The human genome is an elegant but cryptic store of information. Its roughly three billion bases encode, either directly or indirectly, the instructions for synthesizing nearly all the molecules that form each human cell, tissue, and organ. Sequencing the human genome^1–3 provided highly accurate DNA sequences for each of the 24 chromosomes. At present, however, we have an incomplete understanding of the protein-coding portions of the genome, and markedly less understanding of both non-protein-coding transcripts and genomic elements that temporally and spatially regulate gene expression. To understand the human genome, and by extension the biological processes it orchestrates and the ways in which its defects can give rise to disease, we need a more transparent view of the information it encodes.

The molecular mechanisms by which genomic information directs the synthesis of different biomolecules has been the focus of much of molecular biology over the last three decades. Previous studies have typically concentrated on individual genes, with the resulting general principles then providing insights into, for example, transcription, chromatin remodeling, mRNA splicing, DNA replication and numerous other genomic processes. Although many such principles appear valid as additional genes are investigated, they typically have not provided genome-wide insights about biological function.

The first genome-wide analyses that shed light on human genome function made use of observing the actions of evolution. The ever-growing set of vertebrate genome sequences^4–8 is providing increasing power to reveal the genomic regions that have been most and least acted upon by the forces of evolution. However, while these studies convincingly indicate the presence of numerous genomic regions under strong evolutionary constraint, they have less power in identifying the precise bases that are constrained and provide little, if any, insight into why those bases are biologically important. Further, although we have good models for how protein-coding regions evolve, our present understanding about the evolution of other functional genomic regions is poorly developed. Experimental studies that augment what we learn from evolutionary analyses are key for solidifying our insights about genome function.

The Encyclopedia of DNA Elements (ENCODE) Project⁹ aims to provide a more biologically informative representation of the human genome by using high-throughput methods to identify and catalogue the functional elements it encodes. In its pilot phase, 35 groups provided more than 200 experimental and computational datasets that examined in unprecedented detail a targeted 29.998 Mb of the human genome. This roughly 30 Mb— ~1% of the human genome— is sufficiently large and diverse to allow for rigorous pilot testing of multiple experimental and computational methods. These 30 Mb are divided among 44 genomic regions; roughly 15 Mb reside in 14 regions for which there is already substantial biological knowledge, while the other roughly 15 Mb reside in 30 regions chosen by a stratified random-sampling method (see http://www.genome.gov/10506161).

The highlights of our findings to date include:

• The human genome is pervasively transcribed, such that the majority of its bases are associated with at least one primary transcript and many transcripts link distal regions to established protein-coding loci.
• Many novel non-protein-coding transcripts have been identified, with many of these overlapping protein-coding loci and others located in regions of the genome previously thought to be transcriptionally silent.
• Numerous previously unrecognised transcription start sites have been identified, many of which show chromatin structure and sequence-specific protein-binding properties similar to well-understood promoters.
• Regulatory sequences that surround transcription start sites are symmetrically distributed, with no bias towards upstream regions.
• Chromatin accessibility and histone-modification patterns are highly predictive of both the presence and activity of transcription start sites.
• Distal DNaseI hypersensitive sites have characteristic histone modification patterns that reliably distinguish them from promoters; some of these distal sites show marks consistent with insulator function.
• DNA-replication timing is correlated with chromatin structure.
• A total of 5% of the bases in the genome can be confidently identified as being under evolutionary constraint in mammals; for approximately 60% of these constrained bases, there is evidence of function based on the results of the experimental assays performed to date.
• While there is general overlap between genomic regions identified as functional by experimental assays and those under evolutionary constraint, not all bases within these experimentally-defined regions show evidence of constraint.
• Different functional elements vary greatly in their sequence variability across the human population and in their likelihood of residing within a structurally variable region of the genome.
• To our surprise, many functional elements are seemingly unconstrained across mammalian evolution. This suggests the possibility of a large pool of neutral elements that are biologically active but provide no specific benefit to the organism. This pool may serve as a ‘warehouse’ for natural selection, potentially acting as the source of lineage-specific elements and functionally conserved but non-orthologous elements between species.

