on September 11, 2012
`
`www.sciencemag.org
`
`Downloaded from
`
`R E P O R T S
`
`3. S. Georgi, www.batteriesdigest.com/id380.htm (accessed
`June 2005).
`4. R. M. Alexander, J. Exp. Biol. 160, 55 (1991).
`5. G. A. Cavagna, N. C. Heglund, C. R. Taylor, Am. J.
`Physiol. 233, R243 (1977).
`6. J. Drake, Wired 9, 90 (2001).
`7. S. Stanford, R. Pelrine, R. Kornbluh, Q. Pei, in Proceedings
`of the 13th International Symposium on Unmanned
`Untethered Submersible Technology (Autonomous
`Undersea Systems Institute, Lee, NH, 2003).
`8. T. Starner, J. Paradiso, in Low Power Electronics Design
`(CRC Press, Boca Raton, FL, 2004), p. 45–1.
`9. G. A. Cavagna, M. Kaneko, J. Physiol. 268, 647 (1977).
`10. S. A. Gard, S. C. Miff, A. D. Kuo, Hum. Mov. Sci. 22,
`597 (2004).
`11. Supporting material is available on Science Online.
`12. Because it is a prototype, there has been no attempt
`to reduce the weight of the backpack—indeed, it is
`substantially ‘‘overdesigned.’’ Further, the 5.6 kg
`includes the weight of six load cells and one 25-cm-
`long transducer, each with accompanying brackets
`and cables, as well as other components that will not
`be present on a typical pack. In future prototypes, we
`estimate that the weight will exceed that of a normal
`backpack by no more than 1 to 1.5 kg.
`13. Under high-power conditions (5.6 km hourj1 with
`20- and 29-kg loads and 4.8 km hourj1 with a 38-kg
`load), power generation on the incline was the same
`as on the flat. Under low-power conditions (4.8 km
`hourj1 with 20- and 28-kg loads), electricity gener-
`ation on the incline was actually substantially greater
`than that on the flat (table S1).
`14. R. Margaria, Biomechanics and Energetics of Muscular
`Exercise (Clarendon, Oxford, 1976).
`
`15. R. A. Ferguson et al., J. Physiol. 536, 261 (2001).
`16. G. A. Cavagna, P. A. Willems, M. A. Legramandi, N. C.
`Heglund, J. Exp. Biol. 205, 3413 (2002).
`17. A. Grabowski, C. T. Farley, R. Kram, J. Appl. Physiol.
`98, 579 (2005).
`18. J. M. Donelan, R. Kram, A. D. Kuo, J. Exp. Biol. 205,
`3717 (2002).
`19. J. M. Donelan, R. Kram, A. D. Kuo, J. Biomech. 35, 117
`(2002).
`20. J. S. Gottschall, R. Kram, J. Appl. Physiol. 94, 1766 (2003).
`21. Because this savings in metabolic energy represents
`only 6% of the net energetic cost of walking with the
`backpack (492 W) (table S3) (17, 18), accurate de-
`terminations of the position and movements of the
`center of mass, as well as the direction and magnitude
`of the ground reaction forces, are essential to discern
`the mechanism. This will require twin–force-platform
`single-leg measurements, as well as a complete kine-
`matics and mechanical energy analysis (19, 20). The
`energy analysis is made more complex because the
`position of the load with respect to the backpack
`frame and the amount of energy stored in the back-
`pack springs vary during the gait cycle. Finally, elec-
`tromyogram measurements are also important to
`test whether a change in effective muscle moment
`arms may have caused a change in the volume of
`activated muscle and hence a change in metabolic
`cost (20, 27, 28).
`22. K. Schmidt-Nielsen, Animal Physiology: Adaptation
`and Environment (Cambridge Univ. Press, Cambridge,
`ed. 3, 1988).
`23. This assumes that electronic devices are being pow-
`ered in real time. If there were a power loss of 50%
`associated with storage (such as in batteries) and re-
`
`covery of electrical energy, then these factors would
`be halved.
`24. When not walking, the rack can be disengaged and
`the generator cranked by hand or by foot. Electrical
`powers of È3 W are achievable by hand, and higher
`wattage can be achieved by using the leg to power it.
`25. R. Kram, J. Appl. Physiol. 71, 1119 (1991).
`26. A. E. Minetti, J. Exp. Biol. 207, 1265 (2004).
`27. A. A. Biewener, C. T. Farley, T. J. Roberts, M. Temaner,
`J. Appl. Physiol. 97, 2266 (2004).
