Sanger_paper Sanger Paper

User Manual: sanger_paper

Open the PDF directly: View PDF .
Page Count: 6

The Wellcome Trust Sanger Institute is one of the

world’s largest genome centers, and a substantial

amount of our sequencing is performed with

‘next-generation’ massively parallel sequencing

technologies: in June 2008 the quantity of purity-

filtered sequence data generated by our Genome

Analyzer (Illumina) platforms reached 1 terabase,

and our average weekly Illumina production output

is currently 64 gigabases. Here we describe a set of

improvements we have made to the standard Illumina

protocols to make the library preparation more reliable

in a high-throughput environment, to reduce bias,

tighten insert size distribution and reliably obtain

high yields of data.

Next-generation DNA sequencers, such as the 454-FLX

(Roche), SOLiD (Applied Biosystems) and Genome

Analyzer (Illumina) have transformed the landscape of

genetics through their ability to produce hundreds of

megabases of sequence information in a single run. This

has enabled us to design genome-wide and ultra-deep

sequencing projects that, because of their enormity, would

not otherwise have been possible (for reviews see refs. 1,

2 and for an evaluation of the performance of these three

platforms see ref. 3).

At the Wellcome Trust Sanger Institute we currently

have all three of these sequencing platforms, though the

Genome Analyzer is the platform we have invested most

heavily in: we have 28 machines on site, all capable of

generating paired-end data. The Illumina data analysis

pipeline performs ‘purity filtering’ of these data to elimi-

nate sequence data from clusters that appear to be mixed

as a consequence of their proximity on the flowcell. We

typically generate 4–5 gigabases (Gb) of filtered sequence

data, with an error rate of <0.9% per seven-day, 36-cycle

paired-end run, making us one of the world’s largest and

most productive users of Illumina sequencers.

Sequencing library preparation involves the produc-

tion of a random collection of adapter-modified DNA

fragments, with a specific range of fragment sizes, which

are ready to be sequenced. We have found the standard

Illumina sequencing library preparation protocols (Fig. 1)

to be suboptimal in several respects, and we enhanced our

output by developing and implementing many modifica-

tions and improvements to these protocols, all with the

aim of obtaining the maximum number of high-quality

sequence reads per run from the lowest mass of starting

DNA, in a robust and reproducible way. The modifica-

tions and improvements we describe here can be adopted

en masse as an alternative library preparation pipeline.

However, because some steps are additional rather than

alternative, we tend to select different modifications for

different sequencing projects, depending on the specific

requirements of that project (Supplementary Table 1

online). Here we have attempted to describe each modifi-

cation in the order in which it would fit in to the standard

library preparation pipeline.

Fragmentation

The first stage in a standard genomic DNA library prepa-

ration for the Genome Analyzer is DNA fragmentation

by nebulization (in 30–60% glycerol at 30–35 p.s.i.). This

generates fragments with a typical size range of 0–1200

base pairs (bp) and a peak around 5–600 bp. Nebulization

is a fairly reproducible technique, is sequence-indepen-

dent, and is rapid and inexpensive4. However, the wide

size distribution of generated fragments is uneconomical:

by mass, the 200 ± 20-bp fragments represent only ~10%

of the total DNA after nebulization. Moreover, approxi-

mately half of the DNA vaporizes during nebulization,

meaning that only 5% of the original DNA is used for

subsequent library generation. Even under much more

extreme nebulization conditions (for example, 90 p.s.i.

for 18 min) it is not possible to ‘move’ the fragment-

size peak below around 400 bp, and doing so still does

not improve the yield at 200 bp (ref. 4 and unpublished

observations). Thus we have evaluated alternative meth-

ods of sample fragmentation.

