UBC Faculty Research and Publications

Sequencing platform and library preparation choices impact viral metagenomes Solonenko, Sergei A; Ignacio-Espinoza, J C; Alberti, Adriana; Cruaud, Corinne; Hallam, Steven; Konstantinidis, Kostas; Tyson, Gene; Wincker, Patrick; Sullivan, Matthew B May 10, 2013

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


52383-12864_2013_Article_5001.pdf [ 361.12kB ]
JSON: 52383-1.0220626.json
JSON-LD: 52383-1.0220626-ld.json
RDF/XML (Pretty): 52383-1.0220626-rdf.xml
RDF/JSON: 52383-1.0220626-rdf.json
Turtle: 52383-1.0220626-turtle.txt
N-Triples: 52383-1.0220626-rdf-ntriples.txt
Original Record: 52383-1.0220626-source.json
Full Text

Full Text

RESEARCH ARTICLE Open AccessSequencing platform and library preparationchoices impact viral metagenomesSergei A Solonenko1, J César Ignacio-Espinoza2, Adriana Alberti3, Corinne Cruaud3, Steven Hallam4,Kostas Konstantinidis5, Gene Tyson6, Patrick Wincker3 and Matthew B Sullivan1,2*AbstractBackground: Microbes drive the biogeochemistry that fuels the planet. Microbial viruses modulate their hostsdirectly through mortality and horizontal gene transfer, and indirectly by re-programming host metabolisms duringinfection. However, our ability to study these virus-host interactions is limited by methods that are low-throughputand heavily reliant upon the subset of organisms that are in culture. One way forward are culture-independentmetagenomic approaches, but these novel methods are rarely rigorously tested, especially for studies ofenvironmental viruses, air microbiomes, extreme environment microbiology and other areas with constrainedsample amounts. Here we perform replicated experiments to evaluate Roche 454, Illumina HiSeq, and Ion TorrentPGM sequencing and library preparation protocols on virus metagenomes generated from as little as 10pg of DNA.Results: Using %G + C content to compare metagenomes, we find that (i) metagenomes are highly replicable,(ii) some treatment effects are minimal, e.g., sequencing technology choice has 6-fold less impact than varyinginput DNA amount, and (iii) when restricted to a limited DNA concentration (<1μg), changing the amount ofamplification produces little variation. These trends were also observed when examining the metagenomes forgene function and assembly performance, although the latter more closely aligned to sequencing effort and readlength than preparation steps tested. Among Illumina library preparation options, transposon-based librariesdiverged from all others and adaptor ligation was a critical step for optimizing sequencing yields.Conclusions: These data guide researchers in generating systematic, comparative datasets to understand complexecosystems, and suggest that neither varied amplification nor sequencing platforms will deter such efforts.BackgroundAdvances in sequencing technologies have revolutionizedthe life sciences. For example, ecology and evolution cannow be examined across the tree of life [1], and at reso-lutions ranging from broad analyses (e.g., BGI’s 10,000Microbial Genomes Project, http://ldl.genomics.cn/page/M-research.jsp) to focused investigation of populationstructure within particular species [2]. These analyses, how-ever, center on genomes as the unit of interest and repre-sent a “bottom-up approach” to exploring the diversity oflife [3].Concurrently, metagenomics provides a “top-downapproach” for studying complex microbial assemblagesin nature [3]. Recent reviews cover next generation se-quencing applications [4-6], but rarely acknowledge thefactors that generate quantitative data needed formetagenomics. For example, sequence quality evaluatedacross benchtop systems did not consider library prepar-ation [7], and recommendations of amplification-free pro-tocols that require >2 μg of DNA to minimize biases [8]are not meaningful for DNA-limited applications. Thereare also numerous sequencing platform options, thoughmicrobial metagenomes generated across commonly-usedsequencing platforms only minimally differ in taxonomicdistributions or contig assembly quality [9].Some fields, such as viral ecology or microbial ecologyof permafrost soils or the atmosphere, are routinely DNA-limited (<1 ng) and thus require optimization and quanti-tation assessment at each step in the metagenomicsample-to-sequence pipeline [10]. Towards this end, em-pirical data are now available to guide researchers in* Correspondence: mbsulli@email.arizona.edu1Department of Ecology and Evolutionary Biology, University of Arizona,Tucson, AZ, USA2Department of Molecular and Cellular Biology, University of Arizona, Tucson,AZ, USAFull list of author information is available at the end of the article© 2013 Solonenko et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of theCreative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use,distribution, and reproduction in any medium, provided the original work is properly cited.Solonenko et al. BMC Genomics 2013, 14:320http://www.biomedcentral.com/1471-2164/14/320concentrating and purifying viruses [11,12] prior to DNAextraction. Once DNA is extracted, small yields requireamplification to obtain enough material for sequencing.While whole genome amplification was an attractive op-tion, it is now documented to result in non-quantitativemetagenomes due to both stochastic [13] and systematicbiases [14]. In contrast, linker-amplification-based libraries[15-17] provide a nearly quantitative alternative, evenfrom sub-nanogram amounts of DNA [15]. Togetherthese advances allowed the compilation of the first large-scale, systematically prepared comparative metagenomicdataset for quantitative viral ecology [18] with new toolsand analytical platforms now emerging to handle suchdatasets [19,20]. Beyond viral ecology, these studiesprovide a roadmap for generating quantitative meta-genomic datasets from any low (<100 ng) input DNAsamples.Here we expand upon these efforts to focus on the finalsteps in viral metagenomic sequencing (overview inFigure 1, and sequencing statistics summarized in Table 1).The first experiment evaluates co-varied input DNA andamplification cycle amounts, as well as sequencing plat-form choice on the resulting metagenomes. These datawere derived from DNA extracted from a 1,080L Bio-sphere 2 Ocean viral concentrate and included small-insert metagenomes prepared from varied low-input DNAamounts (10 pg—100 ng) and amplification conditionsfor commonly used sequencing platforms (IlluminaHiSeq2000, herein ‘Illumina’ and Roche 454 Titanium,herein ‘454’). Additionally, these low-input samples werecomplemented by standard input DNA(≥1,000ng) small-insert metagenomes to compare three sequencing plat-forms (Illumina, 454, Ion Torrent) and limited large-insertclone library Sanger end-sequencing (8,000ng fosmid li-brary). The second experiment focuses on Illumina se-quencing only. Here, viral DNA derived from two separateocean samples (Tara Oceans [21] stations 41 and 109) wasused to examine the effect of amplification conditions(e.g., cycle number) and input DNA amount indepen-dently, as well as compare standard Illumina libraries totransposon-based Nextera libraries [22].ResultsExperiment 1: The impact of input DNA, amplification,and sequencing platform on metagenomesLibrary success varies by sequencing protocolAs expected, the fosmid library and all 6 libraries madefrom ≥1,000 ng DNA were successful in generating suffi-cient DNA for sequencing regardless of sequencing plat-form (Table 1). Additionally, low DNA input libraries for454 (linker-amplified [15] to obtain sufficient genetic ma-terial) were all successful, with highest read yields per ngof input DNA of any method (Additional file 1: Figure S1).In contrast, Experiment 1 Illumina libraries constructedfrom low starting DNA amounts were less successful(Table 1). Specifically, 3 of 6 libraries, one 10ng and both1ng libraries, failed library construction, even with theaddition of carrier DNA and adaptor concentration ad-justment to increase ligation efficiencies. Two of theremaining 3 low input DNA libraries, one 10ng and two100ng, were sequenced, but yielded fewer and more vari-able numbers of reads and abundant adaptor sequence(see * in Table 1).