`28. T. M. Griffin, T. J. Roberts, R. Kram, J. Appl. Physiol.
`95, 172 (2003).
`29. This work was supported by NIH grants AR46125 and
`AR38404. Some aspects of the project were supported
`by Office of Naval Research grant N000140310568
`and a grant from the University of Pennsylvania Re-
`search Foundation. The authors thank Q. Zhang, H.
`Hofmann, W. Megill, and A. Dunham for helpful dis-
`cussions; R. Sprague, E. Maxwell, R. Essner, L. Gazit, M.
`Yuhas, and J. Milligan for helping with the experimen-
`tation; and F. Letterio for machining the backpacks.
`
`Supporting Online Material
`www.sciencemag.org/cgi/content/full/309/5741/1725/
`DC1
`Materials and Methods
`SOM Text
`Figs. S1 and S2
`Tables S1 to S4
`References
`
`14 February 2005; accepted 25 July 2005
`10.1126/science.1111063
`
`Accurate Multiplex Polony
`Sequencing of an Evolved
`Bacterial Genome
`Jay Shendure,1*. Gregory J. Porreca,1*. Nikos B. Reppas,1
`Xiaoxia Lin,1 John P. McCutcheon,2,3 Abraham M. Rosenbaum,1
`Michael D. Wang,1 Kun Zhang,1 Robi D. Mitra,2 George M. Church1
`
`We describe a DNA sequencing technology in which a commonly available,
`inexpensive epifluorescence microscope is converted to rapid nonelectrophoretic
`DNA sequencing automation. We apply this technology to resequence an evolved
`strain of Escherichia coli at less than one error per million consensus bases. A
`cell-free, mate-paired library provided single DNA molecules that were amplified
`in parallel to 1-micrometer beads by emulsion polymerase chain reaction.
`Millions of beads were immobilized in a polyacrylamide gel and subjected to
`automated cycles of sequencing by ligation and four-color imaging. Cost per
`base was roughly one-ninth as much as that of conventional sequencing. Our
`protocols were implemented with off-the-shelf instrumentation and reagents.
`
`The ubiquity and longevity of Sanger sequenc-
`ing (1) are remarkable. Analogous to semicon-
`ductors, measures of cost and production have
`followed exponential trends (2). High-throughput
`centers generate data at a speed of 20 raw bases
`per instrument-second and a cost of $1.00 per
`raw kilobase. Nonetheless, optimizations of elec-
`
`1Department of Genetics, Harvard Medical School,
`Boston, MA 02115, USA. 2Department of Genetics,
`3Howard Hughes Medical Institute, Washington Uni-
`versity, St. Louis, MO 63110, USA.
`
`*These authors contributed equally to this work.
`.To whom correspondence should be addressed.
`E-mail: shendure@alumni.princeton.edu (J.S.),
`gregory_porreca@student.hms.harvard.edu (G.J.P.)
`
`trophoretic methods may be reaching their lim-
`its. Meeting the challenge of the $1000 human
`genome requires a paradigm shift in our under-
`lying approach to the DNA polymer (3).
`Cyclic array methods, an attractive class
`B
`[
`of alternative technologies, are
`multiplex
`in
`that they leverage a single reagent volume to
`enzymatically manipulate thousands to mil-
`lions of immobilized DNA features in paral-
`lel. Reads are built up over successive cycles
`of imaging-based data acquisition. Beyond
`this common thread, these technologies di-
`versify in a panoply of ways: single-molecule
`versus multimolecule features, ordered versus
`disordered arrays, sequencing biochemistry,
`
`scale of miniaturization, etc. (3). Innovative
`proof-of-concept experiments have been re-
`ported, but are generally limited in terms of
`throughput, feature density, and library com-
`plexity (4–9). A range of practical and tech-
`nical hurdles separate these test systems from
`competing with conventional sequencing on
`genomic-scale applications.
`Our approach to developing a more mature
`alternative was guided by several consider-
`ations. (i) An integrated sequencing pipeline
`includes library construction, template ampli-
`fication, and DNA sequencing. We therefore
`sought compatible protocols that multiplexed
`each step to an equivalent order of magnitude.
`(ii) As more genomes are sequenced de novo,
`demand will likely shift toward genomic rese-
`quencing; e.g., to look at variation between in-
`dividuals. For resequencing, consensus accuracy
`increases in importance relative to read length
`because a read need only be long enough to
`correctly position it on a reference genome.
