Several genome centres have established very effective "pipelines" for the sequencing of entire genomes, often using a mixed strategy drawing data from mapped clones and whole genome shotgun efforts 1. However, such pipelines are rapidly evolving through the incorporation of new, high-throughput sequencing platforms, that obviate conventional clone libraries, or Sanger sequencing chemistries 23, but yield sequence reads of limited length. As such, current and emerging sequencing strategies benefit from the availability of compatible physical maps guiding clone selection, or facilitating assemblies that may also enhance validation efforts 134. Furthermore, physical maps become a critical necessity when spanning repeat-rich genomic regions, typified by telomeric or centromeric portions of chromosomes. Although clone fingerprint 5 or end-sequence maps 6 are widely used in whole-genome squencing efforts, gaps and sequence assembly errors may persist stemming from "clone drop-outs," or uncertainties in the map assembly process caused by the presence of repeated sequence elements. Clearly, physical maps will continue to be an important feature of large-scale sequencing projects, but new mapping approaches must advance in ways that effectively deal with the trend towards obviation of traditional libraries and the abundance of modestly sized reads. The main issues will likely center on the validation of "strategically" unfinished genome sequences and comprehensive description of genome structure. These issues become acute when sequenced genomes are selected from nascently described organisms lacking genetic resources or an associated scientific community. Consequently, future comparative studies could suffer from sequencing errors, and not fully discern structural variation – a major feature of genome evolution and a source of disease genotypes.

The rice (Oryza sativa) genome was originally chosen for sequencing because it is a staple food crop for more than half of the world's population, in addition to its many genetic attributes, or resources that include: a compact genome (~400 Mb), well-defined genetic maps, Yeast Artificial Chromosome (YAC) and Bacterial Artificial Chromosome (BAC)/P1-derived Artificial Chromosome (PAC) map resources, comprehensive sequence-tagged or transcript maps, and efficient genetic transformation techniques 7891011121314151617181920. Rice also shares extensive syntenic relationships with other cereal plants bearing huge genome sizes such as maize (~2,500 Mb), barley (~4,900 Mb), and wheat (~15,000 Mb) 212223242526. The large-scale sequencing of rice (O. sativa ssp. japonica cv Nipponbare) was initiated in 1998 under the auspices of the International Rice Genome Sequencing Project (IRGSP), with joint efforts from Japan, the United Sates, China, Brazil, Great Britain, France, India, Korea, and Thailand 27. IRGSP members decided at the time to pursue the collaboration-friendly, clone-by-clone, or BAC/PAC-by-BAC/PAC strategy supported by extensive map resources. For example, BAC/PAC draft sequences or contigs were anchored and oriented on the rice genetic maps – these contigs were further augmented by BAC-end sequencing (via Sequence-Tagged Connectors) and contig-end walking. BAC maps and fibre Fluorescent in situ Hybridization (FISH) were also used for characterization of gaps present within low recombination regions or genomic portions showing modest BAC/PAC coverage 17192829. The IRGSP release of the rice genome is now finished with publication of the analysis and annotation of these data 29303132333435. Here, the IRGSP sequence for each chromosome – Build 4.0, released in August, 2005 – is represented as a "pseudomolecule," or a virtual contig. Each pseudomolecule is constructed by joining PAC/BAC sequences according to their order determined by comparison with a previously constructed physical map 30. Finishing steps include identification and removal of overlapping sequences with resulting physical gaps replaced by a variable number of successive "N's", reflecting their estimated breadth. There are 62 physical gaps including 17 telomeric gaps, and 9 centromeric gaps with a total size of ~18.1 Mb 30, with one gap closed in chromosome 1 and some of them partially filled 36 within the current build.

In a parallel effort, TIGR (The Institute for Genomic Research) also constructed similar pseudomolecules for each of 12 rice chromosomes 37 to enlarge the span of regions comprising blocks of contiguous sequence – their approach included: the resolution of discrepancies between overlapping BAC/PAC clones, trimming of overlap regions, and linking of unique sequences. These efforts relied on 3,450 rice BAC/PAC clone sequences obtained from the IRGSP; of these, 3,408 BAC/PAC clones (98.8%) were finished, and 42 BAC/PAC (1.2%) clones were unfinished (phase 2), as defined by Genbank. As such, the data show many gaps between clones, i.e., physical gaps, denoted by "1000N's" in the final pseudomolecules. Finally, centromeres were identified using the "CentO" centromeric sequence 38. There are 48 physical gaps within the 12 pseudomolecules including gaps at 10 centromeres.