A large genome center’s improvements to

the Illumina sequencing system

Michael A Quail, Iwanka Kozarewa, Frances Smith, Aylwyn Scally, Philip J Stephens,

Richard Durbin, Harold Swerdlow & Daniel J Turner

Wellcome Trust Sanger Institute, Wellcome Trust Genome Campus, Hinxton, Cambridgeshire, CB10 1SA, UK. Correspondence should be

addressed to D.J.T. (djt@sanger.ac.uk).

PUBLISHED ONLINE 25 NOVEMBER 2008; DOI:10.1038/NMETH.1270

NATURE METHODS | VOL.5 NO.12 | DECEMBER 2008 | 1005

PERSPECTIVE

tocol: (i) bias in the base composition of

sequences; (ii) high frequency of chimeric

sequences, produced when two template

strands are ligated during the adapter liga-

tion step; and (iii) imperfect distribution of

insert sizes. These have all been overcome by

the use of several protocol modifications,

described here.

Paired-end oligonucleotides. We no longer

use the Illumina single-end adapters or PCR

primers because paired-end oligos generate

sequencing libraries that are compatible with

both single and paired-end flowcells. The

adapters themselves are modified to confer

protection from digestion at the 3′ thymine

(T) overhang. Though we do not have the

details of the modification used by Illumina,

we have obtained comparable results using

our own adapters and PCR primers modi-

fied with a phosphorothioate between the

two bases at the 3′ end (Supplementary

Protocol 2 online).

Gel extraction. During the size selection step

of the standard library prep protocol, a gel

slice is selected and the DNA extracted. We

identified that melting this gel slice by heating

to 50 °C in chaotropic buffer decreased the

representation of A+T-rich sequences, pos-

sibly reflecting a higher affinity of spin col-

umns for double-stranded DNA, as strands

with a high A+T content will be most likely to

become denatured during this step and may not reanneal. To improve

the representation of these A+T-rich sequences, we modified the gel-

extraction protocol, melting agarose gel slices in the supplied buffer

at room temperature (18–22 °C). This reduces G+C bias considerably

(Supplementary Protocol 3 online and Fig. 3a,b).

Double size selection. Partially complementary adapters, which

essentially consist of the sequences to which the sequence primers

hybridize during the sequencing reaction, are ligated onto the A-tailed

fragments7 via a T overhang (Fig. 1). Their structure ensures that each

template strand receives different sequences at the opposite ends8 and

works in much the same way as a vectorette9.

Inefficient end-repair or A-tailing reactions will result in a lower

concentration of template to which adapters can be successfully ligat-

ed, and so the relative concentration of adapters is increased, which

will promote the formation of adapter dimers. If these dimers are not

removed, they will ultimately be sequenced along with the intended

template, wasting the capacity of the flowcell. Additionally, inefficient

A-tailing will result in a high proportion of blunt-ended template

molecules, which can self-ligate, generating chimeric sequences.

It is likely that the efficiency of the A-tailing step will be improved

by the use of alternative polymerases and higher concentrations of

magnesium ions. Additionally, the efficiency of the ligation step

appears to be improved by the use of ultrapure ligases, such as

those from Enzymatics, which are virtually free of the contaminat-

ing exonuclease activity generally found in standard commercial

Adaptive focused acoustics. We now routinely fragment DNA sam-

ples using adaptive focused acoustics technology in a 24-well format

(AFA; Covaris). In this process, acoustic energy is controllably focused

into the aqueous DNA sample by a dish-shaped transducer, resulting

in cavitation events within the sample. The collapse of bubbles in the

suspension creates multiple, intense, localized jets of water, which dis-

rupt the DNA molecules in a reproducible and predictable way.

After disruption, 200-bp fragments comprise 17% of the total

fractionated DNA by mass, but in contrast to nebulization, very

little DNA is lost during the fragmentation process, generating a

four- to fivefold higher yield of the intended fragment size range

(Supplementary Protocol 1 online and Fig. 2). Additionally, because

of its high-throughput capability, AFA has enabled highly multiplexed

sequencing using indexing tags, where each sample needs to be pro-

cessed separately until after PCR amplification (Fig. 1). Also, because

the size distribution of DNA fragmented by AFA is narrow, for some

applications, such as array enrichment of targeted loci5,6, we are able

to omit the gel electrophoresis–based size selection step altogether

from the library preparation, decreasing the workload and increasing

yields further.