%G + C content variation within treatments is minimalThe replicates’ read %G+C distributions were correlatedusing the Pearson product–moment correlation coeffi-cient (Pearson’s r). There is little variation in %G+Cacross replicate libraries from any 454, Illumina, or IonTorrent sequencing data – replicates have pairwise correl-ation values from 0.99 to 1 and cluster together >94% ofthe time (Figure 2). This indicates that, at least for therange of %G +C in this B2O sample, intra-replicate vari-ation is minimal and therefore there is high power to de-tect statistically significant differences across treatments.05101520250.001 0.01 0.10 1 10 100 1,000Starting DNA amount (ng)Library amplification cycles3322 222 1Roche 454 GS FLX TiIllumina HiSeq2000Ion Torrent PGM Sanger ABI 3730xl210,000Sequencing technology0510152025Library amplification cycles1 10 100 1,000Starting DNA amount (ng)21111 10 100 1,000211 1Station 109 Station 41StandardNexteraLibrary Prep.Experiment 1 Experiment 2Figure 1 Experimental design overview. Library preparation treatments were done at varying levels of replication, as indicated by the numbers(1 to 3) next to each treatment. The number of amplification cycles (see y axis) includes those necessary to generate enough DNA for librarypreparation, but does not include the emPCR (454, Ion Torrent) or bridge (Illumina) amplification cycles used to build large enough populationsof reads for nucleotide sequencing signal detection.Solonenko et al. BMC Genomics 2013, 14:320 Page 2 of 12http://www.biomedcentral.com/1471-2164/14/320Input DNA amount, decision to amplify impact %G+ C contentHierarchical clustering of sample %G+C distribution cor-relations shows consistent differences. First, all ≥1,000 ngmetagenomes cluster together 100% of the time (Figure 2).Of the treatments tested, input DNA most strongly im-pacts the resulting metagenomes, with ≥1,000 ng next-generation metagenomes clearly separated from the rest.Among these ≥1,000 ng samples, Illumina metagenomeshave higher %G +C than 454 and Ion Torrent metage-nomes (see Additional file 1: Figure S2 for example %G +Cdistribution plots), but differences between sequencingplatforms are much less than differences between DNA in-puts, with UPGMA branch length distances of 0.02 and0.16, respectively (Figure 2). While of limited sampling, thelargest shift towards higher %G +C sequences (Pearson’s r<0.8) was in the fosmid library relative to the unamplifiedlibraries (Figure 2, Additional file 1: Figure S3).Among the <1,000ng metagenomes, there are minimaldifferences between platforms and the only supported re-lationship, with bootstraps greater than the intra-replicate94% value, was the clustering of Illumina 100ng sampleswith Illumina 10 ng samples (Figure 2). This implies that,among amplified metagenomes, the degree of amplifica-tion and sequencing platform choice only minimally im-pact the resulting metagenomes. The fact that thesediversely prepared metagenomes were nearly indistin-guishable by %G+C distribution metrics (Pearson’sr values >0.99, Figure 2) is promising for comparability ofamplified metagenomes across sequencing platforms.Duplicate reads uncorrelated with any single variableDuplicates in metagenomes are derived from either natur-ally occurring duplicates in genomes and communities, orartificial duplicates generated during 454’s emPCR step orat some unknown point in Illumina preparations that isinconsistent across replicate libraries [23,24].Here, hierarchical clustering of duplicate frequencies(Figure 2) and raw duplicate distributions, normalized tometagenome size (Additional file 1: Figure S3), suggest apattern of three duplication groups. The first, composedof unamplified 454 and 10ng Illumina metagenomes, con-tains intermediate levels of duplication (14.6 to 42.7%)and few high-frequency (>10 fold) duplicate reads (0.06 to5.1%). The second cluster, composed of most Illuminametagenomes, has an intermediate level of duplication(27.1 to 37.3%), but also an excess of highly duplicatedreads (10.4 to 15.6%). The third includes the amplified 454metagenomes, both Ion Torrent metagenomes, and thepoorly amplified 100ng Illumina library, all of which havefew duplicate reads (0.9 to 16.6%) and very few high-Table 1 Summary statistics for all metagenomic libraries used in analysisDNA source Technology StartingDNA (ng)Libraryamplification(# cycles)Replicates Raw reads(millions)Raw quality+/-SD (PHRED)Rawlength(bp)Failed QC+/- SD (%)Experiment1Biosphere 2OceanIllumina HiSeq20001,000 14 2 65.5, 51.8 34.2 +/- 0.0 100 PE 29.9 +/- 0.5100 14 2 6.7, 0.3 33.8 +/- 0.2 100 PE 28.3 +/- 0.2 *10 18 2 2.5, 0 32 100 PE 31.9 *1 18 2 0, 0 0 0 0Roche 454 GSFLX1,500 0 2 0.30, 0.38 32.5 +/- 0.7 408 +/-1115.4 +/- 0.410 15 (LA) 3 0.91, 0.90,0.8532.8 +/- 0.8 377 +/-1531.5 +/- 4.00.01 25 (LA) 3Ion TorrentPGM 316 chip1,000 5 2 2.3, 2.4 16.3 +/- 0.2 105 +/- 5 40.3 +/- 7.6ABI 3730xl 8,000 0 1 0.7 44.6 603 7.9Experiment2Tara OceansStation 41Illumina HiSeq200010 9 (N) 1 20.3 34.8 101 PE 36.310 12 2 18.6, 31.3 34.2 +/- 0.2 101 PE 36.2 +/- 0.910 15 1 15.4 34.3 101 PE 35.7100 12 1 17.7 34.6 101 PE 35.0Tara OceansStation 109Illumina HiSeq200010 9 (N) 1 2.6 34.9 101 PE 35.410 12 1 20.4 34.9 101 PE 34.310 15 2 28.6, 16.2 34.4 +/- 0.5 101 PE 33.6 +/- 0.6100 12 1 16.7 34.8 101 PE 34.3Starting DNA refers to the amount of pre-size selection DNA used in library construction; Library amplification abbreviations are LA = linker amplification andN = Nextera; Raw quality scores reported are PHRED scores; Raw length ‘PE’ denotes paired end reads. * denotes the successful 10ng library and one of the 100 nglibraries had an additional 40% of QC-passed reads that were lost due to removal of TruSeq adaptor sequence contaminants.Solonenko et al. BMC Genomics 2013, 14:320 Page 3 of 12http://www.biomedcentral.com/1471-2164/14/320frequency duplicate reads (0.0005 to 0.9%) (Additional file1: Figure S4). However, these deep internal nodes lackedsupport, with bootstraps less than the intra-replicate 90%value, and duplication frequencies do not obviously correl-ate to any single metagenome category (e.g., technology,amplification, DNA amount, or paired end).Some duplicate sequences may be real. For example,one 100bp sequence is overrepresented in multiple li-braries including 1,000ng Illumina (0.14% of the totalreads), Ion Torrent (0.006%), and unamplified 454(0.36%) libraries. Artificial duplicate frequency corre-lations (see Online Methods) match overall duplicatefrequencies for all treatments except a single 10ng,poorly-amplified, adapter-containing Illumina library(Additional file 1: Figure S5-7), where 40% of the readswere predominantly artificial, high frequency duplicates(Additional file 1: Figure S8 and S9).Gene function and read assembly parallel %G + C findingsTo evaluate variations