Coming Of Age: Ten Years Of Next-generation Sequencing Technologies

REVIEWSTemplateA DNA fragment to besequenced. The DNA istypically ligated to one or moreadapter sequences where DNAsequencing will be initiated.FragmentationThe process of breaking largeDNA fragments into smallerfragments. This can beachieved mechanically (bypassing the DNA through anarrow passage), by sonicationor enzymatically.ClustersGroups of DNA templates inclose spatial proximity,generated either thoughbead-based amplification orby solid-phase amplification.Bead-based approaches relyon emulsions to maintaintemplate isolation duringamplification. Solid-phaseapproaches rely on thetemplate-to-bound-adapterratio to probabilistically bindtemplate molecules at asufficient distance fromeach other.Flow cellsDisposable parts of anext-generation sequencingroutine. Template DNA isimmobilized within the flow cellwhere fluid reagents can bestreamed into the cell andflushed away.Rolling circle amplification(RCA). A method of DNAamplification using a circulartemplate. Briefly, DNApolymerase binds to a primedsection of a circular DNAtemplate. As the polymerasetraverses the template, a newstrand is synthesized. Whenthe polymerase completes afull circle and encounters thedouble-stranded DNA(dsDNA) template, it displacesthe template withoutdegradation, thus creatinga long ssDNA fragmentcomposed of many copiesof the template sequence.an elongating strand. In most SBL and SBS approaches,DNA is clonally amplified on a solid surface. Havingmany thousands of identical copies of a DNA fragmentin a defined area ensures that the signal can be distinguished from background noise. Massive parallelizationis also facilitated by the creation of many millions ofindividual SBL or SBS reaction centres, each with its ownclonal DNA template. A sequencing platform can collect information from many millions of reaction centressimultaneously, thus sequencing many millions of DNAmolecules in parallel.There are several different strategies used to generate clonal template populations: bead-based, solid-stateand DNA nanoball generation (FIG. 1). The first step ofDNA template generation is fragmentation of the sample DNA, followed by ligation to a common adaptorset for clonal amplification and sequencing. For beadbased preparations, one adaptor is complementary toan oligo nucleotide fragment that is immobilized ona bead (FIG. 1a). Using emulsion PCR (emPCR)8, theDNA template is amplified such that as many as onemillion clonal DNA fragments are immobilized ona single bead9. These beads can be distributed onto aglass surface10 or arrayed on a PicoTiterPlate (RocheDiagnostics)11. Solid-state amplification12 eschews theuse of emPCR in favour of amplification directly on aslide13 (FIG. 1b,c). In this approach, forward and reverseprimers are covalently bound to the slide surface, eitherrandomly or on a patterned slide. These primers provide complementary ends to which single-strandedDNA (ssDNA) templates can bind. Precise controlover template concentration enables the amplificationof templates into localized, non-overlapping clonalc lusters, thus maintaining spatial integrity. Recently,several NGS platforms have utilized patterned flow cells.By defining precisely where primers are bound to theslide, more DNA templates can be spatially resolved,enabling higher densities of reaction centre clusters and increasing sequencing throughput.The Complete Genomics technology used by theBeijing Genomics Institute (BGI) is currently the onlyapproach that achieves template enrichment in solution. In this case, DNA undergoes an iterative ligation,circularization and cleavage process to create a circulartemplate, with four distinct adaptor regions. Throughthe process of rolling circle amplification (RCA), up to 20billion discrete DNA nanoballs are generated (FIG. 1d).The nanoball mixture is then distributed onto a patterned slide surface containing features that allow a s ingle n anoball to associate with each location14.After ligation, the template is imaged and the knownbase or bases in the probe are identified16. A new cyclebegins after complete removal of the anchor–probe complex or through cleavage to remove the fluorophore andto regenerate the ligation site.The SOLiD platform utilizes two-base-encodedprobes, in which each fluorometric signal representsa dinucleotide17. Consequently, the raw output is notdirectly associated with the incorporation of a knownnucleotide. Because the 16 possible dinucleotide combinations cannot be individually associated with spectrally resolvable fluorophores, four fluorescent signalsare used, each representing a subset of four dinucleotidecombinations. Thus, each ligation signal represents oneof several possible dinucleotides, leading to the termcolour-space (rather than base-space), which must bedeconvoluted during data analysis. The SOLiD sequen cing procedure is composed of a series of