`However, a consensus accuracy of 99.99%, i.e.,
`the Bermuda standard, would still result in hun-
`dreds of errors in a microbial genome and hun-
`dreds of thousands of errors in a mammalian
`genome. To avoid unacceptable numbers of
`false-positives, a consensus error rate of 1 
`10j6 is a more reasonable standard for which
`to aim. (iii) We sought to develop sequencing
`chemistries compatible with conventional epi-
`fluorescence imaging. Diffraction-limited optics
`with charge-coupled device detection achieves
`an excellent balance because it not only pro-
`vides submicrometer resolution and high sen-
`sitivity for rapid data acquisition, but is also
`inexpensive and easily implemented.
`
`1728
`
`9 SEPTEMBER 2005 VOL 309 SCIENCE www.sciencemag.org
`
`Ariosa Exhibit 1040, pg. 1
`IPR2013-00277
`
`

`
` on September 11, 2012
`
`www.sciencemag.org
`
`Downloaded from
`
`situ polonies (6) was easily applied to ePCR
`beads, resulting in a È1.5-cm2 array of dis-
`ordered, monolayered, immobilized beads (Note
`S4, Fig. 2A).
`With few exceptions (18), sequencing bio-
`chemistries rely on the discriminatory capaci-
`ties of polymerases and ligases (1, 6, 8, 19–22).
`We evaluated a variety of sequencing protocols
`in our system. A four-color sequencing by
`B
`[
`ligation scheme (
`degenerate ligation
`) yielded
`the most promising results (Fig. 2, B and C). A
`detailed graphical description of this method is
`shown in fig. S7. We begin by hybridizing an
`B
`[
`anchor primer
`to one of four positions
`(immediately 5¶ or 3¶ to one of the two tags).
`We then perform an enyzmatic ligation reaction
`of the anchor primer to a population of degen-
`erate nonamers that are labeled with fluorescent
`dyes. At any given cycle, the population of
`nonamers that is used is structured such that the
`identity of one of its positions is correlated with
`the identity of the fluorophore attached to that
`nonamer. To the extent that the ligase discrim-
`inates for complementarity at that queried po-
`sition, the fluorescent signal allows us to infer
`
`R E P O R T S
`
`the identity of that base (Fig. 2, B and C). After
`performing the ligation and four-color imaging,
`the anchor primer:nonamer complexes are
`stripped and a new cycle is begun. With T4
`DNA ligase, we can obtain accurate sequence
`when the query position is as far as six bases
`from the ligation junction while ligating in
`the 5¶Y3¶ direction, and seven bases from
`
`
`the ligation junction in the 3¶Y5¶ direction.
`This allows us to access 13 bp per tag (a
`hexamer and heptamer separated by a 4- to
`5-bp gap) and 26 bp per amplicon (2 tags 
`13 bp) (fig. S7).
`Although the sequencing method presented
`here can be performed manually, we benefited
`from fully automating the procedure (fig. S3).
`Our integrated liquid-handling and microscopy
`setup can be replicated with off-the-shelf com-
`ponents at a cost of about $140,000. A detailed
`description of instrumentation and software is
`provided in Notes S5 and S7.
`As a genomic-scale challenge, we sought a
`microbial genome that was expected, relative to
`a reference sequence, to contain a modest num-
`ber of both expected and unexpected differences.
`
`Fig. 1. A multiplex approach to genome sequencing. (A) Sheared, size-selected genomic fragments
`(yellow) are circularized with a linker (red) bearing Mme I recognition sites (Note S1). Subsequent
`steps, which include a rolling circle amplification, yield the 134- to 136-bp mate-paired library
`molecules shown at right. (B) ePCR (14) yields clonal template amplification on 1-mm beads (Note
`S2). (C) Hybridization to nonmagnetic, low-density ‘‘capture beads’’ (dark blue) permits enrichment
`of the amplified fraction (red) of magnetic ePCR beads by centrifugation (Note S3). Beads are
`immobilized and mounted in a flowcell for automated sequencing (Note S4). (D) At each sequencing
`cycle, four-color imaging is performed across several hundred raster positions to determine the
`sequence of each amplified bead at a specific position in one of the tags. The structure of each
`sequencing cycle is discussed in the text, Note S6, and fig. S7.
`
`Conventional shotgun libraries are con-
`structed by cloning fragmented genomic DNA
`of a defined size range into an Escherichia coli
`vector. Sequencing reads derived from opposite
`B
`[
`ends of each fragment are termed
`mate-pairs.