in gene function, metagenomicreads were compared to an expansive database of marinevirus protein sequences (>456K protein clusters derivedfrom over 6M reads from 32 diverse pelagic ocean viruscommunities [18]). As is common for viral metagenomes(reviewed in ref. [18]) only 3—7% of the reads mappedto protein clusters without self-clustering. However, theresulting gene frequency patterns were well-supportedand mirror patterns observed in the above %G + C ana-lyses (Figure 3A). Replicate metagenomes were mostsimilar (pairwise r-values >0.95), while the biggest di-fference was between metagenomes prepared from ≥1,000 ng of starting DNA and those prepared from100ng or less (r-values <0.8). Within these two largeclusters, sequencing technology choice contributed add-itional, but minor, divergences (r-values 0.8—0.9). Not-ably, these protein cluster pairwise r-values are lowerthan those for either %G + C or duplicate frequency.This likely reflects increased analytical resolution, as1,500 protein clusters correlated per metagenome in thefunction analysis, while only 50 or 10 bins were resolvedin the %G + C and duplicate analyses, respectively.Finally, assembly experiments (see Methods, Figure 3B)revealed that total assembly size positively correlated tothe number of reads used in assembly. In turn, the max-imum and N50 contig sizes were relatively insensitive toincreased read numbers in the larger datasets. This wasI F B 14 1000I R A 14 1000I F A 14 1000I R B 14 1000I F B 14 100I R B 14 100I F A 18 10I R A 18 104 F A NA 15004 F B NA 1500S F A NA 80004 F A 25 0.014 F A 15 104 F C 15 104 F B 15 104 F B 25 0.014 F C 25 0.01I F A 14 100I R A 14 100T F A 5  1000T F B 5  1000S F A NA 80004 F A 15 104 F B 15 104 F C 15 104 F A 25 0.014 F B 25 0.014 F C 25 0.01I F A 14 100I R A 14 100I F B 14 100I F A 18 10I R B 14 100I R A 18 10I R B 14 1000I F A 14 1000I R A 14 1000I F B 14 1000T F A 5  1000T F B 5  10004 F A NA 15004 F B NA 15000.000 0.010 0.020UPGMA Distance969699998489640. Distance1001001009410010010010098989497726410010067TechPairRep AmpngTechPairRep AmpngS F A NA 80004 F A 15 104 F B 15 104 F C 15 104 F A 25 0.014 F B 25 0.014 F C 25 0.01I F A 14 100I R A 14 100I F B 14 100I F A 18 10I R B 14 100I R A 18 10I R B 14 1000I F A 14 1000I R A 14 1000I F B 14 1000T F A 5  1000T F B 5  10004 F A NA 15004 F B NA 1500I F B 14 1000I R A 14 1000I F A 14 1000I R B 14 1000I F B 14 100I R B 14 100I F A 18 10I R A 18 104 F A NA 15004 F B NA 1500S F A NA 80004 F A 25 0.014 F A 15 104 F C 15 104 F B 15 104 F B 25 0.014 F C 25 0.01I F A 14 100I R A 14 100T F A 5  1000T F B 5  1000%G+C Distribution Duplicate Frequency0.97 0.98 0.99 1Pearson’s r010305070Count0.8 0.85 0.9 0.95 1Pearson’s r01020304050CountFigure 2 %G+ C and duplication plots for Experiment 1 metagenomes. Heatmap coloring indicates the relative pairwise correlations(Pearson’s r) in the %G + C distributions (red-to-yellow) and duplicates (blue-to-green) where red and blue colors indicate the lowest levels ofcorrelation, while white represents highly correlated data. The %G + C distribution correlations were UPGMA clustered with 100 bootstrap runs toindicate statistical support (only >60% support shown). Abbreviations are as follows: “Tech” is sequencing technology represented by 4 (454),T (Ion Torrent), I (Illumina), S (Sanger); “Pair” is the forward or reverse paired end sequence data; “Rep” is the arbitrarily labeled replicate rangingfrom two (A and B) to three (A, B, or C); “ng” is the nanograms of input DNA from which the viral metagenome was derived. The most reliableestimate of the true %G + C distribution is the unamplified 454 metagenomes. Relative to these, fosmid end sequences generated using Sangersequencing were the most shifted toward high %G + C, while problematic <1000ng input DNA metagenomes were less shifted towardhigh %G + C, and reliable 1000ng Illumina metagenomes were only slightly shifted toward high %G + C.Solonenko et al. BMC Genomics 2013, 14:320 Page 4 of 12http://www.biomedcentral.com/1471-2164/14/320true for both k-mer and overlap-based assembly algo-rithms (see Methods).Experiment 2: The independent effects of input DNA andlibrary amplification on Illumina-sequencedmetagenomesLow input DNA library success improved with optimizationIn contrast to Experiment 1, all 10 Experiment 2 Illuminalibraries (eight 10ng and two 100ng libraries) were suc-cessful. Replicate libraries did not cluster together consist-ently, but this reflected the extremely minimal varianceacross the replicates rather than poor replication (Figure 4,note reduced axis scales relative to Figure 2).Transposon-based library preparation slightly impacts%G + CIn both Tara Oceans station 41 and 109 datasets, theamount of input DNA (10 or 100 ng) and amplification(12 or 15 cycles) resulted in less variation than was ob-served in replicate library preparations (Figure 4). Theonly exception was transposon-based libraries, which di-verged from the relatively invariant standard Illumina li-braries. For all samples, duplicate frequencies varied asmuch between as within treatments (Figure 4) and muchless duplication was observed in Experiment 2 than 1. Thedendrogram topology observed in pairwise %G+C ana-lyses was recovered in analyses of function (Figure 5A),I F A 14 100I F A 18 10I R A 14 100I R A 18 10I R B 14 100I F B 14 1004 F A 25 0.014 F C 15 104 F C 25 0.014 F B 25 0.014 F A 15 104 F B 15 104 F A NA 15004 F B NA 1500I R A 14 1000I R B 14 1000I F A 14 1000I F B 14 1000T F B 5  1000T F A 5  10000.7 0.85 1Pearson’s r01020Count0. Dist.78969585100100100736110099926297TechPairRepAmpngI F A 14 100I F A 18 10I R A 14 100I R A 18 10I R B 14 100I F B 14 1004 F A 25 0.014 F C 15 104 F C 25 0.014 F B 25 0.014 F A 15 104 F B 15 104 F A NA 15004 F B NA 1500I R A 14 1000I R B 14 1000I F A 14 1000I F B 14 1000T F B 5  1000T F A 5  1000100101102103104105106107108109I F A 18   10 I R A 18   10 I F A 14  100 I R A 14  100 I F B 14  100 I R B 14  100 I F A 14 1000 I R A 14 1000 I F B 14 1000 I R B 14 1000 T F A 5  1000 T F B 5  1000 T F A 5  1000 T F B 5  1000 R F A NA 1500 R F B NA 1500 R F A 15   10 R F B 15   10 R F C 15   10 R F A 25 0.01 R F B 25 0.01 R F C 25 0.01 Contig N50 Contig MaxAssembly bp QC’d reads Velvet    NewblerProtein ClustersFigure 3 Protein cluster functional analysis and assembly statistics for Experiment 1 metagenomes. Metagenomic reads were mapped toPOV protein clusters (see text) and hit frequencies were used to produce pairwise correlation heat maps. Details as described in Figure 2,including bootstrap analysis of statistical support for correlations across metagenomes. Assembly performance of each sample across the datasetwas evaluated using metrics of n50 and maximum contig size, as well as the number of reads and base pairs that were assembled. Note thatinferior assembly performance was restricted to samples with reduced read yields. Lastly, the Newbler assembler yielded larger contigs andsmaller total assemblies when compared to Velvet assembly of the same Ion Torrent dataset.Solonenko et al. BMC Genomics 2013, 14:320 Page 5 of 12http://www.biomedcentral.com/1471-2164/14/320but not assembly (Figure 5B), where the transposon-basedtreatment for the Station 109 sample produced manyfewer reads than other metagenomes.DiscussionReplication is fundamental to rigorous experimental de-sign [25], but it is only now becoming financially possiblefor metagenomic studies [26,27]. Here we examined repli-cate metagenomes across varied DNA input amounts, li-brary