In addition to the rice genome sequence from the IRGSP and TIGR, the draft rice genome sequence of the same cultivar Nipponbare was generated by two separate private sources: Pharmacia, Inc. (previously part of Monsanto, Inc., Peapack, NJ) and Syngenta, Inc. (San Diego, CA) 39. A 259 Mb draft sequence from Pharmacia was also generated by a clone-by-clone based strategy 40, while the 390 Mb Syngenta rice genome draft sequence consisting of 42,109 sequence contigs was obtained using a whole-genome shotgun sequencing approach with an estimation of 32,000 to 50,000 genes for this cultivar. A draft sequence (361 Mb out of the estimated genome size of 466 Mb) of the O. sativa L. ssp. indica cultivar (93-11) was also obtained by the Beijing Genomics Institute (BGI) 41 using a whole-genome shotgun sequencing approach with an estimated gene count of 46,022 to 55,615 for this subspecies. Using TIGR's pseudomolecules and publicly available rice EST sequence data, Affymetrix has recently constructed a rice gene expression array with 46,115 rice gene models (Affymetrix, personal communication). In this regard, amongst sequenced genomes rice comprises the largest number of predicted genes, with more genes than human, and almost double that of Arabidopsis thaliana.

Although the current releases of rice sequence from IRGSP, and TIGR are of very high quality, difficult gaps remain to be spanned for each of the 12 chromosomal pseudomolecules. These gaps persist because some reside within genomic regions showing sparse coverage of genetic markers used for anchoring BAC/PAC clones, and others suffer from library construction, which may bias against heavily repetitive regions. Importantly, existing gaps probably contain many functional genes 36, even within centromeric regions 42, in addition to information describing chromosomal structure. Furthermore, sequence contigs, within the pseudomeolecules, may still contain errors, in part, arising from assemblies conducted in repeat-rich regions of the rice genome. Given these issues, we constructed a genome wide optical map for determination of the size of sequence gaps and for identification of problematic sequence assemblies (discordances) through the analysis of sequence build alignments against our optical restriction map.

Optical mapping is now a robust, automated system for the construction of whole genome ordered restriction maps from ensembles of individual genomic DNA molecules 43444546474849505152535455. Library construction, PCR amplification, hybridization, and their attending artefacts are obviated in optical mapping, since genomic DNA is the analyte and restriction enzymes are used to generate reliable markers. Therefore, high-resolution physical maps are created on a whole genome basis presenting an organism's genetic constitution in a form directly linkable to sequence data. Using a newly automated version of optical mapping system, we have constructed a whole genome physical map for the rice genome using the optical mapping system, employing schemes akin to whole genome shotgun sequencing approaches. Given the results presented here, we find that there is no practical limit to the optical mapping of large, complex genomes, even in the absence of any sequence information, since the level of automation we have achieved provides ample data sets for our assembly techniques to span entire genomes. However, accessible sequence data for any genome provides direct links to a multitude of annotation and analysis tools that are especially facilitated when genomes are chosen to be strategically sequenced, such as maize.

More specifically, we present the construction of a whole-genome shotgun optical restriction map, using SwaI, of the rice (Oryza sativa) genome, and its comparison to sequence builds. Because physical distances (in kb) between restriction sites are accurately determined by optical mapping, alignments of an "in silico" restriction map, constructed from sequence data, against our optical map reveal discordances characterized by features commonly associated with ordered restriction maps. They include: missing or extra restriction "cuts," missing or extra restriction fragments, and significant alterations of restriction fragment sizes or patterns. Also, large-scale discordances covering hundreds of kilobases are discoverable and described here.

As such, we show that a high-resolution physical map based on the direct analysis of genomic DNA, spans existing sequence physical gaps, validates the genome sequence assembly, characterizes gaps, corrects sequence misassemblies, and creates a physical scaffold for sequence finishing. We expect that this map will also secure a resource for the genome sequencing communities at-large in their investigation of rice subspecies and cultivars. We also think that the maps presented here will facilitate the final validation of the rice sequence data, which should strengthen the important role that this genome is already playing as an accessible model system for other plants and cereal crops.

We have constructed a whole genome SwaI restriction map for the rice genome (O. sativa ssp. japonica cv. Nipponbare) using a modified whole genome shotgun optical mapping approach that was used to identify problematic regions within the current sequence build. The whole genome optical map consists of 14 optical map contigs, of which 9 are finished chromosome optical map contigs. Among the remaining 3 chromosomes, chromosomes 6 and 11 each composed of two optical map contigs with gaps at the centromeric regions, and 9 only has one optical map contig with gap at the nucleolar organizer ribosomal repeat and telomeric regions. Maps, by definition, do not posses the same resolution as sequence, so that map alignments to sequence reveal discordances governed by experimental factors that hinge on the average restriction fragment size, or "resolution" of the final map, which is limited by reliable detection of small restriction fragments, sizing errors, and the extent of genome coverage 55. Alignments between the optical and in silico maps, derived from IRGSP and TIGR sequence pseudomolecules or virtual contigs, enabled us to fully place such sequence data along map scaffolds. These alignments revealed a high degree of concordance (Table 2, Table 3, Fig. 3) and provided a largely independent way to assess the errors in the rice genome sequence assembly. Perhaps more importantly, map-sequence alignments have characterized a number of gaps, which may be difficult to assess using other approaches.