A-tailing, ligation and size selection

After close scrutiny of paired-end reads obtained from the Genome

Analyzer, that is, those in which each cluster was sequenced in both

forward and reverse directions, we discovered a number of artifacts

that could be attributed to the standard library preparation pro-

Genomic DNA PCR-amplified

library

Fragmented DNA

sample (0–1200 bp)

Blunt-ended, 5'-

phosphorylated fragments

3' A-tailed

fragments

Adapter-ligated

fragments

Size-selected

fragment library

PCR-amplified

library

Fragmentation

End repair

A-tailing

Ligation

Size selection

PCR

3,4

4,5

6,7,8,9,10

Single-stranded

DNA library

Clonal DNA clusters

on flowcell surface

Single-stranded clusters

ready for sequencing

Incorporation of fluorescent

reversible terminator nucleotides

Cleavage of fluorophores

and blocking groups

Denaturation

Linearization, blocking and hybridization

Flowcell transfer to Genome Analyzer

Imaging

Cycles of cluster amplification

Quantification

Library preparation

Cluster growth

Sequencing by synthesis

Figure 1 | Illumina sequencing workflow. Stages in the library preparation. Steps accompanied by

numbers are those for which we suggest alternatives to the standard Illumina protocols. Numbers

correspond to those given in Supplementary Protocols 1–13 online.

1006 | VOL.5 NO.12 | DECEMBER 2008 | NATURE METHODS

PERSPECTIVE

Paired-end size selection. Using standard protocols, we found single-

end library preparations, using single- or paired-end adapters, to be

considerably more robust at the step of size selection by electropho-

resis than their paired-end counterparts. With single-end preps, we

excise a band of 50 bp or larger, which generally yields more than

enough DNA to give a high yield of PCR products. However, for the

paired-end protocol, to generate as narrow an insert size range as

possible, a scalpel is inserted into the gel at the desired position, the

blade is washed with Tris buffer, and this buffer acts as the template

for the PCR amplification. We found this practice to yield enough

DNA to give successful amplification only in approximately 30–40%

of attempts. To overcome this, we now excise a 2-mm-wide gel slice

containing DNA of the desired size and extract that following the

Illumina protocol, though with no heating during the melting step,

as discussed above (Supplementary Protocol 3). In our hands this

typically yields 10–20 times more DNA than the standard protocol,

has an almost 100% success rate and generates an acceptably narrow

size distribution of paired-end reads (Fig. 3f,g).

PCR

As well as increasing robustness, extracting more DNA from gel slices

enables the DNA to be quantified more accurately before PCR ampli-

fication. The PCR step introduces into the adapter-ligated template

molecules the oligonucleotide sequences required for hybridization

to the flowcell surface.

preparations of ligases. Using this enzyme we have achieved a

20–30% increase in yield of successfully ligated fragments (as deter-

mined by quantitative PCR (qPCR); Supplementary Protocol 4

online), presumably because the reduced exonuclease activity

regenerates fewer blunt-ended fragments after A-tailing.

However, although the steps described

above may reduce the formation of blunt-

ended ligation concatamers, enzymatic

reactions are rarely 100% efficient, and thus

a small proportion of template strands will

still be chimeric. In many applications, a low

frequency of chimeric sequences will pres-

ent no problem and can simply be removed

informatically. In other applications, such as

detection of infrequent de novo recombinant

molecules, chimeric sequences will generate

false positives, so their frequency needs to be

reduced to minimize the amount of subse-

quent confirmatory work.

