanopore sequencing enables molecular diagnosis of FSHD.

However, many consider benchmarking of a sequence technology to be the human genome. Four human genomes have been sequenced to date using nanopore technology. The reference genome GM12878 at 35× (35), two patient genomes at 11× and 16× coverage (34), and the genome of Clive Brown, CTO of Oxford Nanopore Technologies, at 55× (the ‘cliveome’—http://cliveo.me/; date last accessed July 16, 2017). Initially, the GM12878 reference sequence assembled to an NG50 of 3 Mb using 30× of long reads (35). The same assembly methods were used by Koren and Phillippy to assemble the ‘Cliveome’ to an NG50 of 24.5 Mbp (https://genomeinformatics.github.io/cliveome/). Quick and colleagues (35) developed an ultra-long read protocol, defined as read length N50 approaching 100 kb, and Jain et al. (35) generated an additional 5× coverage of the reference genome using these methods (for comparison the ‘cliveome’ read N50 is computed at 15.5 kbp). This increased the genome assembly NG50 from 3 to 6.4 Mbp. Phillippy and Koren predict that a 30× ultra-long read set would assemble to an NG50 in excess of 50 Mbp (https://genomeinformatics.github.io/cliveome/; date last accessed July 16, 2017) (35). Thus, highly contiguous genome assemblies should be enabled by the read lengths currently attainable only on the nanopore platform. These datasets enable the assembly of single contigs spanning many regions of the genome that have been previously undescribed or unassembled. For example, Jain et al. (35) are able to assemble a single contig spanning all the class I HLA genes from the major histocompatibility region on chromosome 6. The ability to investigate and phase such regions is likely to be of great importance in the future.

Obviously long-read sequencing should improve structural variant detection, contiguity and phasing across the human genome (4). For read length, nanopore sequencing currently represents the leader of the field (35). Stancu et al. (34) used nanopore sequencing to investigate the genomes of two patients with congenital disease, using 16× and 11× coverage. These coverage levels were sufficient to identify and resolve all the previously identified breakpoints in both patients and identify additional previously undescribed breakpoints in the second patient. Importantly, additional breakpoint junctions that could not be detected using Illumina reads alone were identified from these Nanopore data. These experiments used significant numbers of flow cells from both R7 and R9 pore types. For patient 1, using R7 technology, 122 sequencing runs resulted in 16× coverage. For patient 2, 11× coverage could be obtained from just 13 runs. This compares well with the 39 flow cells used by Jain et al. (35) to obtain 35× coverage of the reference GM12878. Oxford Nanopore Technologies themselves used 52 flow cells to sequence the ‘Cliveome’, the genome of Clive Brown CTO, to a depth of 55× (http://cliveo.me/; date last accessed July 16, 2017).

Clearly, ultra-long read sequencing is of great interest and many would like to sequence a complete chromosome from end-to-end. Beyond the simple problem of delivering a complete chromosome intact to a single nanopore, it is worth considering that Chromosome 21, at 48 million base pairs, would take almost 30 h to pass through a single nanopore at 450 b/s (the current speed of nanopore sequencing). Chromosome 1 would take just over 6 days! Less obvious is that a component of the residual error of base calling may be due to current generation base callers not being trained to distinguish modified bases (39). The ability to directly sense modified bases has been demonstrated across a wide variety of genomes and some, although not all, of the residual error in base calling may be derived from the effects of modifications. This is an area rich in potential for future analysis and many groups are developing methods to detect such modifications and have shown that these modifications can be detected in native human genomic DNA (35,40,41).

These early studies reveal much potential for nanopore sequencing in human genetics. The ease of use and low purchase cost for the MinION enable many to sequence in a wide variety of environments for a very low cost of entry. However, although technically possible, the MinION is not the ideal tool for sequencing a human genome yet. In principle, five flow cells, each generating 20 Gbp of nanopore sequencing data, will yield just over 30× coverage of a human genome although more likely 10 flow cells would be necessary. On a single MinION, this requires 20 days of sequencing and concomitant base calling of the signal data. Oxford Nanopore is developing rapidly in this area and has a range of GPU (graphical processing unit) and FPGA (field programmable gate array) solutions to base calling in development. For human genomics, scaling up the MinION to a higher throughput device is an obvious step. In early 2017, Oxford Nanopore re-announced the GridION platform (see Table 2). This device, the first unit of which has shipped, contains the compute and sequencing hardware sufficient to run five MinION scale flow cells simultaneously. This platform is perhaps targeted to individual human genome scale sequencing, with no more than two complete runs being required to obtain a long-read 30× human genome assuming 10 Gbp per flow cell. Importantly, this device does not require additional compute for base calling. Oxford Nanopore had previously announced an even larger device, the PromethION, which contains 280 times more sequencing channels than a MinION. This device is projected to generate 50 Gbp of sequence data per flow cell at launch which equates to 25 human genomes at 30× every 48 h. These developments will enable human genome sequencing at scale on a long-read platform both within and outside conventional genome centers.

Table 2.