`To avoid bottlenecks imposed by E. coli
`transformation, we developed a multiplexed,
`cell-free library construction protocol. Our
`strategy (Fig. 1A) uses a type IIs restriction
`endonuclease to bring sequences separated on
`the genome by È1 kb into proximity. Each
`È135–base pair (bp) library molecule contains
`two mate-paired 17- to 18-bp tags of unique ge-
`nomic sequence, flanked and separated by uni-
`versal sequences that are complementary to
`amplification or sequencing primers used in
`subsequent steps. The in vitro protocol (Note
`S1) results in a library with a complexity of È1
`million unique, mate-paired species.
`Conventionally, template amplification has
`been performed by bacterial colonies that must
`be individually picked. Polymerase colony, or
`polony, technologies perform multiplex ampli-
`fication while maintaining spatial clustering of
`identical amplicons (10). These include in situ
`polonies (11), in situ rolling circle amplification
`(RCA) (12), bridge polymerase chain reaction
`(PCR) (13), picotiter PCR (9), and emulsion
`PCR (14). In emulsion PCR (ePCR), a water-
`in-oil emulsion permits millions of noninteract-
`ing amplifications within a milliliter-scale
`volume (15–17). Amplification products of in-
`dividual compartments are captured via in-
`clusion of 1-mm paramagnetic beads bearing
`one of the PCR primers (14 ). Any single bead
`bears thousands of single-stranded copies of the
`same PCR product, whereas different beads bear
`the products of different compartmentalized
`PCR reactions (Fig. 1B). The beads generated
`by ePCR have highly desirable characteristics:
`high signal density, geometric uniformity, strong
`feature separation, and a size that is small but
`still resolvable by inexpensive optics.
`Provided that the template molecules are
`sufficiently short (fig. S1), an optimized version
`of the ePCR protocol described by Dressman
`et al. (14) robustly and reproducibly amplifies
`our complex libraries (Note S2). In practice,
`ePCR yields empty, clonal, and nonclonal
`beads, which arise from emulsion compartments
`that initially have zero, one, or multiple template
`molecules, respectively. Increasing template
`concentration in an ePCR reaction boosts the
`fraction of amplified beads at the cost of greater
`nonclonality (14). To generate populations in
`which a high fraction of beads was both ampli-
`fied and clonal, we developed a hybridization-
`based in vitro enrichment method (Fig. 1C). The
`protocol is capable of a fivefold enrichment of
`amplified beads (Note S3).
`Iterative interrogation of ePCR beads (Fig.
`1D) requires immobilization in a format compat-
`ible with enzymatic manipulation and epifluo-
`rescence imaging. We found that a simple
`acrylamide-based gel system developed for in
`
`www.sciencemag.org
`
`SCIENCE VOL 309 9 SEPTEMBER 2005
`
`1729
`
`Ariosa Exhibit 1040, pg. 2
`IPR2013-00277
`
`

`
`were not found. Of the 1.6 million reads, we
`were able to confidently place È1.16 million
`(È72%) to specific locations on the reference
`genome, resulting in È30.1 million bases of
`resequencing data at a median raw accuracy
`of 99.7%. At this stage of the analysis, the
`data were combined with reads from a pre-
`vious instrument run that contributed an addi-
`tional È18.1 million bases of equivalent quality
`(Fig. 2D). In this latter experiment, È1.8 mil-
`lion reads were generated from È7.6 million
`objects (È24%), of which È0.8 million were
`confidently placed (È40%).
`
`High-confidence consensus calls were de-
`termined for 70.5% of the E. coli genome for
`which sufficient and consistent coverage was
`available (3,289,465 bp; generally positions
`with È4 or greater coverage). There were
`six positions within this set that did not agree
`with the reference sequence, and thus were
`targeted for confirmation by Sanger sequenc-
`ing. All six were correct, although in one case
`we detected the edge of an 8-bp deletion
`rather than a substitution (Table 2). Three of
`these six mutations represent heterogeneities
`in lambda Red or MG1655, or errors in the
`
` on September 11, 2012
`
`www.sciencemag.org
`
`Downloaded from
`
`Fig. 2. Raw data acquisition and base calling. (A) Brightfield images (area shown corresponds to
`0.01% of the total gel area) facilitate object segmentation by simple thresholding, allowing resolution
`even when multiple 1-mm beads are in contact. (B) False-color depiction of four fluorescence images
`acquired at this location from a single ligation cycle. A, gold; G, red; C, light blue; T, purple. (C) Four-
`color data from each cycle can be visualized in tetrahedral space, where each point represents a single
`bead, and the four clusters correspond to the four possible base calls. Shown is the sequencing data
`from position (j1) of the proximal tag of a complex E. coli–derived library. (D) Cumulative distribu-
`tion of raw error as a function of rank-ordered quality for two independent experiments (red tri-
`angles, 18.1-Mb run; blue squares, 30.1-Mb run). The x axis indicates percentile bins of beads, sorted
`on the basis of a confidence metric. The y axis (logarithmic scale) indicates the raw base-calling
`1  10j3
`1  10j2, Q30
`0
`0
`accuracy of each cumulative bin. Equivalent Phred scores are Q20
`0 j10[log10(raw per-base error)]}. Cumulative distribution of raw error with sequencing
`{Phred score
`by ligation cycles considered independently is shown in fig. S8.