preparation procedures, and sequencing platforms.Low input DNA library success depends on adaptorligationWhile all ≥1,000 ng DNA libraries were successful, envi-ronmental samples, particularly for viruses, routinelyyield <1ng of DNA [15]. Libraries constructed from≤100 ng DNA were successful using the linker-am-plification protocol for 454 [15], but Illumina librariesfailed or were low-quality for Experiment 1, but notExperiment 2. Two separate protocols were used – bothoptimized for recovery from column purification steps[28], but employed different template:adaptor ratios inligation [29]. Specifically, Experiment 1 used 170:1, whileExperiment 2 used 22:1 for 10ng starting DNA. Thus lowDNA libraries require adjusted adaptor:template ratiosduring ligation (see Genoscope protocol for guidelines).Presence of library amplification drives biasTwo amplification reactions are common in metagenomicsample preparations. The first, library amplification, in-creases input DNA to balance library preparation lossesfrom purification, size selection, and quality titrations [8].S F B 12 10S R B 12 10N R A 9  10N F A 9  10S R A 12 10S F A 12 100S F A 15 10S R A 12 100S F A 12 10S R A 15 100.00000 0.00010 0.00025UPGMA Dist.8376PrepPairRepAmpngS F B 12 10S R B 12 10N R A 9  10N F A 9  10S R A 12 10S F A 12 100S F A 15 10S R A 12 100S F A 12 10S R A 15 100.9994 0.9996 0.9998 1Pearson’s r0246810CountN F A 9  10N R A 9  10S F B 12 10S R B 12 10S F A 12 100S R A 12 100S F A 15 10S F A 12 10S R A 12 10S R A 15 100.0000.0060.012UPGMA Dist.10010063931008986024681014CountPrepPairRepAmpng N F A 9  10N R A 9  10S F B 12 10S R B 12 10S F A 12 100S R A 12 100S F A 15 10S F A 12 10S R A 12 10S R A 15 10St. 41 %G+C Distribution Duplicate Frequency Distribution0.975 0.985 0.995 1Pearson’s rS R B 15 10N F A 9  10N R A 9  10S F A 12 100S R A 12 100S F B 15 10S R A 12 10S R A 15 10S F A 12 10S F A 15 10PrepPairRepAmpngS R B 15 10N F A 9  10N R A 9  10S F A 12 100S R A 12 100S F B 15 10S R A 12 10S R A 15 10S F A 12 10S F A 15 100.997 0.998 0.999 1Pearson’s r051015Count0.0000 0.0010UPGMA Dist.7063N F A 9  10N R A 9  10S F A 12 100S F A 15 10S F A 12 10S R A 12 10S R A 15 10S R A 12 100S F B 15 10S R B 15 10PrepPairRepAmpng0.0000.0020.004UPGMA Dist.1001001006397N F A 9  10N R A 9  10S F A 12 100S F A 15 10S F A 12 10S R A 12 10S R A 15 10S R A 12 100S F B 15 10S R B 15 100.994 0.996 0.998 1Pearson’s r05101520CountSt. 109 %G+C Distribution Duplicate Frequency DistributionFigure 4 %G+ C and duplication plots for Experiment 2 metagenomes. Details as described in Figure 2, including bootstrap analysis ofstatistical support for correlations across metagenomes. UPGMA clustering bootstrap support >60% shown only.Solonenko et al. BMC Genomics 2013, 14:320 Page 6 of 12http://www.biomedcentral.com/1471-2164/14/320This adaptor-mediated amplification step is used for limit-ing DNA for 454 (15—25 cycles [15]), but is routinelyemployed in Ion Torrent (5 cycles) and Illumina (12—16cycles) to enrich for correctly ligated adaptors. This stepcan alter overall library %G+C [15,17,30]. The secondamplification step is specific to the sequencing technology(e.g., emPCR in 454 or Ion Torrent, bridge amplificationin Illumina) and used for improving signal detection. Thisstep should not alter overall library %G+C, but can artifi-cially over-represent sequences [23,24].In this study, two libraries received no library amplifica-tion: unamplified 454 and fosmid libraries. Fosmids hadelevated %G+C, which is ascribed to a cloning bias [26].Among the remaining libraries, we expected a low %G+Cshift due to the adaptor-mediated amplification step, com-monly attributed to inhibitory effects of high %G+CDNA secondary structures, either during library amplifica-tion [30] or downstream emPCR [31]. However, thesetrends were not observed: in Experiment 1, the 454 un-amplified and amplified Illumina 1,000 ng libraries corre-late well with one another (r-values > 0.99), but poorly(r-values < 0.9) with the amplified (18 cycles) 10ngIllumina libraries. This difference appears to be driven byreduced low %G+C reads relative to the ≥1,000 ng libra-ries, which may implicate low input DNA libraries as moresensitive to loss of low %G+C reads either during gel ex-traction heat steps [32] or preferential fragmentationthrough heating [33]. A possible improvement over gel ex-traction is Sage Science’s Pippin Prep (tested with 65ng ofDNA, see Figure 2B in ref. [15]), which avoids heating.Heat during fragmentation is avoidable with Covarisacoustic shearing. Both techniques also minimize conta-mination, which is crucial for DNA-limited libraries.While amplified ≤100 ng metagenomes displayed dif-ferent %G + C distributions from ≥1,000 ng metage-nomes, the amount of amplification only minimallyimpacts the resulting metagenomes. This was true in Ex-periment 1, where starting DNA amount and amplifica-tion cycling co-varied, as well as Experiment 2, wherethese parameters were independent. Fragment competi-tion resulting from cycling conditions is thought to selectfor higher %G + C and shorter fragments, thus linker-mediated amplification protocols employ tight sizingconditions and %G + C optimized PCR conditions [30].Such careful library construction can produce minimallybiased (<1.5-fold %G + C variation) viral metagenomesfrom sub-nanogram amounts of DNA [10,15]. The %G +C patterns observed in the current larger-scale studywere also paralleled in functional analyses (protein clus-ter mapping) and assembly performance. This suggestsN R A 9  10N F A 9  10S R B 12 10S F A 12 10S R A 12 10S F B 12 10S F A 12 100S R A 12 100S R A 15 10S F A 15 100.975 0.99 1048Count00.0100.02010010010087769076PrepPairRepAmpngN R A 9  10N F A 9  10S R B 12 10S F A 12 10S R A 12 10S F B 12 10S F A 12 100S R A 12 100S R A 15 10S F A 15 10S F A 12 100S R A 12 100S F A 12  10S R A 12  10S F B 12  10S R B 12  10S F A 15  10S R A 15  10N F A  9  10N R A  9  10100101102103104105106107108109N F A 9  10N R A 9  10S R A 15 10S F A 15 10S R A 12 10S F A 12 10S F A 12 100S R A 12 100S F B 15 10S R B 15 100.994 0.998048Count0.0010.0030.005100100100100999972100N F A 9  10N R A 9  10S R A 15 10S F A 15 10S R A 12 10S F A 12 10S F A 12 100S R A 12 100S F B 15 10S R B 15 101PrepPairRepAmpngS F A 12 100S R A 12 100S F A 12  10S R A 12  10S F A 15  10S R A 15  10S F B 15  10S R B 15  10N F A  9  10N R A  9  10100101102103104105106107108109Figure 5 Protein cluster functional analysis and assembly statistics for Illumina-sequenced Experiment 2 metagenomes. Note that onemetagenome from Station 109 DNA yielded significantly fewer reads and thus had a lower total assembly size. Details as described in Figure 3,including bootstrap analysis of statistical support for correlations across metagenomes.Solonenko et al. BMC Genomics 2013, 14:320 Page 7 of 12http://www.biomedcentral.com/1471-2164/14/320that systematically prepared linker-amplified metage-nomes derived from variable input DNA amounts arequantitatively comparable.Some caution is warranted for high-throughput trans-poson-based library preparation options like Nextera. Spe-cifically, Experiment 2 revealed that standard librariesprepared from limiting DNA and under varied conditionswere relatively invariant, whereas the transposon-basedprotocol led to divergent %G+C and protein cluster pro-files for metagenomes from both stations. While thesedeviations were statistically significant (90% bootstrapclustering in Figures 4 and 5), they were minor in magni-tude