Yield summary for nanopore sequencers

	MinION	GridION	PromethION
Flow cell number	1	5	48
Integrated compute	No	Yes	Yes
Channels per flow cell	512	512	3000
Nanopores per flow cella	2048	2048	6000
Yield per flow cell (theoretical/reported)	20 Gbp/2.3 Gbp	20 Gbp/2.3 Gbp	50 Gbp/6.7 Gbpb
Approx time to 30× human genomec	9 days/78 days	2 days/16 days	<2 h/<7 h

Upper yield estimates are based on Oxford Nanopore published figures. Lower estimates are derived from Jain et al. (35). At the time of writing, users in the field are routinely observing yields in the range of 4–10 Gb, with outliers at 15 Gb.

a

This is a maximum value for nanopore count. Each channel can select from more than one nanopore during the course of a run.

b

No public data are yet available for the PromethION device. The lower bound estimate for yield from a single flow cell is scaled according to the number of pores per flow cell from MinION data.

c

These calculations assume the entire device is devoted to a single genome.

Many challenges remain to be addressed. Simply managing data files is a challenge with nanopore data (42). This is an area of active development and it is likely that specific file formats will change rapidly. To reduce this problem it is tempting to base call nanopore data and then keep only the resulting fastq files for subsequent analysis. However, in the rapidly changing world of nanopore sequencing this would prevent subsequent signal level analysis of the data. New generations of base callers, methods to detect modifications and any level of signal-based correction of sequence data require the raw signal data from the pore itself (27,43–46). An example of this is the observation that Nanopore base calls collapse homopolymers down. Indeed, early base callers did do this but the introduction of newer base callers has improved this issue (35). Previously generated datasets can, in principle, be recalled with refined algorithms in the future. Thus, saving only fastq data will result in the inability to extract additional information in the future. In this regard, the availability of a reference dataset from GM12878 and the sharing of the ‘cliveome’, although with some restrictions, provide useful sources of data for future tool development.

The development of appropriate bioinformatics tool chains for handling datasets at human scale is required. A comprehensive survey of nanopore tools is beyond the scope of this article (see Jain et al. (12) for useful tools), but challenges include rapid assembly, mapping of reads greater than 50 kB to a reference at scale, detection of variants and further improvements to accuracy and consensus analysis. Many appropriate tools have already been developed, but do not yet scale well to human sized nanopore datasets (12,35). In addition, nanopore technology enables experiment types that have never previously been possible. As an example, nanopore sequencing is truly real time. Data are available for analysis the moment a read has traversed a nanopore. This enables analysis to begin before an experiment has completed. Even at low levels of sequence depth it has been demonstrated that specific cancers can be classified in as little as 6 h (47). These types of targeted analyses, coupled with the real-time nature of the platform, could one day be used to dramatically speed up diagnoses in the clinic.

However, this is not the limit of real-time analysis. By reversing the voltage over an individual channel, a read can be rejected from a specific pore. In this way, the amount of data sequenced from individual regions of a genome can be controlled. This method is called ‘Read Until’ and could enable targeted sequencing of multiple regions of the human genome in real time, although early work on this was limited by genome size (48). In essence, read until converts nanopore sequencers to targeted long-read platforms with rejected reads being, by definition, short. It is easy to imagine common sites of structural variation being screened by nanopore sequencing, with the sequencer switching from long reads to short reads over each target as specific structural variants are validated. The potential for such applications is dramatically improved by the ability to sequence ultra-long reads over regions of interest.

So far, I have only considered sequencing genomic DNA. Of course, nanopore sequencing can also be applied to cDNA and even native RNA (49–51). These approaches have great potential for unlocking the phase of specific transcript variants and precisely determining splice variants. At this time direct RNA sequencing has not been available for long and few papers are available. However, no other technology can report direct measurements of RNA and enable the detection of both splice variants and modifications within the native RNA molecule, although other competing technologies do exist. It is too early to determine the impact of RNA sequencing with nanopore on more global human studies, but undoubtedly this is an area of great interest.

In the 3 years since nanopore sequencing has been available to customers, progress has been dramatic. The proof of principle that human genomes can be sequenced on a hand-held device has been achieved by multiple groups. The GridION enables sequencing of a human genome within 2–4 days at 30× coverage. The PromethION has the potential to unlock high throughput human-scale sequencing. Assuming the same library was loaded across all 48 flow cells, it would be possible to generate a human genome at 30× within 2 h. Alternatively, the ability to asynchronously load flow cells would enable a clinical laboratory to dedicate two flow cells to an individual patient and generate a genome sequence within 48 h for each patient. The implicit benefits of longer reads are shared with other long-read sequencing technologies. Even so, nanopore sequencing read lengths currently exceed that of any other technology (35). This alone will facilitate the assembly and phasing of previously intractable regions of the human genome. Longer reads will enhance our ability to detect and survey structural variants, even at lower coverage (34). Combined with the ability to directly detect modified DNA and the real-time platform, nanopore sequencing is likely to contribute to molecular diagnosis of human disease (34,37,38). Furthermore, sequencing RNA directly opens up a previously unexplored domain. The potential for nanopore sequencing is promising. It is surely only a matter of time until a complete end-to-end assembly of the human genome is available. The question is if this will require a hybrid assembly based upon multiple technologies, or can be derived from nanopore reads alone?

Acknowledgements

Many papers have been omitted from this summary due to the limitations of space. The nanopore sequencing community has been especially open with much discussion on Twitter, the Nanopore community and face-to-face. The author thanks the entire community for their openness. The author also thanks Teri Evans for comments on this manuscript.

Conflict of Interest statement. The author was a member of the MinION access program (MAP) and has received free-of-charge flow cells and kits for nanopore sequencing, and travel and accommodation expenses to speak at Oxford Nanopore Technologies conferences. The author has also received research funding from Oxford Nanopore Technologies.

Funding

ML is supported by awards from the Biotechnology and Biological Sciences Research Council (BB/N017099/1).