`
`Table 1. Genome Coverage and SNC prediction. Bases with consistent consensus coverage were used to
`make mutation predictions. To assess power, the outcome of consensus calling for the mock SNC
`positions with various levels of coverage was determined. Data from two independent sets of mock
`SNCs are shown. ‘‘86 of 87,’’ for example, means that 87 of the 100 mock SNCs were present in the
`sequence that was covered with 1 or more reads, and 86 of these were called correctly.
`
`Coverage
`1 or greater
`2 or greater
`3 or greater
`4 or greater
`
`Percent of genome
`
`Correctly called mock substitutions
`
`91.4%
`
`83.3%
`
`74.9%
`
`66.9%
`
`86 of 87
`88 of 90
`78 of 78
`75 of 76
`67 of 67
`68 of 68
`58 of 58
`62 of 62
`
`R E P O R T S
`
`We selected a derivative of E. coli MG1655,
`engineered for deficiencies in tryptophan bio-
`synthesis and evolved for È200 generations
`under conditions of syntrophic symbiosis via
`coculture with a tyrosine biosynthesis–deficient
`strain (23). Specific phenotypes emerged during
`the laboratory evolution, leading to the expec-
`tation of genetic changes in addition to inten-
`tionally engineered differences.
`An in vitro mate-paired library was con-
`structed from genomic DNA derived from a
`j strain. To
`single clone of the evolved Trpv
`sequence this library, we performed successive
`instrument runs with progressively higher bead
`densities. In an experiment ultimately yielding
`30.1 Mb of sequence, 26 cycles of sequencing
`were performed on an array containing ampli-
`fied, enriched ePCR beads. At each cycle, data
`were acquired for four wavelengths at 20
`optical magnification by rastering across each
`of 516 fields of view on the array (Fig. 1D). A
`detailed description of the structure of each
`sequencing cycle is provided in Note S6. In
`total, 54,696 images (14 bit, 1000 1000)
`were collected. Cycle times averaged 135 min
`per base (È90 min for reactions and È45 min
`for imaging), for a total of È60 hours per
`instrument run.
`Image processing and base calling algo-
`rithms are detailed in Note S7. In brief, all
`images taken at a given raster position were
`aligned. Two additional image sets were ac-
`quired: brightfield images to robustly identify
`bead locations (Fig. 2A) and fluorescent primer
`images to identify amplified beads. Our algo-
`rithms detected 14 million objects within the
`set of brightfield images. On the basis of size,
`fluorescence, and overall signal coherence over
`the course of the sequencing run, we deter-
`mined 1.6 million to be well-amplified, clonal
`beads (È11%). For each cycle, mean inten-
`sities for amplified beads were extracted and
`normalized to a 4D unit vector (Fig. 2, B and
`C). The Euclidean distance of the unit vector
`for a given raw base call to the median cen-
`troid of the nearest cluster serves as a natural
`metric of the quality of that call.
`The reference genome consisted of the E.
`coli MG1655 genome (GenBank accession code
`U00096.2) appended with sequences corre-
`sponding to the cat gene and the lambda Red
`prophage, which had been engineered into the
`sequenced strain to replace the trp and bio
`operons, respectively. To systematically assess
`our power to detect single-base substitutions,
`we introduced a set of 100 random single-
`nucleotide changes into the reference sequence
`B
`[
`at randomly selected positions (
`mock SNCs
`)
`(Table 1).