Chimeric templates will be longer than the

singletons, which provides a way of prevent-

ing them from contaminating the DNA frac-

tion: for applications in which a low frequency

of chimeric templates is required, we perform

an additional size selection after shearing but

before ligation (Supplementary Protocol 5

online). This results in a narrow size range

being available for ligation. Any blunt-ended

concatemers are appreciably longer than the

singletons, and we remove these in the post-

ligation size-selection step. This additional

size selection reduces the incidence of chime-

ras to 0.02%, compared to up to 5% with the

standard library preparation protocol, and we

have found this step to have the added benefit

of reducing the shoulder of small insert sizes,

giving a tighter insert-size distribution of the

desired fraction (Fig. 3c–e), which leads to

clusters with more uniform diameter.

140

120

100

–20

Relative fluorescence units

15 100 150 300 400 500 700 1,500

Size (bp)

Nebulization

AFA

Ladder

140

120

100

–20

Relative fluorescence units

15 100150 300 400 500 700 1,500

Size (bp)

Mapped depth (bin size, 500 bp)

0 20 40 60 80 100

Percentile of unique sequence

ordered by G+C content

10 30 40 50 60

G+C content (%)

Distribution of reads with

indicated G+C content

Mean read depth

S.d.

Mapped depth (bin size, 500 bp)

0 20 40 60 80 100

Percentile of unique sequence

ordered by G+C content

10 30 40 50 60

G+C content (%)

Distribution of reads with

indicated G+C content

Mean read depth

S.d.

140

120

100

–20

Relative fluorescence units

15 100150 300 400 500 700 1,500

Size (bp)

140

120

100

–20

Relative fluorescence units

15 100150 300 400 500 700 1,500

Size (bp)

3.5

3.0

2.5

2.0

1.5

1.0

0.5

0.0

Number of reads (millions)

0 100 200 300 400 500 600

Size (bp)

3.5

3.0

2.5

2.0

1.5

1.0

0.5

0.0

Number of reads (millions)

0 100 200 300 400 500 600

Size (bp)

cde

Figure 2 | Sample fragmentation. Comparison of fragmentation by

nebulization with AFA technology. We fragmented 4.5 µg of human genomic

DNA by nebulization or AFA, purified the samples using a spin column,

eluted the DNA in 30 µl of 10 mM Tris pH 8.5, and ran 1 µl of each eluate on

an Agilent Bioanalyzer 2100 DNA 100 chip. For a 200-bp (±20 bp) library,

the yield produced by AFA was four- to fivefold greater than that produced

by nebulization.

Figure 3 | A-tailing, ligation and size selection. (a,b) G+C plots before (a) and after (b) optimization

of gel extraction. The figures show the total area in which reads with a particular G+C content are

distributed, with the mean and s.d. The greater width of the shaded area in plot a indicates a wider

dispersion of coverage for all values of G+C content for which sequences were obtained. (c–e) Agilent

Bioanalyzer 2100 traces for three libraries, a 60-bp insert library with optimized PCR (c), the same

60-bp library with excess DNA in PCR (d) and a 200bp insert library, showing shoulder of small

fragments (e). (f,g) Insert size distribution from sequenced human DNA using the standard (f) and

modified (g) paired-end library preparation protocols.

NATURE METHODS | VOL.5 NO.12 | DECEMBER 2008 | 1007

PERSPECTIVE

PCR yield. By the use of alternative high-fidelity polymerases in

a more optimized reaction, we have found it possible to increase

the yield of the enrichment PCR five- to tenfold (Supplementary

Protocol 7 online and Fig. 4a), which allows fewer cycles of amplifi-

cation to be performed.

PCR cleanup. Surplus PCR primers may interfere with quantification

and will compete with the amplicon for hybridization to the flowcell

surface. Consequently, it is necessary to remove surplus oligos after

amplification. We have found that solid-phase reversible immobiliza-

tion (SPRI) technology10 can be used to remove a higher proportion

of primers and adapter dimers than spin columns, while producing

a comparable yield of amplicon DNA, and allows elution in a wider

variety of buffers (Supplementary Protocol 8 online and Fig. 4b).