probe–anchorbinding, ligation, imaging and cleavage cycles to elongate the complementary strand (FIG. 2a). Over the courseof the cycles, single-nucleotide offsets are introduced toensure every base in the template strand is sequenced.Complete Genomics performs DNA sequencingusing combinatorial probe–anchor ligation (cPAL)14or combinatorial probe–anchor synthesis (cPAS; seethe BGISEQ‑500 website). In cPAL (FIG. 2b), an anchorsequence (complementary to one of the four adaptorsequences) and a probe hybridize to a DNA nanoballat several locations. In each cycle, the hybridizing probeis a member of a pool of one-base-encoded probes, inwhich each probe contains a known base in a constantposition and a corresponding fluorophore. After ima ging, the entire probe–anchor complex is removed and anew probe–anchor combination is hybridized. Each subsequent cycle utilizes a probe set with the known base inthe n 1 position. Further cycles in the process also useadaptors of variable lengths and chemistries, allowingsequencing to occur upstream and downstream of theadaptor sequence. The cPAS approach is a modificationof cPAL intended to increase read lengths of CompleteGenomics’ chemistry; however, at present, details aboutthe approach are limited.Sequencing by ligation (SOLiD and CompleteGenomics). Fundamentally, SBL approaches involvethe hybridization and ligation15 of labelled probe andanchor sequences to a DNA strand. The probes encodeone or two known bases (one-base-encoded probes or twobase-encoded probes) and a series of degenerate or universal bases, driving complementary binding betweenthe probe and template, whereas the anchor fragmentencodes a known sequence that is complementary to anadapter sequence and provides a site to initiate ligation.Sequencing by synthesis: CRT (Illumina, Qiagen). CRTapproaches are defined by their use of terminator molecules that are similar to those used in Sanger sequen cing, in which the ribose 3ʹ‑OH group is blocked, thuspreventing elongation19,20. To begin the process, a DNAtemplate is primed by a sequence that is complementary to an adapter region, which will initiate polymerasebinding to this double-stranded DNA (dsDNA) region.During each cycle, a mixture of all four individuallylabelled and 3ʹ‑blocked deoxynucleotides (dNTPs) areSequencing-by-synthesis categories. SBS is a term usedto describe numerous DNA-polymerase-dependentmethods in the literature, but it does not delineate thedifferent mechanisms involved in SBS approaches.For this article, SBS approaches will be classifiedeither as cyclic reversible termination (CRT) or as single-nucleotide addition (SNA)18.NATURE REVIEWS GENETICSVOLUME 17 JUNE 2016 02

REVIEWSa Emulsion PCR(454 (Roche), SOLiD (Thermo Fisher), GeneReader (Qiagen), Ion Torrent (Thermo Fisher))On-bead amplificationTemplates hybridize to bead-bound primers and are amplified;after amplification, the complement strand disassociates,leaving bead-bound ssDNA templatesEmulsionMicelle droplets are loadedwith primer, template,dNTPs and polymeraseb Solid-phase bridge amplificationFinal product100–200 million beads withthousands of bound templatec Solid-phase template walking(Illumina)(SOLiD Wildfire (Thermo Fisher))Template bindingFree templates hybridizewith slide-bound adaptersBridge amplificationDistal ends of hybridized templatesinteract with nearby primers whereamplification can take placeTemplate bindingFree DNA templates hybridizeto bound primers and thesecond strand is amplifiedPrimer walkingdsDNA is partially denatured,allowing the free end tohybridize to a nearby primerTemplate regenerationBound template is amplifiedto regenerate free DNAtemplatesCluster generationAfter several cycles ofamplification, clusters on apatterned flow cell aregeneratedCluster generationAfter several rounds ofamplification, 100–200 millionclonal clusters are formedPatterned flow cellMicrowells on flow celldirect cluster generation,increasing cluster densityd In-solution DNA nanoball generationRolling circle amplificationCircular templates are amplified to generated longconcatamers, called DNA nanoballs; intermolecularinteractions keep the nanoballs cohesive andseparate in solution(Complete Genomics (BGI))CleavageCircular DNAtemplatesare cleaveddownstreamof the adaptersequenceAdapter ligationOne set of adaptersis ligated to eitherend of a DNAtemplate, followedby templatecircularizationN Iterative ligationThree additionalrounds of ligation,circularization andcleavage generate acircular template withfour different adaptersFigure 1 Template amplification strategies. a In emulsion PCR,fragmented DNA templates are ligated to adapter sequences and arecaptured in an aqueous droplet (micelle) along with a bead covered withcomplementary adapters, deoxynucleotides (dNTPs), primers and DNApolymerase. PCR is carried out within the micelle, covering each bead withthousands of copies of the same DNA sequence. b In solid-phase bridgeamplification, fragmented