`An algorithm was developed to place the
`discontinuous reads onto the reference sequence
`(Note S7). The matching criteria required the
`paired tags to be appropriately oriented and
`located within 700 to 1200 bp of one anoth-
`er, allowing for substitutions if exact matches
`
`1730
`
`9 SEPTEMBER 2005 VOL 309 SCIENCE www.sciencemag.org
`
`Ariosa Exhibit 1040, pg. 3
`IPR2013-00277
`
`

`
` on September 11, 2012
`
`www.sciencemag.org
`
`Downloaded from
`
`R E P O R T S
`
`ing. To detect genomic rearrangements, we
`mined the unplaced mate-pairs for consistent
`links between genomic regions that did not fall
`within the expected distance constraints. In
`addition to detecting the expected replacements
`of the trp and bio operons with cat and lambda
`Red prophage (Fig. 3D), we detected and con-
`firmed the absence of a 776-bp IS1 transposon
`(Fig. 3C), a previously described heterogeneity
`in MG1655 strains (24). We also detected and
`confirmed a È1.8-kb region that was heteroge-
`neously inverted in the genomic DNA used to
`construct the library (Fig. 3E), owing to activity
`of pin on the invertible P region (25).
`We observe error rates of È0.001 for the
`better half of our raw base calls (Fig. 2D). Al-
`though high consensus accuracies are still
`achieved with relatively low coverage, our best
`raw accuracies are notably one to two orders of
`magnitude less accurate than most raw bases in a
`conventional Sanger sequencing trace. The PCR
`amplifications before sequencing are potentially
`introducing errors at a rate that imposes a ceiling
`on the accuracies achievable by the sequencing
`method itself. One potential solution is to create
`a library directly from the genomic material to
`be sequenced, such that the library molecules are
`linear RCA amplicons. Such concatemers, where
`each copy is independently derived from the
`original template, would theoretically provide a
`form of error correction during ePCR.
`Our algorithms were focused on detection of
`point substitutions and rearrangements. Increas-
`ing read lengths, currently totaling only 26 bp
`per amplicon, will be critical to detecting a
`wider spectrum of mutation. A higher fidel-
`ity ligase (20) or sequential nonamer ligations
`(20, 21) may enable completion of each 17- to
`18-bp tag. Eco P15 I, which generates È27-bp
`tags, would allow even longer read lengths while
`retaining the same mate-pairing scheme (26).
`We estimate a cost of $0.11 per raw kilo-
`base of sequence generated (Note S8), roughly
`one-ninth as much as the best costs for elec-
`trophoretic sequencing. Raw data in all se-
`quencing methods are generally combined to
`form a consensus. Even though costs are
`generally defined in terms of raw bases, the
`critical metric to compare technologies is con-
`sensus accuracy for a given cost. There is thus
`a need to devise appropriate cost metrics for
`specific levels of consensus accuracy.
`If library construction costs are not in-
`cluded, the estimated cost drops to $0.08 per
`raw kilobase. Higher densities of amplified
`beads are expected to boost the number of bases
`sequenced per experiment. While imaging, data
`were collected at a rate of È400 bp/s. Although
`enzymatic steps slowed our overall through-
`put to È140 bp/s, a dual flowcell instrument
`(such that the microscope is always imaging)
`will allow us to achieve continuous data ac-
`quisition. Enzymatic reagents, which dominate
`our cost equation, can be produced in-house at
`a fraction of the commercial price.
`
`reference sequence; three were only present
`in the evolved variant (Table 2). Of the 100
`mock SNCs, 53 were at positions called with
`high confidence. All of these were correctly
`called as substitutions of the expected nucleo-
`tide (59 of 59 on a second set of mock SNCs).
`The absence of substitution errors in È3.3 Mb
`of reference sequence positions called with
`high confidence suggests that we are achieving
`consensus accuracies sufficient for resequenc-
`ing applications. Percentage of the genome
`covered and mock SNC discovery at various
`levels of coverage are shown in Table 1.
`Despite 10 coverage in terms of raw base
`pairs, only È91.4% of the genome had at least
`
`1 coverage (fig. S4). Substantial fluctuations
`in coverage were observed owing to the sto-
`chasticity of the RCA step of library construc-
`tion. We are currently generating libraries that
`are more complex and more evenly distributed.