Sequencing without PCR. We have found that it is unnecessary to

retain the PCR step to enrich for properly ligated fragments, so long

as only those fragments with an adapter at either end can be quanti-

fied, as only they will yield clusters that can be sequenced. This can

be done by quantitative PCR, discussed below. Thus we can elimi-

nate the PCR step entirely, simply by ligating on appropriate adapters

after A-tailing (Supplementary Protocol 9 online). For this purpose

we use high-performance liquid chromatography–purified, partially

noncomplementary oligos with a phosphorothioate linkage between

the two bases at the 3′ end of one strand. From a starting amount

of 5 µg of DNA and fractionation by AFA, we can obtain sufficient

paired-end DNA for > 400 lanes of high-density clusters, or 100 lanes

if nebulization is used to fragment the DNA. The obvious benefits

of this are that PCR duplicates are absent: the observed duplication

Template quantity. By using optimized quantities of template in the

PCR, we can ensure a clean library, free of adapter-dimer or single-

stranded DNA. We routinely analyze our sequencing libraries after

PCR by microfluidic capillary electrophoresis and have noticed that

the quality of the library obtained decreases with increasing concen-

tration of template DNA: too much template DNA often results in

the accumulation of an apparently higher-molecular-weight peak

(Fig. 3d), which may represent a single-stranded template product

that accumulates as primers become depleted. Conversely, the lower

the mass of DNA used in the PCR, the fewer the number of template

strands for the same fragment size, and the greater the incidence of

PCR product duplicates in the resulting sequences: we have observed

libraries from which as many as 60% of sequences were PCR product

duplicates. Thus it is essential to choose the appropriate set of condi-

tions for each PCR (Supplementary Protocol 6 online).

Adapter dimers

PCR primers

1,400

1,000

900

800

700

600

500

400

300

200

Column

cleanup Ladder

Band

size (bp) Ladder

SPRI

cleanup

280

240

200

160

120

15 100 150 300 400 500 700 1,500

Size (bp)

Optimal PCR conditions

Standard PCR conditions

Figure 4 | PCR. (a) An ~200-bp fragment library was prepared, and 10 ng

was amplified for 18 cycles using standard Illumina PCR conditions or

optimized PCR conditions. (b) A comparison of methods of PCR amplicon

purification. We prepared a paired-end library with phiX DNA using conditions

that would promote the formation of adapter dimers and unextended PCR

primers. After PCR, we divided the library into two: half was purified following

the standard Illumina protocol, through a QiaQuick PCR cleanup column (left),

whereas the other was purified using SPRI technology (right).

35,000

30,000

25,000

20,000

15,000

10,000

5,000

Number of clusers after purity filtering

0 10,000 20,000 30,000 40,000 50,000 60,000 70,000

Total number of clusters detected

500-bp fragments

200-bp fragments

70,000

50,000

30,000

10,000

Unfiltered cluster number per tile

75 150 250 350 450 550 650 750 850

Run number

Median cluster number

Upper and lower quartiles

1.5 interquartile range

Outlier

Fluctuation in median cluste

number from bin to bin

qPCR assay introduced

Figure 5 | Quantification. (a) Cluster throughput as a function of total clusters for 200- and 500-bp inserts. The 500-bp inserts underwent fewer cycles of

cluster amplification (28, compared to 35 for the 200-bp libraries), resulting in smaller clusters, and so a cluster density of 40–44k per tile (GA1) will produce

the maximum yield from either insert size. (b) Standardization of cluster density with qPCR quantification. Runs were grouped into 25-run bins, and a boxplot

was generated. After some initial problems with degradation of standards, cluster number has leveled out at ~35,000–40,000 per tile.