Below, we first provide an overview of the experimental techniques used for our studies, after which we describe the insights gained from analyzing and integrating the generated datasets. We conclude with a perspective of what we have learned to date about this 1% of the human genome and what we believe the prospects are for a broader and deeper investigation of the functional elements in the human genome. To aid the reader, ^{Box 1} provides a glossary for many of the abbreviations used throughout this paper.

Table 1 (expanded in Supplementary Information section S1.1) lists the major experimental techniques used for the studies reported here, relevant acronyms, and references reporting the generated datasets. These datasets reflect over 400 million experimental data points (603 million data points if one includes comparative sequencing bases). In describing the major results and initial conclusions, we seek to distinguish biochemical function from biological role. Biochemical function reflects the direct behaviour of a molecule(s), while biological role is used to describe the consequence(s) of this function for the organism. Genome-analysis techniques nearly always focus on biochemical function but not necessarily on biological role. This is because the former is more amenable to large-scale data-generation methods, while the latter is more difficult to assay on a large scale.

Table 1 Summary of types of experimental techniques used in ENCODE

ENCODE aimed to establish redundancy with respect to the findings represented by different datasets. In some instances, this involved the intentional use of different assays based on a similar technique, whereas in other situations, different techniques assayed the same biochemical function. Such redundancy has allowed methods to be compared and the generation of consensus datasets, much of which is discussed in companion papers, such as the ChIP/chip platform comparison^{10, 11}. All ENCODE data have been released after verification but prior to this publication, as befits a ‘community resource’ project (http://www.wellcome.ac.uk/doc_wtd003208.html). Verification is defined as when the experiment is reproducibly confirmed (see Supplementary Information section S1.2). The main portal for ENCODE data is provided on the UCSC Genome Browser (http://genome.ucsc.edu/ENCODE/); this is augmented by multiple other web sites (see Supplementary Information section S1.1).

A common feature of genomic analyses is the need to assess the significance of the co-occurrence of features or other statistical tests. One confounding factor is the heterogeneity of the genome, which can produce uninteresting correlations of variables distributed across the genome. We have developed and used a statistical framework that mitigates many of these hidden correlations by adjusting the appropriate null distribution of the test statistics. We term this correction procedure “Genome Structure Correction” (GSC) (see Supplementary Information section S1.3).

In the next five sections, we detail the various biological insights of the pilot phase of the ENCODE Project.

These methods recapitulate previous findings, but provide enhanced resolution due to the larger number of tissues sampled and the integration of results across the three approaches. To begin with, our studies show that 14.7% of the bases represented in the unbiased tiling arrays are transcribed in at least one tissue sample. Consistent with previous work^{14, 15}, many (63%) TxFrags reside outside of GENCODE annotations, both in intronic (40.9%) and intergenic (22.6%) regions. GENCODE annotations are richer than the more-conservative RefSeq or Ensembl annotations, with 2,608 transcripts clustered into 487 loci, leading to an average of 5.4 transcripts per locus. Finally, extensive testing of predicted protein-coding sequences outside of GENCODE annotations was positive in only 2% of cases¹⁶, suggesting that GENCODE annotations cover nearly all protein-coding sequences. The GENCODE annotations are categorised both by likely function (mainly, the presence of an open reading frame) and by classification evidence (for example, transcripts based solely on ESTs are distinguished from other scenarios); This classification is not strongly correlated with expression levels (see Supplementary Information sections S2.4.2 and S2.4.3).

Analyses of more biological samples have allowed a richer description of the transcription specificity (see Figure 1 and Supplementary Information section S2.5). We found that 40% of TxFrags are present in only one sample, whereas only 2% are present in all samples. Although exon-containing TxFrags are more likely (74%) to be expressed in more than one sample, 45% of unannotated TxFrags are also expressed in multiple samples. GENCODE annotations of separate loci often (42%) overlap with respect to their genomic coordinates, in particular on opposite strands (33% of loci). Further analysis of GENCODE-annotated sequences with respect to the positions of open read frames revealed that some component exons do not have the expected synonymous vs non synonymous substitution patterns of protein-coding sequence (see Supplementary Information section S2.6) and some have deletions incompatible with protein structure¹⁸. Such exons are on average less expressed (25% vs 87% by RT-PCR, see Supplementary Information section S2.7) than exons involved in more than one transcript (see Supplementary Information section S2.4.3), but when expressed have a tissue distribution comparable to well-established genes.