relative to other treatment effects observed here.Such a %G+C bias in Nextera library preps is not entirelysurprising as previous work demonstrated reduced cover-age in both high and low %G+C regions of virus genomes[34], presumably due to non-random transposition. Evalu-ation of new transposition methods should be consideredif their eventual products require strictly unbiased repre-sentation of input DNA.Finally, while not investigated here, polymerases used inamplification can alter metagenomes. Phi29 polymerase,for example, leads to stochastic and systematic biases thatcan impact resulting coverage [13], while some high-fidelity polymerases (e.g., TAKARA) enrich for rare se-quences and others (e.g., PfuTurbo) do not [11,15]. InExperiment 1, the ≥1,000 ng libraries only minimally dif-fered from each other despite the fact that they employdifferent polymerases across sequencing platforms. Thesepolymerase-specific effects would depend on protocol par-ticulars (e.g., PCR cycler settings and additives) [17,30]and the underlying %G+C distribution (particularly for<20% or >80% G +C fragments) of the DNA to beamplified. Future work to determine the impact of poly-merase choice empirically on metagenomes derived froma wider range of %G +C than those employed here wouldbe informative.Duplicates vary by input DNA, amplification, technologyDuplicated reads are problematic in quantitative applica-tions as they can be real or artificial [23,24,35,36]. Here,Experiment 1’s true distribution of duplicates is presum-ably represented by the first cluster (includes unamplified454 libraries), except the artificial duplicates discussedbelow. By comparison, metagenomes from the secondcluster contained highly duplicated artificial reads that re-duced library complexity during amplification. The lastcluster, which included amplified 454, as well as oneIllumina and two Ion Torrent metagenomes, had lowlevels of duplication. For the 454 libraries, this could bedue to the diversifying effects of the linker amplificationprocess [15], but it is harder to explain this trend in theIon Torrent metagenomes or find a process that ties lowlibrary amplification in the 100ng Illumina metagenometo lower duplication levels. Artificial duplicates in Illuminalibraries were only an issue in the problematic 10ng li-brary, where 40% of the reads were high-frequency, pre-dominantly artificial duplicates. Further study is requiredto determine mechanisms that generate artificial dupli-cates in Illumina data.Sequencing technologies produce comparable outputWhile the metagenomes here were derived from three verydifferent ocean viral communities, the range of %G+Cwas not extreme. Given that, sequencing technology is nota major factor impacting ocean viral metagenomes, whichis consistent with previous microbial metagenomic studies[9]. However, read length can influence many downstreamapplications, from assembly efforts to functional identifica-tion of genes [37,38]. Of widely used next-generationtechnologies, 454 currently has the longest read length of800bp, with paired-end Illumina capable of 250 + bp [7].However, emerging nanopore technologies are likely to betruly transformative [39]. Details are not yet public, butthese technologies promise longer reads, direct observa-tion of fragment sequences, and minimal library prepar-ation enabling low input DNA applications.ConclusionsAs we strive for systematic and quantitative analyses ofcomplex environments, a thorough understanding ofempirically-documented biases in methods is critical. Herewe demonstrate that while sequencing platform choiceand degree of amplification have little impact on resul-ting metagenomes, presence of amplification and startingDNA amounts do influence library success and compos-ition. Our findings are critical both for the interpretationof systematic comparisons of DNA-limited communitymetagenomes, as well as for novel methods of studyingvirus-host interactions [40-42] that generate small a-mounts of DNA. Notably, however, high replicability ob-served here might have been aided by diluting the initialconcentrated DNA sample, and potential inhibitors, to ob-tain ‘low input DNA’ samples. Consideration should bemade of the impact of inhibitors on low input DNA sam-ples, particularly when amplification steps are needed forsample preparation.Given current findings, unamplified libraries are bestwhen DNA is not limiting (>2 ug) [43] while sequencingplatform choice minimally impacts quantitative repre-sentation in the resulting metagenomes. When DNA islimiting, as in viral community samples or microbial com-munities of permafrost soils or air samples, specific rec-ommendations for quantitative metagenomics are asfollows. Low input DNA (1—100 ng) libraries can utilizeeither a linker-amplified protocol [15] optimized for theappropriate sequencing technology of choice [10] or, forIllumina sequencing, standard library preparations whereSolonenko et al. BMC Genomics 2013, 14:320 Page 8 of 12http://www.biomedcentral.com/1471-2164/14/320adaptor:template ratios are carefully controlled. For sam-ples with ultra-low DNA yields (<1 ng), it is best not torisk failure in standard library preparations and to use in-stead a sequencing technology optimized linker-amplifiedprotocol. Future research directions include developing amechanistic understanding of the non-intuitive, but rep-licable differences in linker-amplified metagenomes, aswell as improving understanding of polymerase impactsand developing empirical datasets for a broader range of%G +C samples.MethodsSource DNAs and sample preparation detailsExperimental protocol availabilityAll detailed protocols are listed by name, and are docu-mented and available at http://eebweb.arizona.edu/Fac-ulty/mbsulli/protocols.htm.Briefly, FeCl-precipitated viral concentrates wereobtained from 0.2μm filtered seawater collected from theman-made Biosphere 2 Ocean in December 2010, as wellas Stations 41 (Indian Ocean, 14°34.572 N 70°1 E, deepchlorophyll maximum) and 109 (south Pacific Ocean, 1°58.286 N 84°26.772 W, deep chlorophyll maximum) of theTara Oceans expedition on March 30th, 2010, and May12th, 2011, respectively. The viral concentrate from theformer was purified using both CsCl and DNase, whileonly DNAse was used for the latter.DNA Source for B2O metagenomes (Biosphere 2 Ocean)The B2 Ocean environment is host to a stable microbialcommunity, as measurements of microbial phyletic fre-quencies are consistent across samples taken a year apart(Additional file 2). FeCl precipitation [12] was used to con-centrate viruses from 1,080L of 0.2 μm filtered seawater,which were then DNase I treated [11] to remove freeDNA, cesium chloride purified to remove microbial con-taminants (dsDNA viral band was pulled 1.4—1.52 g/ml[11]), and further concentrated to 4mL using an Amicon30KDa filter. The final yield was 1.26 × 1012 SYBR-stainedvirus particles. DNA was extracted using the Wizard PrepDNA Purification system (Promega, cat# A7211 andA7181).DNA Source for TARA metagenomes20—60L seawater was collected and filtered for two TARAOceans [21] stations using the protocol described above.These samples yielded 690 ng (station 41) and 950 ng(station 109) of DNA, using the Wizard Prep DNA Purifi-cation system. Starting DNA amounts of 10 and 100 ngwere used in Illumina sequencing library construction asdescribed in the Genoscope protocol (Genoscope Illuminaprotocol).454 Library Prep (Sullivan lab)The linker amplification protocol was used to generateamplicon