`A Gaussoid distribution of distances between
`mate-paired tags was observed, consistent with
`the size selection during library construction
`(Fig. 3, A and B). Notably, the helical pitch of
`DNA (È10.6 bp per turn) is evident in the local
`statistics of È1 million circularization events
`(Fig. 3B). As a function of the number of bases
`sequenced, we generated over an order of
`magnitude more mate-pairing data points than
`an equivalent amount of conventional sequenc-
`
`Fig. 3. Mate-paired tags and rearrangement discovery. (A) Diagnostic 6% polyacrylamide gel of the
`sheared, size-selected genomic DNA from which the library was constructed. Lanes 1 and 4 are molecular
`size markers. Lane 2 represents the material used in the library sequenced to generate the paired-tag
`mappings in (B), and lane 3 represents genomic DNA for a different library. (B) Histogram of distances
`between È1 million mapped mate-pair sequences. The probability of circularization favors integrals of
`the helical pitch of DNA, such that the Fourier transform of the distribution (inset) yields a peak at 10.6
`bp (27) (C to E). Consistent, aberrant mapping of unplaced mate-pairs to distal sequences revealed
`information about underlying rearrangements. Top and bottom blue bars indicate genomic positions for
`proximal and distal tags, respectively. Green connections indicate mate-pairings that fall within expected
`distance constraints, whereas red and black connections indicate aberrant connections (red indicates
`connections between the same strand, and black, connections between opposite strands). (C) Detection
`of a 776-bp deletion in the flhD promoter (24). (D) Detection of the replacement of the bio locus with
`the lambda red construct. (E) Detection of the P-region inversion (25). Detection of the inversion on a
`background of normally mate-paired reads indicates that the inversion is heterogeneously present.
`
`Table 2. Polymorphism discovery. Predictions for mutated positions were tested and verified as correct
`by Sanger sequencing. We found three mutations unique to the evolved strain—two in ompF, a porin,
`and one in lrp, a global regulator.
`
`Position
`
`Type
`
`Gene
`
`Context
`
`Confirmation
`
`Comments
`
`Y
`
`G
`T
`986,328
`8-bp deletion
`931,955
`Y
`T
`G
`985,791
`1,976,527j1,977,302 776-bp deletion
`Y
`3,957,960
`C
`T
`Y
`l-red, 3274
`T
`C
`Y
`l-red, 9846
`T
`C
`
`ompF j10 region
`lrp
`Frameshift
`Y
`ompF Glu
`Ala
`flhD
`Promoter
`5¶ UTR
`ppiC
`Y
`ORF61 Lys
`Y
`cI
`Glu
`
`Gly
`Glu
`
`Yes
`Yes
`Yes
`Yes
`Yes
`Yes
`Yes
`
`Evolved strain only
`Evolved strain only
`Evolved strain only
`MG1655 heterogeneity
`MG1655 heterogeneity
`l-red heterogeneity
`l-red heterogeneity
`
`www.sciencemag.org
`
`SCIENCE VOL 309 9 SEPTEMBER 2005
`
`1731
`
`Ariosa Exhibit 1040, pg. 4
`IPR2013-00277
`
`

`
` on September 11, 2012
`
`www.sciencemag.org
`
`Downloaded from
`
`21. S. C. Macevicz, U.S. Patent 5,750,341 (1998).
`22. S. Brenner et al., Nat. Biotechnol. 18, 630 (2000).
`23. N. B. Reppas, X. Lin, in preparation.
`24. C. S. Barker, B. M. Pruss, P. Matsumura, J. Bacteriol.
`186, 7529 (2004).
`25. R. H. Plasterk, P. van de Putte, EMBO J. 4, 237 (1985).
`26. M. Mucke, S. Reich, E. Moncke-Buchner, M. Reuter, D. H.
`Kruger, J. Mol. Biol. 312, 687 (2001).
`27. D. Shore, R. L. Baldwin, J. Mol. Biol. 170, 957 (1983).
`28. For advice, encouragement, and technical assistance,
`we are deeply indebted to J. Zhu, S. Douglas, J. Chou,
`J. Aach, M. Nikku, A. Lee, N. Novikov, and M. Wright
`(Church Lab); A. Blanchard, G. Costa, H. Ebling, J.
`Ichikawa, J. Malek, P. McEwan, K. McKernan, A. Sheridan,
`and D. Smith (Agencourt); S. Skiena (SUNY–Stony
`Brook) C. Felts (RPI); R. Fincher (Alcott); D. Focht
`(Bioptechs); and M. Hotfelder and J. Feng (Washing-
`ton University). We thank B. Vogelstein, J. Edwards,
`and their groups for assistance with emulsion PCR.
`This work was supported by the National Human
`Genome Research Institute–Centers of Excellence in
`Genomic Science and U.S. Department of Energy–
`Genomes to Life grants.