Figure 1 Annotated and unannotated TxFrags detected in different cell lines. The proportion of different types of transcripts detected in the indicated number of cell lines (from 1/11 at the far left to 11/11 at the far right) is shown. The data for annotated (more ...)

Critical questions are raised by the presence of a large amount of unannotated transcription with respect to how the corresponding sequences are organised in the genome – do these reflect longer transcripts that include known loci, do they link known loci, or are they completely separate from known loci? We further investigated these issues using both computational and new experimental techniques.

Figure 2 Length of genomic extensions to GENCODE-annotated genes based on RACE experiments followed by array hybridisations (RxFrags). The indicated bars reflect the frequency of extension lengths among different length classes. The solid line shows the cumulative (more ...)

To further characterise the 5′ RxFrag extensions, we performed RT-PCR followed by cloning and sequencing for 550 of the 5′ RxFrags (including the 261 longest extension identified for each locus). The approach of mapping RACE products using microarrays is a combination method previously described and validated in several studies^{14, 17, 20}. Hybridization of the RT-PCR products to tiling arrays confirmed connectivity in almost 60% of the cases. Sequenced clones confirmed transcript extensions. Longer extensions were harder to clone and sequence, but 5 of 18 RT-PCR-positive extensions over 100-kb were verified by sequencing (see Supplementary Information section S2.9.7 and Denoeud et al¹⁷). The detection of numerous RxFrag extensions coupled with evidence of considerable intronic transcription indicates that protein-coding loci are more transcriptionally complex than previously thought. Instead of the traditional view that many genes have one or more alternative transcripts that code for alternative proteins, our data suggest that a given gene may both encode multiple protein products as well as produce other transcripts that include sequences from both strands and from neighbouring loci (often without encoding a different protein). Figure 3 illustrates such as case, where a new fusion transcript is expressed in the small intestine, and consists of at least three coding exons from the ATP50 gene and at least two coding exons from the DONSON gene, with no evidence of sequences from two intervening protein-coding genes (ITSN1 and CRYZL1).

Figure 3 Overview of RACE experiments showing a gene fusion. Transcripts emanating from the region between the DONSON and ATP50 genes. A 330-kb interval of human chromosome 21 (within ENm005) is shown, which contains four annotated genes: DONSON, CRYZL1, ITSN1 (more ...)

Many known ncRNAs are characterised by a well-defined RNA-secondary structure. We applied two de-novo ncRNA-prediction algorithms – EvoFold and RNAz – to predict structured ncRNAs (as well as functional structures in mRNAs) using the multi-species sequence alignments (see below, Supplementary Information section S2.11, and Washietl et al²⁶). Using a sensitivity threshold capable of detecting all known miRNAs and snoRNAs, we identified 4986 and 3707 candidate ncRNA loci with EvoFold and RNAz, respectively. Only 268 loci (5% and 7%, respectively) were found with both programs, representing a 1.6-fold enrichment over that expected by chance; the lack of more extensive overlap is due to the two programs having optimal sensitivity at different levels of GC content and conservation. We experimentally examined 50 of these targets using RACE/tiling-array analysis and brain and testis tissues (see Supplementary Information sections S2.11 and S2.9.3); the predictions were validated at a 56%, 65%, and 63% rate for Evofold, RNAz, and dual predictions, respectively.

Figure 4 Coverage of primary transcripts across ENCODE regions. Three different technologies [integrated annotation from GENCODE, RACE-array experiments (RxFrags), and PET tags] were used to assess the presence of a nucleotide in a primary transcript. Use of these (more ...)

Previous studies have suggested a similar broad amount of transcription across the human^{14, 15} and mouse²⁷ genomes. Our studies confirm these results, and have investigated the genesis of these transcripts in greater detail, confirming the presence of substantial intragenic and intergenic transcription. At the same time, many of the resulting transcripts are neither traditional protein-coding transcripts nor easily explained as structural non-coding RNAs. Other studies have noted complex transcription around specific loci or chimeric-gene structures (for example refs^28–30), but these have often been considered exceptions; our data show that complex intercalated transcription is common at many loci. The results presented in the next section show extensive amounts of regulatory factors around novel TSSs, which is consistent with this extensive transcription. The biological relevance of these unannotated transcripts remains unanswered by these studies. Evolutionary information (detailed below) is mixed in this regard, for example, it indicates that unannotated transcripts show weaker evolutionary conservation than many other annotated features. As with other ENCODE-detected elements, it is difficult to identify clear biological roles for the majority of these transcripts; such experiments are challenging to perform on a large scale, and furthermore, it seems likely that many of the corresponding biochemical events may be evolutionarily neutral (see below).