libraries for 454 sequencing, as well as am-plification-free libraries, as previously described [15].Briefly, genomic DNA was Covaris-sheared, unidirection-ally ligated to an adaptor, and amplified using adaptor-specific primers using 15 to 25 amplification cycles,depending on the starting DNA amount (a description ofthe amount of cycling and relationship to input DNA weredocumented previously [15]). Following the addition ofbarcodes, sequencing libraries were ligated to 454-specificadaptors.Fosmid Library Prep (Hallam lab)8μg of B2O viral DNA was used in large-insert fosmid li-brary construction using the Epicentre CopyControlFosmid Library Production Kit (CCFOS110) as previouslydescribed [44]. A total of 17 384-well plates of clones werepicked, and 384 fosmids were sequenced bi-directionallywith Sanger sequencing.Ion Torrent Library Prep (University of ArizonaGenomics Core)2μg of B2O viral DNA was used for sequencing librarypreparation following the Ion Fragment Library Kit UserGuide (Rev July 2011), loaded onto beads, emPCR-ed, thensequenced using the 316 chip on the Ion Torrent PGM.Illumina Library Prep for B2O metagenomes (EmoryGenomics Core)DNA samples were Covaris-sheared and size-selected to300—600bp using SPRI Size Selection chemistry, enrich-ment amplified using Phusion DNA polymerase accordingto starting amount of DNA (14—18 cycles), and pairedend sequenced. Two libraries starting with 1ng of DNAfailed to amplify to sufficient amounts, even with the useof a carrier DNA protocol (Emory carrier DNA protocol).One 10 ng library experienced the same problem, and wasnot sequenced. The libraries were multiplexed on two se-quencing lanes, with one replicate of each starting amountlibrary present together on each lane.Illumina Library Prep for TARA metagenomes (Genoscope)DNA samples were Covaris-sheared and size selected to160—180bp, amplified according to starting amount ofDNA (9—15 cycles) and paired-end sequenced. Severalmodifications of the standard Illumina protocol [32] wereintroduced in order to minimize losses of ultra-low DNAamounts. The low-fragment-size shearing settings, cou-pled with Ampure beads to remove very short fragments,ensured the recovery of appropriately sized fragmentswithout the need for gel sizing. The Pfx Platinum poly-merase was used to increase amplification efficiency andthus decrease the number of total library amplificationSolonenko et al. BMC Genomics 2013, 14:320 Page 9 of 12http://www.biomedcentral.com/1471-2164/14/320cycles. During ligation, proper adaptor ratios were chosento correspond to 2—3 fold more adaptor ends than frag-ment ends in the working ligation reaction (GenoscopeIllumina protocol). Transposon-based Nextera librarieswere prepared per manufacturer’s instructions using theIllumina compatible Nextera DNA Sample Prep Kit(Epicentre Biotechnologies, cat#GA09115).Bioinformatics methodsScript availabilityAll custom scripts are listed by name and available athttp://code.google.com/p/tmpl/.Sequencing dataAll metagenomic sequences are publically availablethrough the CAMERA portal at http://camera.calit2.net/[CAMERA: CAM_P_00001027]. 454 and Ion Torrentdata, provided by UAGC, were delivered in .sff format andconverted for downstream processing to FASTA andQUAL formats using sffinfo (roche454 v2.6) and then toFASTQ format using BioPerl 1.6.1. B2 Ocean Illuminadata, by Emory Genomics Core, and TARA OceansIllumina data, by Genoscope, were provided in FASTQformat. Each library was examined for raw qualityusing FastQC (v0.9, downloaded Aug 2012 from http://www.bioinformatics.babraham.ac.uk/projects/fastqc/) andFastx_Toolkit (v0.0.13 downloaded Feb 2010 from http://hannonlab.cshl.edu/fastx_toolkit/). The FastQC reportwas the source of duplication data used in the figures.Adapter sequences were detected in two metagenomes(I1A18N10 and I1A14N100) through the overrepresentedsequences functionality of FastQC. The fastx_toolkit util-ity ‘fastx_clipper’ was used with the –C option to removeall reads matching the above adapter motif from the for-ward paired end reads, removing approximately 40% ofthe reads that passed QC in each of these libraries.Quality controlNext, procedures for quality control were established toremove suspect sequence data, either by filtering wholereads or trimming reads in accordance with known se-quencing technology artifacts. For 454 and Ion Torrentdata, whole-read filtering was used, as is common formetagenomics [11,15,45,46] (Additional file 1: Figure S10).In contrast, because Illumina errors are localized to par-ticular parts of a read [47,48], these data were trimmedusing a threshold predicted quality score to removesuspect regions of the read at both the 3’ and 5’ ends usingDynamicTrim.pl, part of the SolexaQA package [49](Additional file 1: Figure S11). After QC steps, 69—85% ofthe 454 reads remained, compared to 60% for Ion Torrentand 63—74% for Illumina (Table 1). The fastx_toolkit soft-ware package was also used to remove Illumina readsunder 50bp, while the 454 and Ion Torrent reads werecleaned using a custom pipeline [18]. This processingensured that the data analyzed would be analogous tothat used for metagenomic inference. FastQC andFastx_Toolkit were used to check the QC process of eachmetagenome.%G + C analyticsThe mean read %G+C was chosen as the focus of ouranalysis, rather than the %G+C of sequence subsets of aread or the larger genome regions from which the readfragment originated, since mean fragment %G+C the bestpredictor of GC bias [50]. QC-ed reads were processedusing the BioPerl 1.6.1 script bp_gc_calc.pl to obtain aver-age %G+C values for each read. Given the large readlength differences across these libraries (90bp to 350bp),only the first 50bp of each read are used in all %G + C dis-tribution analyses to match the shortest QC-ed Illuminadata, while normalizing for read length. Reads were trun-cated to 50bp using fastx_toolkit and processed withbp_gc_calc.pl. Phage metagenomic reads were cut intonon-overlapping 50bp fragments using a bash script andalso processed with bp_gc_calc.pl.Statistical analysis and figuresR 2.14.1 (http://www.R-project.org/) was used to run acustom script, 0.02gc.R, which calculated frequencies ofreads in 2% G +C bins for each metagenome. Pearsons’s rpairwise correlation values were calculated using the cov()function, and heatmap figures were generated usingthe heatmap.2() function found in the gplots library(http://CRAN.R-project.org/package=gplots). Lastly, boot-strapped UPGMA clustering values for each node wereobtained using the pvclust() function in the pvclust library(http://CRAN.R-project.org/package=pvclust), with pair-wise distances calculated from Pearson’s correlation valuesand hierarchical clustering done using the “average”method.Duplicate analysesDuplication levels were assessed in raw reads by countingthe occurrence of duplicates only in the starting 50bp ofeach read using the FastQC duplication level utility out-put, normalized to total metagenome size to reflect rela-tive frequencies. Artificial duplicates were defined as thosewith identical starts and >95% identity throughout theread, and were detected using CD-HIT-454 [51] andCD-HIT-DUP [52] with default parameters.Protein cluster analysesFunctional differences within and between metagenomeswere assessed in Experiment 1 by mapping metagenomicreads to the Pacific Ocean Virome database [18]. The hitfrequencies of the 1,500 protein clusters that were mostabundant across all metagenomes were then