DNA is ligated to adapter sequences and bound toa primer immobilized on a solid support, such as a patterned flow cell. The freeend can interact with other nearby primers, forming a bridge structure. PCRis used to create a second strand from the immobilized primers, and unboundDNA is removed. c In solid-phase template walking154, fragmented DNA isligated to adapters and bound to a complementary primer attached to a solidsupport. PCR is used to generate a second strand. The now double-stranded3 2 1 HybridizationDNA nanoballs areimmobilized on apatterned flow celltemplate is partially denatured, allowing the free end of the original templateto drift and bind to another nearby primer sequence.areNatureReverseReviewsprimers Geneticsused to initiate strand displacement to generate additional free templates,each of which can bind to a new primer. d In DNA nanoball generation,DNA is fragmented and ligated to the first of four adapter sequences. Thetemplate is amplified, circularized and cleaved with a type II endonuclease.A second set of adapters is added, followed by amplification, circularizationand cleavage. This process is repeated for the remaining two adapters. Thefinal product is a circular template with four adapters, each separated by atemplate sequence. Library molecules undergo a rolling circle amplificationstep, generating a large mass of concatamers called DNA nanoballs, whichare then deposited on a flow cell. Parts a and b are adapted from REF. 18,Nature Publishing Group.336 JUNE 2016 VOLUME silbuPnallimcaM6102

REVIEWSOne-base-encoded probesOligonucleotides that containa single interrogation base in aknown position. The basecorresponds to a fluorescentlabel on each probe. Theremaining bases are eitherdegenerate (any of the fourbases) or universal (unnaturalbases with nonspecifichybridization), allowing theprobe to interact with manydifferent possible templatesequences.Two-base-encoded probesOligonucleotides that containtwo adjacent interrogationbases in a known position.The bases correspond to afluorescent label on eachprobe. The remaining basesare either degenerate (any ofthe four bases) or universal(unnatural bases withnonspecific hybridization)allowing the probe to interactwith many different possibletemplate sequences.Colour-spaceA system exclusively usedby SOLiD. When atwo-base-encoded probe isused, the bound labelcorresponds to two basesrather than one. Thus, thesignal derived from a SOLiDrun is in a series of coloursthat represent overlappingdinucleotides, rather than eachcolour being directly correlatedto a single base. A referencebased alignment is the mostefficient way to translatecolour-space into base-space.For example, in the sequenceATGT the first probe will matchAT, the second will match TGand the third GT. If the AT isknown, then the subsequentcolour order is uniquely solvedas TG and GT, leading to areadout of ATGT. Finalsequence deconvolution ofcolour-space is achieved withthe knowledge of the secondbase identity in one round andthe colour of the subsequentround in which the ligation isoffset by one nucleotide,allowing for the identificationof the next base.Base-spaceA system used by mostnext-generation sequencingplatforms. When aone-base-encoded probe ora sequencing-by‑synthesisapproach is used, each signal iscorrectly correlated to a base.added. After the incorporation of a single dNTP to eachelongating complementary strand, unbound dNTPs areremoved and the surface is imaged to identify whichdNTP was incorporated at each cluster. The fluorophore and blocking group can then be removed and anew cycle can begin.The Illumina CRT system (FIG. 3a) accounts for thelargest market share for sequencing instruments compared to other platforms21. Illumina’s suite of instruments for short-read sequencing range from small,low-throughput benchtop units to large ultra-highthroughput instruments dedicated to population-levelwhole-genome sequencing (WGS). dNTP identificationis achieved through total internal reflection fluorescence (TIRF) microscopy using either two or four laserchannels. In most Illumina platforms, each dNTP isbound to a single fluorophore that is specific to thatbase type and requires four different imaging channels, whereas the NextSeq and Mini-Seq systems use atwo-fluorophore system.In 2012, Qiagen acquired the Intelligent BioSystemsCRT platform, which was commercialized andrelaunched in 2015 as the GeneReader 22 (FIG. 3b) .Unlike other systems, this platform is intended to be anall‑in‑one NGS platform, from sample preparation toanalysis. To accomplish this, the GeneReader system isbundled with the QIAcube sample preparation systemand the Qiagen Clinical Insight platform for vari ant ana lysis. The GeneReader uses virtually the same approachas that used by Illumina; however, it does not aim toensure that each template incorporates a fluorophore- labelled dNTP23. Rather, GeneReader aims to ensurethat just enough