To better understand transcriptional regulation, we sought to begin cataloguing the regulatory elements residing within the 44 ENCODE regions. For this pilot project, we mainly focused on the binding of regulatory proteins and chromatin structure involved in transcriptional regulation. We analysed over 150 datasets, mainly from ChIP-chip^37–39, Chip-PET and STAGE^{40, 41} studies (see Supplementary Information sections S3.1 and S3.2). These methods use chromatin immunoprecipitation (ChIP) with specific antibodies to enrich for DNA in physical contact with the targeted epitope. This enriched DNA can then be analysed using either microarrays (ChIP-chip) or high-throughput sequencing (ChIP-PET and STAGE). The assays included 18 sequence-specific transcription factors and components of the general transcription machinery [for example, RNA polymerase II (PolII), TAF1, and TFIIB]. In addition, we tested more than 600 potential promoter fragments for transcriptional activity by transient-transfection reporter assays that utilized 16 human cell lines³³. We also examined chromatin structure by studying the ENCODE regions for DNaseI sensitivity (via quantitative PCR⁴² and tiling arrays^{43, 44}, see Supplementary Information section S3.3), histone composition⁴⁵, histone modifications (via ChIP-chip assays)^{37, 46}, and histone displacement (using FAIRE, see Supplementary Information section 3.4). Below, we detail these analyses, starting with the efforts to define and classify the 5′ ends of transcripts with respect to their associated regulatory signals. Following that are summaries of generated data about sequence-specific transcription-factor binding and clusters of regulatory elements. Finally, we describe how this information can be integrated to make predictions about transcriptional regulation.

The above effort yielded 7,157 TSSs clusters in the ENCODE regions. We classified these TSSs into three categories: Known (present at the end of GENCODE-defined transcripts), Novel (supported by other evidence), and Unsupported. The Novel TSSs were further subdivided based on the nature of the supporting evidence (see Table 3 and Supplementary Information section S3.5), with all four of the resulting subtypes showing significant overlap with experimental evidence using the GSC statistic. Although there is a larger relative proportion of singleton tags in the Novel category, when analysis is restricted to only singleton tags, the Novel TSSs continue to have highly significant overlap to supporting evidence (see Supplementary Information section S3.5.1).

Table 3 Different categories of TSSs defined on the basis of support from different transcript-survey methods.

Figure 5 Aggregate signals of tiling-array experiments from either ChIP-chip or chromatin structure assays, represented for different classes of TSS and DHS. For each plot, the signal was first normalised with a mean of 0 and standard deviation of 1, and then (more ...)

In the case of the three TSS categories (Known, Novel, and Unsupported), Known and Novel TSSs are both associated with similar signals for multiple factors (ranging from histone modifications through DNaseI accessibility), whereas Unsupported TSSs are not. The enrichments seen with chromatin modifications and sequence-specific factors, along with the significant clustering of this evidence, indicate that the Novel TSSs do not reflect false positives and likely utilise the same biological machinery as other promoters. Sequence-specific transcription factors show a marked increase in binding across the broad region that encompasses each TSS. This increase is notably symmetric, with binding equally likely upstream or downstream of a TSS (see Supplementary Information section S3.7 for an explanation of why this symmetrical signal is not an artefact due to the analysis of the signals). Further, there is enrichment of BAF155 binding (a member of the swi/snf chromatin-modifying complex), which persists across a broader extent than other factors. The broad signals with this factor indicate that the ChIP-chip results reflect both specific enrichment at the TSS and broader enrichments across ~5kb regions; (this is not due to technical issues, see Supplementary Information section S3.8).