used toSolonenko et al. BMC Genomics 2013, 14:320 Page 10 of 12http://www.biomedcentral.com/1471-2164/14/320obtain pairwise correlation values. A range of 3—7% ofthe metagenomic reads mapped to these POV PCs, whilethe ‘top 1,500 PCs’ subsample represented >99% of thedata that mapped. Because the Experiment 1 dataset rep-resented a large diversity of read lengths, greatly impactinginference capacity [38], the dataset was normalized to as-sess sequencing platform biases rather than read lengthimpacts as follows: (i) the longer Ion Torrent and 454reads were trimmed to 100bp, and (ii) only reads ≥100 bpwere used from Illumina data.Assembly analysesThe short reads derived from Illumina and Ion Torrentdata were assembled using Velvet v 1.2.03 [53] using de-fault parameters across a range of kmer sizes (23, 27,31bp), but only 31-mer data are reported as kmer sizedid not impact assemblies. The longer 454 reads wereassembled using GS De Novo Assembler v2.6 (http://my454.com/products/analysis-software/index.asp) withdefault parameters.Additional filesAdditional file 1: Figures S1-S11. A log-log plot of all B2 Oceanmetagenome read yields per starting DNA amount (Figure S1).% G + C histogram of several ‘problematic’ and ‘reliable’ libraries,and GC distribution of full dsDNA bacteriophage genomes for reference(Figure S2). %G + C distribution differences between whole-read mean% G + C in unamplified 454 metagenome, in green, and Sanger-sequenced fosmid library, in blue, shows a shift toward high %G + C inthe fosmid library (Figure S3). Duplicate frequencies in Experiment 1metagenomes (Figure S4). Heatmap of Pearson’s r pairwise correlationvalues for artificial duplicate frequencies, as detected using CD-HIT-454for 454 and Ion Torrent data and CD-HIT-DUP for Illumina data(Figure S5). CD-HIT-454 artificial duplicate frequencies in Experiment 1metagenomes generated using 454 and Ion Torrent sequencing(Figure S6). Duplicate frequency minus artificial duplicate frequencyfor Experiment 1 CD-HIT-454 –processed metagenomes (Figure S7).CD-HIT-DUP artificial duplicate frequencies in Experiment 1 Illuminametagenomes(Figure S8). Duplicate frequency minus artificial duplicate frequency forExperiment 1 CD-HIT-DUP –processed metagenomes (Figure S9).Ion Torrent QC length distribution (Figure S10). Methods for TrimmingIllumina Reads (Figure S11).Additional file 2: Pyrotag data for microbial composition of Biosphere 2Ocean in Nov 2008 and Sep 2009. The Biosphere 2 Ocean was the sourceof the DNA sample used in Experiment 1 metagenomes. The distributionof microbial phyla in the B2 Ocean community appears stable across twosamples taken a year apart.Competing interestsThe authors declare that they have no competing interests.Authors’ contributionsSAS and MBS conceived the project and designed the experiments withcontributions from AA, SH, KK, GT and PW. AA and CC performed experiments.SAS and JCIE collected and analyzed the results. SAS, MBS, SH, KK, GT wrote themanuscript. All authors read and approved the final manuscript.AcknowledgementsWe thank B. Poulos, N. Solonenko, A. Gregory, and C. Decker for technicalassistance, as well as TMPL members and two anonymous reviewers forcomments on the manuscript. Funding for this particular study was providedby BIO5, Biosphere 2 and the Gordon and Betty Moore Foundation to MBS.We thank the coordinators and members of the Tara Oceans consortium(http://www.embl.de/tara_oceans/start/) for organizing sampling and dataanalysis. We thank the commitment of the following people and sponsorswho made this singular expedition possible: CNRS, EMBL, Genoscope/CEA,VIB, Stazione Zoologica Anton Dohrn, UNIMIB, ANR (projects POSEIDON/ANR-09-BLAN-0348, BIOMARKS/ANR-08-BDVA-003, PROMETHEUS/ANR-09-GENM-031, and TARA-GIRUS/ANR-09-PCS-GENM-218), EU FP7 (MicroB3/No.287589),FWO, BIO5, Biosphere 2, agne`s b., the Veolia Environment Foundation,Region Bretagne, World Courier, Illumina, Cap L’Orient, the EDF FoundationEDF Diversiterre, FRB, the Prince Albert II de Monaco Foundation, EtienneBourgois, the Tara schooner and its captain and crew. Tara Oceans wouldnot exist without continuous support from 23 institutes (http://oceans.taraexpeditions.org). This article is contribution number 0005 of the TaraOceans Expedition 2009–2012.Author details1Department of Ecology and Evolutionary Biology, University of Arizona,Tucson, AZ, USA. 2Department of Molecular and Cellular Biology, Universityof Arizona, Tucson, AZ, USA. 3CEA, DSV, IG, Genoscope, 2 rue GastonCrémieux CP5706, 91057, Evry, Cedex, France. 4Department of Microbiologyand Immunology, University of British Columbia, Vancouver, BC, Canada.5Department of Civil and Environmental Engineering, Georgia Institute ofTechnology, Atlanta, GA, USA. 6Austalian Center for Ecogenomics, Universityof Queensland, Brisbane, QLD, Australia.Received: 4 February 2013 Accepted: 2 May 2013Published: 10 May 2013References1. Chaffron S, Rehrauer H, Pernthaler J, von Mering C: A global network ofcoexisting microbes from environmental and whole-genome sequencedata. Genome Res 2010, 20:947–959.2. Shapiro BJ, Friedman J, Cordero OX, Preheim SP, Timberlake SC, Szabo G,Polz MF, Alm EJ: Population genomics of early events in the ecologicaldifferentiation of bacteria. Science 2012, 336:48–51.3. Handelsman J, Tiedje JM, Alvarez-Cohen L, Ashburner M, Cann IKO, DelongEF, Doolittle WF, Fraser-Liggett CM, Godzik A, Gordon JI, et al: New Scienceof Metagenomics: Revealing the Secrets of Our Microbial Planet. Nat ResCouncil Report 2007:13.4. Glenn TC: Field guide to next-generation DNA sequencers. Mol EcolResour 2011, 11:759–769.5. Kircher M, Kelso J: High-throughput DNA sequencing–concepts andlimitations. BioEssays : news and reviews in molecular, cellular anddevelopmental biology 2010, 32:524–536.6. Metzker ML: Sequencing technologies - the next generation. Nat RevGenet 2010, 11:31–46.7. Loman NJ, Misra RV, Dallman TJ, Constantinidou C, Gharbia SE, Wain J,Pallen MJ: Performance comparison of benchtop high-throughputsequencing platforms. Nat Biotechnol 2012, 30:434–439.8. Linnarsson S: Recent advances in DNA sequencing methods - generalprinciples of sample preparation. Exp Cell Res 2010, 316:1339–1343.9. Luo C, Tsementzi D, Kyrpides N, Read T, Konstantinidis KT: Directcomparisons of Illumina vs. Roche 454 sequencing technologies on thesame microbial community DNA sample. PLoS One 2012, 7:e30087.10. Duhaime M, Sullivan MB: Ocean viruses: Rigorously evaluating themetagenomic sample-to-sequence pipeline. Virology 2012, 434:181–186.11. Hurwitz BH, Deng L, Poulos B: Sullivan. Evaluation of methods toconcentrate and purify wild ocean virus communities through comparative,replicated metagenomics. Environ Microbiol: MB; 2012. doi:10.1111/j.1462-2920.2012.02836.x.12. John SG, Mendez CB, Deng L, Poulos B, Kauffman AKM, Kern S, Brum J, PolzMF, Boyle EA, Sullivan MB: A simple and efficient method forconcentration of ocean viruses by chemical flocculation. Environ MicrobiolRep 2011, 3:195–202.13. Yilmaz S, Allgaier M, Hugenholtz P: Multiple displacement amplificationcompromises quantitative analysis of metagenomes. Nat Methods 2010,7:943–944.14. Kim KH, Bae JW: Amplification methods bias metagenomic libraries ofuncultured single-stranded and double-stranded DNA viruses. ApplEnviron Microbiol 2011, 77:7663–7668.Solonenko et al. BMC Genomics 2013, 14:320 Page 11 of 12http://www.biomedcentral.com/1471-2164/14/32015. Duhaime M, Deng L, Poulos B, Sullivan MB: Towards quantitativemetagenomics of wild