1008 | VOL.5 NO.12 | DECEMBER 2008 | NATURE METHODS

PERSPECTIVE

library to 10 pM, based on the concentration as measured using

an Agilent Bioanalyzer 2100. With this assay, we take the Agilent-

derived concentration values to be arbitrary, allowing us to dilute

unknowns to a concentration that lies within the 1–100 pM range,

but Agilent values also provide a useful double-check. We have found

that cluster density can be predicted reliably in this way (Fig. 5b).

We have found that the ability to quantify DNA in the picomolar

concentration range also opens up the opportunity for sequencing

much lower DNA concentrations than those permitted by the stan-

dard protocol, such as unamplified array eluates from a sequence-

capture experiment5,6.

Denaturation

Being single-stranded, array eluates require no particular steps to

denature the DNA12 before sequencing. However, for low concen-

trations (<1 nM) of double-stranded DNA it is more problematic:

denaturation by heating has the potential both to damage the DNA

and to introduce anti-(G+C) bias13.

Modified hybridization buffers. For all denaturation we pre-

fer the use of 0.1 M NaOH to heating, though for subnanomolar

libraries this requires an alternative hybridization buffer to be used

(Supplementary Protocol 12 online). We have found the addi-

rate in mapped gorilla DNA sequences prepared without PCR was

approximately 0.5%. This rate of duplication is caused by noise in the

cluster detection and sequence analysis software.

Direct sequencing of short amplicons. To avoid unnecessary PCR

amplification steps, which would potentially exacerbate biases, we can

perform extremely deep sequencing of short amplicons using locus-

specific primers that possess tails that can hybridize to the oligos teth-

ered to the flowcell surface. The tailless forward and reverse oligos

are then used as primers in the sequencing steps (Supplementary

Protocol 10 online).

Quantification

At close to neutral pH, the concentration of DNA going onto the

flowcell governs the number of clusters produced. Thus, for differ-

ent fragment sizes undergoing a given number of cycles of cluster

amplification, there is an optimal concentration range of DNA that

will yield clusters in the optimal density range, enabling the maxi-

mum amount of data to be obtained. For fragments with a mean

insert size of 500 bp or lower, we aim for 40,000–44,000 clusters per

imaged area (tile) on the Genome Analyzer model 1, giving an aver-

age of 20,000–25,000 filtered clusters per tile, equating to 2.0–2.4

Gb per single end run (150–170,000 clusters per tile for the Genome

Analyzer model 2; Fig. 5a).

Overestimation of DNA concentration results in too few clus-

ters, which may make the flowcell uneconomical to sequence.

Underestimation results in too high a cluster density, which can

greatly reduce the amount of data obtained, owing to cluster overlap.

Quantification of DNA before sequencing is thus one of the key fac-

tors in the process.

Electrophoresis. We found the accuracy of spectrophotometry to be

inadequate for quantification: cluster density based on this method

tended to be inconsistent, but typically five- to tenfold lower than

anticipated, presumably because spectrophotometry analysis mea-

sures not only the intended amplicon but also adapter dimers and

unextended primers, with no way of distinguishing between them,

and also struggles to measure low DNA concentrations accurately. By

quantifying libraries electrophoretically, with an Agilent Bioanalyzer,

we have been able to achieve a much more consistent cluster den-

sity. Additionally, because electrophoresis can be used to distinguish

between DNA species on the basis of size, it provides a way to check

the quality of the library preparation. However, for a small proportion

of libraries, we obtained far higher cluster densities, and consequently

far less useful data, than anticipated. We assume that this is a result

of single-stranded DNA generated in the PCR that cannot be easily

quantified when mixed with double-stranded DNA.

Quantitative PCR. This led us to develop a qPCR quantification

assay (for discussion see ref. 11) because such an approach should

be capable of detecting and quantifying all amplifiable molecules.