We selected 577 GENCODE-defined TSSs at the 5′ ends of a protein-coding transcript with over 3 exons to assess expression status. Each transcript was classified as: (1) ’active’ (gene on) or ‘inactive’ (gene off) based on the unbiased transcript surveys, and (2) residing near a ‘CpG island’ or not near a CpG island (‘non-CpG island’) (see Supplementary Information section S3.17). As expected, the aggregate signal of histone modifications is mainly attributable to active TSSs (Figure 5), in particular those near CpG islands. Pronounced doublet peaks at the TSS can be seen with these large signals, similar to previous work in yeast⁴⁸, due to the chromatin accessibility at the TSS. Many of the histone marks and PolII signals are now clearly asymmetrical, with a persistent level of PolII into the genic region, as expected. However, the sequence specific factors remain largely symmetrically distributed. TSSs near CpG islands show a broader distribution of histone marks than those not near CpG islands (see Supplementary Information section S3.6). The binding of some transcription factors (E2F1, E2F4, and cMyc) is extensive in the case of active genes, and is lower (or absent) in the case of inactive genes.

Figure 6 Distribution of RFBRs relative to GENCODE TSSs. Different RFBRs from Sequence Specific factors (Red) or general factors (Blue) are plotted showing their relative distribution near TSSs. The x-axis indicates the proportion of TSSs close (within 2.5KB) (more ...)

To investigate the relationship between chromatin structure and gene expression, we examined transcript levels in two cell lines using a transcript-tiling array. We compared this transcript data with the results of ChIP-chip experiments that measured histone modifications across the ENCODE regions. From this, we developed a variety of predictors of expression status using chromatin modifications as variables; these were derived using both Decision Trees and SVMs (see Supplementary Information section S3.17). The best of these correctly predicts expression status (transcribed vs. non-transcribed) in 91% of cases. This success rate did not decrease dramatically when the predicting algorithm incorporated the results from one cell line to predict the expression status of another cell line. Interestingly, despite the striking difference in histone-modification enrichments in TSSs residing near versus those more distal to CpG islands (see Figure 5 and Supplementary Information section S3.6), including information about the proximity to CpG islands did not improve the predictors. This suggests that despite the marked differences in histone modifications among these TSS classes, a single predictor can be made, using the interactions between the different histone modification levels.

In summary, we have integrated many datasets to provide a more complete view of regulatory information, both around specific sites (TSSs and DHSs) and in an unbiased manner. Based on analysing multiple datasets, we find 4,491 Known and Novel TSSs in the ENCODE regions, almost ten-fold more than the number of established genes. This large number of TSSs might explain the extensive transcription described above; it also begins to change our perspective about regulatory information – without such a large TSS catalogue, many of the regulatory clusters would have been classified as residing distal to promoters. In addition to this revelation about the abundance of promoter-proximal regulatory elements, we also identified a considerable number of putative distal regulatory elements, particularly based on the presence of DHSs. Our study of distal regulatory elements was probably most hindered by the paucity of data generated using distal element-associated transcription factors; nevertheless, we clearly detected a set of distal DHS-associated segments bound by CTCF or cMyc. Finally, we showed that information about chromatin structure alone could be used to make effective predictions about both the location and activity of TSSs.

The ENCODE Project provided an unique opportunity to examine whether individual histone modifications on human chromatin can be the correlated with the time of replication and whether such correlations support the general relationship of active, open chromatin with early replication. Our studies also tested whether segments showing interallelic variation in time of replication have two different types of histone modifications consistent with an interallelic variation in chromatin state.

Figure 7 Correlation between replication timing and histone modifications. (a) Comparison of two histone modifications (H3K4me2 and H3K27me3), plotted as enrichment ratio from the Chip-chip experiments and the time for 50% of the DNA to replicate (TR50), indicated (more ...)

While most genomic regions replicate in a temporally specific window in S phase other regions demonstrate an atypical pattern of replication (Pan-S) where replication signals are seen in multiple parts of S phase. We have suggested that such a pattern of replication stems from interallelic variation in the chromatin structure^{59, 75}. If one allele is in active chromatin and the other in repressed chromatin, both types of modified histones are expected to be enriched in the Pan-S segments. An ENCODE region was classified as non-specific (or Pan-S) regions when >60% of the probes in a 10 kb window replicated in multiple intervals in S phase. The remaining, regions were sub-classified into early, mid, or late replicating based on the average TR50 of the temporally-specific probes within a 10kb window⁷⁵. For regions of each class of replication timing, we determined the relative enrichment of various histone-modification peaks in HeLa cells (Figure 7b; Supplemental material S4.4). The correlations of activating and repressing histone-modification peaks with TR50) are confirmed by this analysis (Figure 7b). Intriguingly, the Pan-S segments are unique in being enriched for both activating (H3K4me2, H3ac, and H4ac) and repressing (H3K27me3) histones, consistent with the suggestion that the Pan-S replication pattern arises from interallelic variation in chromatin structure and time of replication⁷⁵. This observation is also consistent with the Pan-S replication pattern seen for the H19/IGF2 locus, a known imprinted region with differential epigenetic modifications across the two alleles⁷⁶.