labelled dNTPs are i ncorporated toachieve identification.Sequencing by synthesis: SNA (454, Ion Torrent). UnlikeCRT, SNA approaches rely on a single signal to mark theincorporation of a dNTP into an elongating strand. As aconsequence, each of the four nucleotides must be addediteratively to a sequencing reaction to ensure only onedNTP is responsible for the signal. Furthermore, thisdoes not require the dNTPs to be blocked, as the absenceof the next nucleotide in the sequencing reaction prevents elongation. The exception to this is homopolymerregions where identical dNTPs are added, with sequenceidentification relying on a proportional increase in thesignal as multiple dNTPs are incorporated.The first NGS instrument developed was the 454pyrosequencing 24 device. This SNA system distributestemplate-bound beads into a PicoTiterPlate along withbeads containing an enzyme cocktail. As a dNTP isincorporated into a strand, an enzymatic cascade occurs,resulting in a bioluminescence signal. Each burst of light,detected by a charge-coupled device (CCD) camera, can beattributed to the incorporation of one or more identicaldNTPs at a particular bead (FIG. 4a).The Ion Torrent was the first NGS platform without optical sensing 25. Rather than using an enzymaticcascade to generate a signal, the Ion Torrent platformdetects the H ions that are released as each dNTP isincorporated. The resulting change in pH is detected byan integrated complementary metal-oxide- semiconductor(CMOS) and an ion-sensitive field-effect transistor (ISFET)(FIG. 4b) . The pH change detected by the sensor isimperfectly proportional to the number of nucleotidesdetected, allowing for limited accuracy in measuringhomopolymer lengths.Comparison of short-read platforms. Individual shortread sequencing platforms vary with respect to throughput, cost, error profile and read structure (TABLE 1).Despite the existence of several NGS technology providers, NGS research is increasingly being conductedwithin the Illumina suite of instruments21. Althoughthis implies high confidence in their data, it also raisesconcerns about systemic biases derived from using asingle sequencing approach26–28. As a consequence, newapproaches are being developed and researchers increasingly have the choice to integrate multiple sequencingmethods with complementary strengths.The SBL technique used by both the SOLiD andComplete Genomics systems affords these technologiesa very high accuracy ( 99.99%)7,14, as each base is probedmultiple times. Although accurate, both platforms alsoshow evidence of a trade-off between sensitivity andspecificity, such that true variants are missed whilefew false variants are called29–31. There is also evidencethat the platforms share some under- representation ofAT‑rich regions26,32, and the SOLiD platform displayssome substitution errors and some GC‑rich under- representation32. Perhaps the feature most limiting to thewidespread adoption of these technologies is the veryshort read lengths. Although both platforms can generate single-end and paired-end sequencing reads, the maximum read length is just 75 bp for SOLiD and 28–100 bpfor Complete Genomics33, limiting their use for genomeassembly and structural variant detection applications.Unfortunately, owing to these limitations, along withruntimes on the order of several days, the SOLiD systemhas been relegated to a small niche within the industry.Furthermore, although the cPAL-based Revolocity system was intended to compete with the Illumina HiSeqin terms of cost and throughout, its launch was suspended in 2016 and it is now only available as a serviceplatform for human WGS33,34, whereas the cPAS-basedBGISEQ‑500 platform is limited to mainland China.Illumina dominates the short-read sequencingindustry owing, in part, to its maturity as a technology, a high level of cross-platform compatibility andits wide range of platforms. The suite of instrumentsavailable ranges from the low-throughput MiniSeq tothe ultra-high-throughput HiSeq X, which is capableof sequen cing 1,800 human genomes to 30 coverage per year. Further diversification is derived from themany options available for runtime, read structure andread length (up to 300 bp). As the Illumina platformrelies on a CRT approach, it is much less susceptible tothe homo poly mer errors observed in SNA platforms.Although it has an overall accuracy rate of 99.5%35,the platform does display some under-representationin AT-rich32,36 and GC‑rich regions32,37, as well as a tendency towards substitution errors38. In 2008, BentleyNATURE REVIEWS GENETICSVOLUME 17 JUNE 2016 02

REVIEWSet al.35 r

limited throughput and the high costs of sequencing remained major barriers. The release of the first truly high-throughput sequencing platform in the mid-2000s heralded a 50,000-fold drop in the cost of human genome sequencing since the Human Genome Project 3 and led to the moniker: next-generation sequencing (NGS).