viruses and other ultra-low concentration DNAsamples: a rigorous assessment and optimization of the linkeramplification method. Environ Microbiol 2012, 14:2526–2537.16. Hoeijmakers WA, Bartfai R, Francoijs KJ, Stunnenberg HG: Linearamplification for deep sequencing. Nat Protoc 2011, 6:1026–1036.17. Oyola SO, Otto TD, Gu Y, Maslen G, Manske M, Campino S, Turner DJ,Macinnis B, Kwiatkowski DP, Swerdlow HP, Quail MA: Optimizing Illuminanext-generation sequencing library preparation for extremely AT-biasedgenomes. BMC Genomics 2012, 13:1.18. Hurwitz BH, Sullivan MB: The Pacific Ocean Virome (POV): a marine viralmetagenomic dataset and associated protein clusters for quantitativeviral ecology. PLoS One 2012. submitted.19. Roux S, Faubladier M, Mahul A, Paulhe N, Bernard A, Debroas D, Enault F:Metavir: a web server dedicated to virome analysis. Bioinformatics 2011,27:3074–3075.20. Wommack KE, Polson SW, Bhaysar J, Srinivasiah S, Jamindar S, Dumas M:VIROME: a standard operating procedure for classification of viralmetagenome sequences. Stand Genomic Sci 2011, 4:427–439.21. Karsenti E, Acinas SG, Bork P, Bowler C, De Vargas C, Raes J, Sullivan M,Arendt D, Benzoni F, Claverie JM, et al: A holistic approach to marine eco-systems biology. PLoS Biol 2011, 9:e1001177.22. Adey A, Morrison HG, Asan, Xun X, Kitzman JO, Turner EH, Stackhouse B,MacKenzie AP, Caruccio NC, Zhang X, Shendure J: Rapid, low-input, low-bias construction of shotgun fragment libraries by high-density in vitrotransposition. Genome Biol 2010, 11:R119.23. Dong H, Chen Y, Shen Y, Wang S, Zhao G, Jin W: Artificial duplicate readsin sequencing data of 454 Genome Sequencer FLX System. Acta BiochimBiophys Sin 2011, 43:496–500.24. Kozarewa I, Ning Z, Quail MA, Sanders MJ, Berriman M, Turner DJ:Amplification-free Illumina sequencing-library preparation facilitatesimproved mapping and assembly of (G + C)-biased genomes. NatMethods 2009, 6:291–295.25. Hurlbert SH: Pseudoreplication and the design of ecological fieldexperiments. Ecological Monographs 1984, 54:187–211.26. Danhorn T, Young CR, Delong EF: Comparison of large-insert, small-insertand pyrosequencing libraries for metagenomic analysis. ISME J 2012,6:2056–2066.27. Knight R, Jansson J, Field D, Fierer N, Desai N, Fuhrman JA, Hugenholtz P,van der Lelie D, Meyer F, Stevens R, et al: Unlocking the potential ofmetagenomics through replicated experimental design. Nat Biotechnol2012, 30:513–520.28. Kishore R, Reef Hardy W, Anderson VJ, Sanchez NA, Buoncristiani MR:Optimization of DNA extraction from low-yield and degraded samplesusing the BioRobot EZ1 and BioRobot M48. J Forensic Sci 2006,51:1055–1061.29. Sambrook J, Fritsch EF, Maniatis T: Molecular Cloning, a laboratory manual.Cold Harbor Spring Press; 1989.30. Aird D, Ross MG, Chen WS, Danielsson M, Fennell T, Russ C, Jaffe DB,Nusbaum C, Gnirke A: Analyzing and minimizing PCR amplification biasin Illumina sequencing libraries. Genome Biol 2011, 12:R18.31. Schwientek P, Szczepanowski R, Ruckert C, Stoye J, Puhler A: Sequencing ofhigh G + C microbial genomes using the ultrafast pyrosequencingtechnology. J Biotechnol 2011, 155:68–77.32. Quail MA, Kozarewa I, Smith F, Scally A, Stephens PJ, Durbin R, Swerdlow H,Turner DJ: A large genome center's improvements to the Illuminasequencing system. Nat Methods 2008, 5:1005–1010.33. Quail MA: DNA: Mechanical breakage. In Encyclopedia of Life Sciences.Chichester: John Wiley & Sons, Ltd; 2010.34. Marine R, Polson SW, Ravel J, Hatfull G, Russell D, Sullivan M, Syed F, DumasM, Wommack KE: Evaluation of a transposase protocol for rapidgeneration of shotgun high-throughput sequencing libraries fromnanogram quantities of DNA. Appl Environ Microbiol 2011, 77:8071–8079.35. Gomez-Alvarez V, Teal TK, Schmidt TM: Systematic artifacts inmetagenomes from complex microbial communities. ISME J 2009,3:1314–1317.36. Jerome M, Noirot C, Klopp C: Assessment of replicate bias in 454pyrosequencing and a multi-purpose read-filtering tool. BMC Res Notes2011, 4:149.37. Kristensen DM, Mushegian AR, Dolja VV, Koonin EV: New dimensions of thevirus world discovered through metagenomics. Trends Microbiol 2010,18:11–19.38. Wommack KE, Bhavsar J, Ravel J: Metagenomics: read length matters. ApplEnviron Microbiol 2008, 74:1453–1463.39. Wanunu M: Nanopores: A journey towards DNA sequencing. Phys Life Rev2012, 9:125–158.40. Allers E, Moraru C, Duhaime M, Beneze E, Solonenko N, Barerro-Canosa J,Amann R, Sullivan MB: Single-cell and population level viral infectiondynamics revealed by phageFISH, a method to visualize intracellular andfree viruses. Environ Microbiol 2013. in press.41. Deng L, Gregory A, Yilmaz S, Poulos BT, Hugenholtz P, Sullivan MB:Contrasting strategies of viruses that infect photo- and hetero- trophicbacteria revealed by viral-tagging. mBio 2012, 3:e00373–00312.42. Tadmor AD, Ottesen EA, Leadbetter JR, Phillips R: Probing individualenvironmental bacteria for viruses by using microfluidic digital PCR.Science 2011, 333:58–62.43. Luo C, Tsementzi D, Kyrpides NC, Konstantinidis KT: Individual genomeassembly from complex community short-read metagenomic datasets.ISME J 2012, 6:898–901.44. Taupp M, Lee S, Hawley A, Yang J, Hallam SJ: Large insert environmentalgenomic library production. J Visualized Exp: JoVE 2009. doi:10.3791/1387.45. Huse SM, Dethlefsen L, Huber JA, Welch DM, Relman DA, Sogin ML:Exploring microbial diversity and taxonomy using SSU rRNAhypervariable tag sequencing. PLoS Genet 2008, 4:e1000255.46. Rothberg JM, Hinz W, Rearick TM, Schultz J, Mileski W, Davey M, Leamon JH,Johnson K, Milgrew MJ, Edwards M, et al: An integrated semiconductordevice enabling non-optical genome sequencing. Nature 2011,475:348–352.47. Dohm JC, Lottaz C, Borodina T, Himmelbauer H: Substantial biases in ultra-short read data sets from high-throughput DNA sequencing. NucleicAcids Res 2008, 36:e105.48. Minoche AE, Dohm JC, Himmelbauer H: Evaluation of genomic high-throughput sequencing data generated on Illumina HiSeq and genomeanalyzer systems. Genome Biol 2011, 12:R112.49. Cock PJ, Fields CJ, Goto N, Heuer ML, Rice PM: The Sanger FASTQ fileformat for sequences with quality scores, and the Solexa/Illumina FASTQvariants. Nucleic Acids Res 2010, 38:1767–1771.50. Benjamini Y, Speed TP: Summarizing and correcting the GC content biasin high-throughput sequencing. Nucleic Acids Res 2012, 40:e72.51. Niu B, Fu L, Sun S, Li W: Artificial and natural duplicates inpyrosequencing reads of metagenomic data. BMC Bioinformatics2010, 11:187.52. Li W, Fu L, Niu B, Wu S, Wooley J: Ultrafast clustering algorithms formetagenomic sequence analysis. Brief Bioinform 2012, 13:656.53. Zerbino DR, McEwen GK, Margulies EH, Birney E: Pebble and rock band:heuristic resolution of repeats and scaffolding in the velvet short-readde novo assembler. PLoS One 2009, 4:e8407.doi:10.1186/1471-2164-14-320Cite this article as: Solonenko et al.: Sequencing platform and librarypreparation choices impact viral metagenomes. BMC Genomics 201314:320.Submit your next manuscript to BioMed Centraland take full advantage of: • Convenient online submission• Thorough peer review• No space constraints or color figure charges• Immediate publication on acceptance• Inclusion in PubMed, CAS, Scopus and Google Scholar• Research which is freely available for redistributionSubmit your manuscript at www.biomedcentral.com/submitSolonenko et al. BMC Genomics 2013, 14:320 Page 12 of 12http://www.biomedcentral.com/1471-2164/14/320


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items