The extensive rearrangements in the genome of HeLa cells led us to ask whether the detected correlations between TR50 and chromatin state are seen with other cell lines. The histone-modification data with GM06990 cells allowed us to test whether the time of replication of genomic segments in HeLa cells correlated with the chromatin state in GM06990 cells. Early- and late-replicating segments in HeLa cells are enriched and depleted, respectively, for activating marks in GM06990 cells (Figure 7b). Thus, despite the presence of genomic rearrangements (see Supplementary Information section S2.12), the TR50 and chromatin state in HeLa cells are not far from a constitutive baseline also seen with a cell line from a different lineage. The enrichment of multiple activating histone modifications and the depletion of a repressive modification from segments that replicate early in S phase extends previous work in the field at a level of detail and scale not attempted before in mammalian cells. The duality of histone modification patterns in Pan-S areas of the HeLa genome, and the concordance of chromatin marks and replication time across two disparate cell lines (HeLa and GM06990) show the coordination of histone modifications with replication in the human genome.

Previous studies comparing the human, mouse, rat, and dog genomes examined bulk evolutionary properties of all nucleotides in the genome, and provided little insight about the precise positions of constrained bases. Interestingly, these studies indicated that the majority of constrained bases reside within the non-coding portion of the human genome. Meanwhile, increasingly rich datasets of polymorphisms across the human genome have been used extensively to establish connections between genetic variants and disease, but far fewer analyses have sought to use such data for assessing functional constraint⁸⁵.

The ENCODE Project provides an excellent opportunity for more fully exploiting inter- and intra-species sequence comparisons to examine genome function in the context of extensive experimental studies on the same regions of the genome. We consolidated the experimentally-derived information about the ENCODE regions and focused our analyses on 11 major classes of genomic elements. These classes are listed in Table 4 and include two non-experimentally-derived datasets: Ancient Repeats (ARs; mobile elements that inserted early in the mammalian lineage, have subsequently become dormant, and are assumed to be neutrally evolving) and Constrained Sequences (CSs; regions that evolve detectably more slowly than neutral sequences).

Table 4 Eleven classes of genomic elements subjected to evolutionary and population-genetics analyses.

Intra-species variation studies mainly used single-nucleotide polymorphism (SNP) data from Phases I and II, and the 10 resequenced regions in ENCODE regions with 48 individuals of the HapMap Project¹⁰³, nucleotide insertion or deletion (indel) data were from the SNP Consortium and HapMap. We also examined the ENCODE regions for the presence of overlaps with known segmental duplications¹⁰⁴ and copy-number variants (CNVs).

Figure 10 Relative proportion of different annotations among constrained sequences. The 4.9% of bases in the ENCODE regions identified as constrained is subdivided into the portions that reflect known coding regions, UTRs, other experimentally-annotated regions, (more ...)

Figure 11 Overlap of constrained sequences and various experimental annotations. (a) A schematic depiction shows the different tests used for assessing overlap between experimental annotations and constrained sequences, both for individual bases and for entire (more ...)

We also examined measures of human variation (heterozygosity, derived allele-frequency spectra, and indel rates) within the sequences of the experimentally-identified functional elements (Figure 12). For these studies, ARs were used as a marker for neutrally evolving sequence. Most experimentally-identified functional elements are associated with lower heterozygosity compared to ARs, and a few have lower indel rates compared to ARs. Striking outliers are 3′ UTRs, which have dramatically increased indel rates without an obvious cause. This is discussed in more depth in Clark et al¹⁰⁷.

Figure 12 Relationship between heterozygosity and polymorphic indel rate for a variety of experimental annotations.. 3′UTRs are an expected outlier for the indel measures due to the presence of low-complexity sequence (leading to a higher indel rate).