The estimated genome size of rice genome is 382.80 Mb – after summing the masses of all the 12 estimated chromosome sizes estimated based on optical mapping (378.31 Mb), three optical map gaps estimated based on other studies 86465 (3.80 Mb), and all the missing small fragments (0.69 Mb) in the optical maps based on the map alignments between optical maps and the in silico maps of IRGSP sequence pseudomolecules. Comparisons with other genome or chromosome size estimations based on BAC/PAC or YAC physical maps, show that genome and chromosome size estimates based on optical mapping are mostly smaller, but were very close to the most recent estimation of genome size (388.82 Mb) based on the minimum tiles of BAC/PAC clone sequences for each chromosome 30. However, the size estimate for chromosome 9 is quite different from that estimated by optical mapping (3.97 Mb difference, or ~15.0%). The difference between our optical map-based chromosome size and the IRGSP sequence and map-based chromosome is mainly due to the different number of copies of rice rRNA genes used for the size estimation. IRGSP used 850 copies of rice rRNA genes to estimate the length of chromosome 9 short arm nucleolar organizer DNA, which is calculated to be 6.95 Mb, but in fact, this copy number is for a diploid genome 65. We used half of this number, which is 425 copies of rRNA genes to calculate the size of the ribosomal repeat region (3.48 Mb), because the genome sequence is haploid.

The chromosome size measurements and the arm ratios determined by pachytene FISH 61 and optical mapping are quite different. Here chromosome sizes were mostly overestimated except for chromosome 4, 9, and 10, and the arm ratios were overestimated for chromosome 4, 5, 7, 8, 9 and 10 (>10%), but were underestimated for chromosomes 6 (7.2%) and 11 (5.0%). Our results confirmed that the long and short arms of chromosome 3 should be reversed, but not for chromosome 6. The sizing discrepancies between pachytene FISH and optical mapping estimates for chromosome sizes and arm ratios reflect fundamental differences in how DNA intervals are measured and the number of markers used for analysis. In pachytene FISH, distances between markers is measured as an actual distance (microns or image pixels) using difficult to control chromatin substrates – varying degrees of condensation can affect the accuracy of distance measurements in a locus-specific way. For example, size estimate for the short arm of chromosome 4 was significantly underestimated by pachytene FISH probably due to the presence of heterochromatic or AT-rich sequences; while the sizes of chromosomes 1, 2, and 3 were significantly overestimated by pachytene FISH, because these chromosomes are mostly composed of euchromatic sequences 6162. These issues are largely obviated in optical mapping since the mapping substrate is just naked, fully deproteinized DNA molecules, and "distances" are robustly estimated by measurement of integrated fluorescence intensity 5363 using a dye (YOYO-1), whose measured fluorescence intensity is somewhat insensitive to base composition or extent of DNA elongation. Consequently, the chromosome sizes presented in this paper are likely to be more accurate than those determined by pachytene FISH.

Comparisons between the optical maps and the in silico maps from the IRGSP and TIGR sequence pseudomolecules showed that additional gaps exist in the sequence pseudomolecules than what has been reported. There are 62 gaps (including 9 centromeric and 17 telomeric gaps) recorded in the IRGSP sequence pseudomolecules which are represented by consecutive "N" with gap sizes defined 3031. Sixty of the 62 gaps were bridged by optical maps, and the two gaps not bridged were located at the centromeric regions of chromosome 6 and 11. There are additional 90 gaps (including 5 telomeric and 3 centromeric gaps) plus 53 misassemblies present in the IRGSP sequence pseuodomolecules detected by the comparative analysis between optical maps and the in silico maps of the IRGSP sequence pseudomolecules. These misassemblies usually contain SwaI restriction site differences (Table 2, Fig. 4), and some of them may bear gaps. In TIGR rice genome sequence pseudomolecules 37, the 48 physical gaps (including 10 centromeric gaps) were recorded, and 46 of these gaps were spanned by optical maps based on the map aligments with the two unfilled gaps also located at the centromeric regions of chromosome 6 and 11. Comparative analysis between optical maps and the in silico maps of the TIGR sequence pseuodomolecules showed that there are additional 117 gaps (including 23 telomeric and 11 centromeric gaps), plus 39 misassemblies existed in the TIGR rice genome sequence pseudomolecules. Overall, the TIGR sequence pseudomolecules show less discordance with our findings than the IRGSP sequence pseudomolecules as judged by comparison of in silico vs. optical mapping fragment masses and reduced prevalence of sequence misassemblies despite discovery of additional gaps within sequence assemblies. About 70% of these gaps and 50% of of misassemblies are common to both IRGSP and TIGR sequence pseudomolecules (Fig. 4). As alignments between the optical and in silico maps of the sequence pseudomolecules locate gaps and their span, or possible misassembled sequences, those BAC/PAC clones from such problematic regions can be flagged for further sequence analysis. Gaps can be closed using PCR or other molecular techniques, and misassembled sequences can be corrected based on the SwaI optical restriction maps.