We designed amplification primers and a dual-labeled probe to

target the Illumina paired-end adapter sequences (Supplementary

Protocol 11 online). We quantify unknown libraries against stan-

dard libraries that have been sequenced previously, and for which

we know the accurate cluster number, and how this relates to the

Agilent concentration of that library. Because amplification in the

qPCR with these primers is not perfectly efficient, we use 3 dilutions

of standard libraries (100, 10 and 1 pM), and dilute the unknown

0 5 10 15 20

Volume 0.1 M NaOH added (µl)

30,000

20,000

10,000

Cluster/tile

0 5 10 15

Volume denatured template added (µl)

Water

5× SSC

5× SSC + 2 mM Tr is

5× SSC + 10 mM Tr is

5× SSC + 50 mM Tr is

5× SSC

5× SSC + 5 mM Tr is

Figure 6 | Denaturation. (a) pH titration of hybridization buffers. Following

denaturation, the concentration of NaOH in DNA templates is 0.1 M

NaOH. Adding more than 8 µl of this denatured template to the 1 ml of

Hybridization Buffer (5× SSC, 0.1% Tween-20), before loading DNA onto the

flowcell, increases the pH to above 10. This prevents efficient hybridization,

and thus the cluster density falls. The addition of Tris-HCl pH 7.3 to the

supplied bottles of Hybridization Buffer dramatically increases buffering

capacity, making template hybridization more robust. (b) The addition

of 5 mM Tris-HCl pH 7.3 to Illumina Hybridization Buffer allows a greater

volume of denatured template to be added before high pH prevents effective

annealing of templates to the oligos on the flowcell surface. This increases

the robustness of cluster generation by counteracting pipetting errors in the

denaturation step.

NATURE METHODS | VOL.5 NO.12 | DECEMBER 2008 | 1009

PERSPECTIVE

software and a faster run time, both of which we are currently testing,

a conservative prediction is that by the end of 2008, our output will

reach 6–10 terabases of high-quality sequence per year, equivalent to

180 human genomes at 15-fold coverage, or approximately 200,000

bases per second.

The improved workflow and high yield should maintain the

Genome Analyzer as our next-generation sequencing platform of

choice for the immediate future. How long this remains true depends

upon the performance of existing rival technologies: Roche’s 454,

ABI’s SOLiD, Helicos’ ‘True Single Molecule Sequencing’ and Dover

Systems’ Polonator, and those that are on the horizon, such as nano-

pore technologies, for example Oxford Nanopore Technologies, the

Harvard Nanopore Group, and Pacific Biosciences’ Single Molecule

Real Time technology, which promise to bring us closer to the eagerly

anticipated $1,000 genome.

Note: Supplementary information is available on the Nature Methods website.

ACKNOWLEDGMENTS

We thank all the staff at Illumina for their support, particularly T. Ost, M. Gibbs, J.

Smith, N. Gormley, V. Smith and K. Hall. We also thank C. Brown, A. Brown,

R. Pettett, T. Skelly, N. Whiteford, L. Mamanova, E. Sheridan and E. Huckle for helpful

discussions and assistance.

COMPETING INTERESTS STATEMENT

The authors declare competing financial interests: details accompany the full-text

HTML version of the paper at http://www.nature.com/naturemethods/.

Published online at http://www.nature.com/naturemethods/

Reprints and permissions information is available online at http://npg.

nature.com/reprintsandpermissions/

1. Bentley, D.R. Whole-genome re-sequencing. Curr. Opin. Genet. Dev. 16, 545–552

(2006).

2. Mardis, E.R. The impact of next-generation sequencing technology on genetics.

Trends Genet. 24, 133–141 (2008).

3. Smith, D.R. et al. Rapid whole-genome mutational profiling using next-

generation sequencing technologies. Genome Res. 18, 1638–1642 (2008).

4. Surzycki, S. DNA sequencing. in Basic Techniques in Molecular Biology 377–380

(Springer-Verlag, Berlin, 2000).

5. Albert, T.J. et al. Direct selection of human genomic loci by microarray

hybridization. Nat. Methods 4, 903–905 (2007).

6. Hodges, E. et al. Genome-wide in situ exon capture for selective resequencing.

Nat. Genet. 39, 1522–1527 (2007).