`
`Supporting Online Material
`www.sciencemag.org/cgi/content/full/1117389/DC1
`SOM Text
`Figs. S1 to S8
`
`14 July 2005; accepted 27 July 2005
`Published online 4 August 2005;
`10.1126/science.1117389
`Include this information when citing this paper.
`
`(2–7). In parallel, p53 also accumulates in the
`cytoplasm, where it directly activates the pro-
`apoptotic protein BAX to promote mitochondri-
`al outer-membrane permeabilization (MOMP)
`(8–10). Once MOMP occurs, proapoptogenic
`factors (for example, cytochrome c) are released
`from mitochondria, caspases are activated, and
`apoptosis rapidly ensues (11). Thus, p53 pos-
`sesses a proapoptotic function that is indepen-
`dent of its transcriptional activity (12–15).
`If p53 directly engages MOMP in coop-
`eration with BAX, no further requirement for
`p53-dependent
`transcriptional regulation of
`additional proapoptotic Bcl-2 proteins would
`be expected. Nevertheless, PUMA (p53–up-
`regulated modifier of apoptosis), a proapoptotic
`BH3-only protein, is a direct transcriptional
`target of p53. Furthermore, mice deficient in
`Puma are resistant to p53-dependent, DNA
`damage–induced apoptosis even though p53
`is stabilized and accumulates in the cyto-
`plasm (6, 16–18). A better understanding of
`the distinct nuclear and cytoplasmic proapo-
`ptotic functions of p53 may reveal strategies
`for the prevention and treatment of cancer.
`
`R E P O R T S
`
`We demonstrate low costs of sequencing,
`mate-paired reads, high multiplicities, and high
`consensus accuracies. These enable applications
`including BAC (bacterial artificial chromosome)
`and bacterial genome resequencing, as well as
`SAGE (serial analysis of gene expression) tag
`and barcode sequencing. Simulations suggest
`that the current mate-paired libraries are com-
`patible with human genome resequencing, pro-
`vided that the read length can be increased to
`cover the full 17- to 18-bp tag (fig. S5).
`What are the limits of this approach? As
`many as 1 billion 1-mm beads can potentially be
`fit in the area of a standard microscope slide
`(fig. S6). We achieve raw data acquisition rates
`of È400 bp/s, more than an order of magnitude
`faster than conventional sequencing. From anoth-
`er point of view, we collected È786 gigabits of
`image data from which we gleaned only È60
`megabits of sequence. This sparsity—one useful
`bit of information per 10,000 bits collected—is a
`ripe avenue for improvement. The natural limit
`of this direction is single-pixel sequencing,
`in which the commonplace analogy between
`bytes and bases will be at its most manifest.
`
`References and Notes
`1. F. Sanger et al., Nature 265, 687 (1977).
`2. F. S. Collins, M. Morgan, A. Patrinos, Science 300, 286
`(2003).
`3. J. Shendure et al., Nat. Rev. Genet. 5, 335 (2004).
`4. I. Braslavsky, B. Hebert, E. Kartalov, S. R. Quake, Proc.
`Natl. Acad. Sci. U.S.A. 100, 3960 (2003).
`5. T. S. Seo et al., Proc. Natl. Acad. Sci. U.S.A. 102,
`5926 (2005).
`6. R. D. Mitra, J. Shendure, J. Olejnik, O. Edyta Krzymanska,
`G. M. Church, Anal. Biochem. 320, 55 (2003).
`7. M. J. Levene et al., Science 299, 682 (2003).
`8. M. Ronaghi, S. Karamohamed, B. Pettersson, M. Uhlen,
`P. Nyren, Anal. Biochem. 242, 84 (1996).
`9. J. H. Leamon et al., Electrophoresis 24, 3769 (2003).
`10. http://arep.med.harvard.edu/Polonator/Plone.htm
`11. R. D. Mitra, G. M. Church, Nucleic Acids Res. 27, e34
`(1999).
`12. P. M. Lizardi et al., Nat. Genet. 19, 225 (1998).
`13. C. P. Adams, S. J. Kron, U.S. Patent 5,641,658 (1997).
`14. D. Dressman, H. Yan, G. Traverso, K. W. Kinzler, B.
`Vogelstein, Proc. Natl. Acad. Sci. U.S.A. 100, 8817
`(2003).
`15. D. S. Tawfik, A. D. Griffiths, Nat. Biotechnol. 16, 652
`(1998).
`16. F. J. Ghadessy, J. L. Ong, P. Holliger, Proc. N