The gap sizes estimated by optical mapping are likely to be more accurate than that estimated by genetic mapping or fibre FISH and pachytene FISH physical mapping 2933. The genome sequence and structure of rice chromosome 1 was reported in 2002 33 with 9 large sequence contigs and 8 gaps. Two gaps were closed and five gaps in the arm regions still remain in the current IRGSP build 4 pseudomolecules 31 The gap sizes previously estimated by genetic marker or fibre FISH and pachytene FISH are shown in Figure 6. Alignments between the in silico maps of the rice chromosome 1 sequence pseudomolecules and the optical maps can easily reveal the gap sizes between the contigs based on where the consecutive "Ns" located, and the gap sizes are also shown in Figure 6 based on the measurements of the optical maps. The centromeric and telomeric gaps, and the five internal chromosome arm gaps between the sequence contigs were measured to be 1,850 kb in total by genetic markers and fibre or pachytene FISH, while these gaps were measured by optical mapping were shown to be only 792 kb, which is less than one half of the estimation by genetic markers and fibre or pachytene FISH. In the TIGR sequence pseudomolecule release 4, chromosome 1 sequence pseudomolecule has six chromosome internal gaps, and five out of them are at the same locations as in IRGSP chromosome 1 sequence pseudomolecule based on the optical map (Fig. 6), however, there are 346 kb sequence inserted in gap 3 (Fig. 6), which does not belong to this region and appears to be combined from sequences of multiple chromosomes based on the optical map (data not shown). We have tried to use PCR to close the gap 4 and 5 based on IRGSP chromosome 1 sequence, and failed to generate expected unique amplicons probably due to sequence repeat elements. Other approaches are likely required in order to close these gaps.

The centromeric regions of chromosomes found in higher eukaryotes are a complex motif of repetitive sequences. Consequently, comprehensive basepair level knowledge of centromeric regions is at the frontier of genome sequencing technology presenting challenges to cloning, mapping and assembly 386667. Although rice centromeres are analyzable by optical mapping, mammalian centromeres, being larger and more complex, may challenge our ability to span regions bearing few restriction sites with ~500 kb molecules. As such, it is understandable that from previous studies that the centromeres of only chromosomes 4 and 8 have been fully mapped and sequenced. Here, we have fully mapped 10 complete centromeric regions, with the remaining chromosomes being partly characterized (Fig. 2 and Fig. 4). Accordingly, comparisons of optical map findings for the centromeric region of chromosome 8 against IRGSP and TIGR results showed that a ~24 kb gap still exists both in the IRGSP and TIGR pseudomolecules (Table 2, and Table 3), while chromosome 4 showed no discordances.

A whole-genome SwaI restriction map of rice genome was constructed and comparison between the in silico maps of rice chromosome pseudomolecule sequences revealed not only gap characteristics, but also putatively misassembled parts of the genome sequence. We think that this whole genome optical map will greatly facilitate gap closure efforts and guide correction of misassemblies to provide an accurate and completely sequenced model for plant genome research – especially for cereal genomes. As our optical mapping system has achieved a high-degree of automation, now we can produce a whole genome optical map like rice genome with only a fraction of the cost for sequencing. We envision optical mapping employed as a new platform for comparative genomics to study closely related varieties, or cultivars, that would leverage sequence and annotation information across a broad range of rice subspecies and varieties.

SZ contributed to the design of the study, carried out partial data collection, performed optical map assembly and comparison between optical map and sequence, and drafted the manuscript. MCB contributed to the data collection and manual curation of the discordance between optical map and sequence. MP contributed to data collection. CPC involved in data analysis. LP developed new protocols. SAL provided rice seed and assisted in DNA preparations. RR maintained and advanced the imaging system used for this study. SG assisted map assembly efforts and statistical analysis. DKF and ML involved in cluster computing. DCS conceived the study, contributed to experimental design and analysis, and coordinated final writing.