7. Sambrook, J., Fritsch, E. & Maniatis, T. Molecular Cloning: A Laboratory Manual,

(Cold Spring Harbor Laboratory Press, Cold Spring Harbor, New York, 1989).

8. Smith, D. & Malek, J. Asymmetrical adapters and uses thereof. US patent

0172839 (2007).

9. Riley, J. et al. A novel, rapid method for the isolation of terminal sequences

from yeast artificial chromosome (YAC) clones. Nucleic Acids Res. 18, 2887–2890

(1990).

10. Hawkins, T.L., O’Connor-Morin, T., Roy, A. & Santillan, C. DNA purification and

isolation using a solid-phase. Nucleic Acids Res. 22, 4543–4544 (1994).

11. Meyer, M. et al. From micrograms to picograms: quantitative PCR reduces the

material demands of high-throughput sequencing. Nucleic Acids Res. 36, e5

(2008).

12. Thomas, R. The denaturation of DNA. Gene 135, 77–79 (1993).

13. Mandel, M. & Marmur, J. Use of ultraviolet absorbance-temperature profile for

determining the guanine plus cytosine content of DNA. Methods Enzymol. 12,

195–206 (1968).

tion of Tris to the standard Illumina Hybridization Buffer to be

beneficial to the robustness of the initial hybridization of DNA

to the flowcell for all libraries, because it can counteract pipetting

errors during the denaturation stage that would otherwise raise

the pH to a level that would prevent efficient hybridization (Fig. 6).

Additionally, diluting the supplied 2 M NaOH and adding a greater

volume to the 20 µl denaturation reaction helps to reduce fluctua-

tion in cluster number owing to pipetting errors (Supplementary

Protocol 12).

Amplification quality control

After cluster amplification, DNA on the flowcell is double-stranded,

which allows clusters to be stained by an intercalating dye and to be

detected using a fluorescence microscope (Supplementary Protocol

13 online). This is a valuable quality control step, that we use for all

flowcells before linearization and blocking to confirm that the clus-

ter density is appropriate. We generally do not sequence data from

flowcells that have too high or too low a cluster density beyond the

amplification stage.

Conclusion

The Genome Analyzer is a powerful sequencing technology, yet still

relatively new, and consequently it has not yet reached its full sequenc-

ing potential. Here we have described modifications that allow for

more efficient library preparation and enable a stable workflow in a

production environment.

At the Sanger Institute, in addition to a sequencing research and

development team we have several teams who are responsible for

keeping the production instruments running. A library-making

group processes samples, and generates, quality-controls and quan-

tifies libraries. A production group, working in shifts, prepares and

quality-controls flowcells by SybrGreen staining, prepares reagents for

sequencing and manages washing, priming and loading the instru-

ments seven days per week. Informatics teams are responsible for

facilitating sample tracking, for handling the sequence data and for

performing pipeline analyses. All steps in the process are recorded

using custom-written lab-tracking and run-tracking database soft-

ware. All Genome Analyzers are networked, and the generated image

data are continually uploaded to a large compute and disk-storage

cluster for image and base-calling analysis, alignment and assembly,

and other informatics tasks. We keep images for about 1 month on a

disk server, but we store the run quality control and other run details

in a database and deposit short-read sequences for permanent stor-

age in a large repository. A team of project managers coordinate and

oversee individual sequencing projects.

We have recently upgraded all of our Genome Analyzers to the

model 2. The wider flowcells used by upgraded machines offer a 40%

greater imaging area, with the potential for increased read lengths

(>70 bases) of a higher quality (below 1% error in a phiX control lane

for 1–50 bases). Combined with improvements to the image analysis

1010 | VOL.5 NO.12 | DECEMBER 2008 | NATURE METHODS

PERSPECTIVE

Sanger_paper Sanger Paper

Navigation menu

Versions of this User Manual:

Views

Navigation