These findings indicate that the majority of the evolutionarily-constrained, experimentally-identified functional elements show evidence of negative selection both across mammalian species and within the human population. Furthermore, we have assigned at least one molecular function to the majority (60%) of all constrained bases in the ENCODE regions.

We also examined structural variation (i.e., CNVs, inversions, and translocations¹¹⁴; Supplementary Information section S5.2.2). Within these polymorphic regions, we encountered significant overrepresentation of CDSs, TxFrags, and intra-species constrained sequences (P<10⁻³, Figure 13), and also detected a statistically significant under-representation of ARs (P=10⁻³). A similar over-representation of CDSs and intra-species constrained sequences was found within non-polymorphic segmental duplications.

Figure 13 CNV enrichment. The relative enrichment of different experimental annotations in ENCODE regions associated with CNVs. CS_non-CDS are constrained sequences outside of coding regions. A value of 1 or less indicates no enrichment, and values greater than (more ...)

There are two methodological reasons that might explain the apparent excess of unconstrained experimentally-identified functional elements: the underestimation of sequence constraint or overestimation of experimentally-identified functional elements. We do not believe that either of these explanations fully accounts for the large and varied levels of unconstrained experimentally functional sequences. The set of constrained bases analysed here is highly accurate and complete due to the depth of the multiple alignment. Both by bulk fitting procedures and by comparison of SNP frequencies to constraint there is clearly a proportion of constrained bases not captured in the defined 4.9% of constrained sequences, but it is small (see Supplementary Information section S5.4 and S5.5). More aggressive schemes to detect constraint only marginally increase the overlap with experimentally-identified functional elements, and do so with considerably less specificity. Similarly, all experimental findings have been independently validated and, for the least constrained experimentally-identified functional elements (Un.TxFrags and binding sites of sequence-specific factors), there is both internal validation and cross-validation from different experimental techniques. This suggests that there is not likely a significant overestimation of experimentally-identified functional elements. Thus, these two explanations may contribute to the general observation about unconstrained functional elements, but cannot fully explain it.

Instead, we hypothesize five biological reasons to account for the presence of large amounts of unconstrained functional elements. The first two are particular to certain biological assays, where the elements being measured are connected to but do not perfectly coincide with the analysed region. An example of this is the parent transcript of an miRNA, where the current assays detect the exons (some of which are not under evolutionary selection), whereas the intronic miRNA actually harbours the constrained bases. Nevertheless, the transcript sequence provides the critical coupling between the regulated promoter and the miRNA. The sliding of transcription factors (which might bind a specific sequence but then migrate along the DNA) or the processivity of histone modifications across chromatin are more exotic examples of this. A related, second hypothesis is that delocalised behaviours of the genome, such a general chromatin accessibility, may be maintained by some biochemical processes (such as transcription of intergenic regions or specific factor binding) without the requirement for specific sequence elements. These two explanations of both connected components and diffuse components related to, but not coincident with, constrained sequences are particularly relevant for the considerable amount of unannotated and unconstrained transcripts.

The other three hypotheses may be more general. - the presence of neutral (or near neutral) biochemical elements, of lineage-specific functional elements, and of functionally conserved but non-orthologous elements. We believe there are a considerable proportion of neutral biochemically active elements that do not confer a selective advantage or disadvantage to the organism. This neutral pool of sequence elements may turn over during evolutionary time, emerging via certain mutations and disappearing by others. The size of the neutral pool would largely be determined by the rate of emergence and extinction via chance events; low information-content elements, such as transcription factor-binding sites¹¹⁰ will have larger neutral pools. Second, from this neutral pool, some elements might occasionally acquire a biological role and so come under evolutionary selection. The acquisition of a new biological role would then create a lineage-specific element. Finally, a neutral element from the general pool could also become a peer of an existing selected functional element and either of the two elements could then be removed by chance. If the older element is removed, the newer element has, in essence, been conserved without using orthologous bases, providing a conserved function in the absence of constrained sequences. For example, a common HNF4A binding site in the human and mouse genomes may not reflect orthologous human and mouse bases, though the presence of an HNF4A site in that region was evolutionarily selected for in both lineages. Note that both the neutral turnover of elements and the ‘functional peering’ of elements has been suggested for cis-acting regulatory elements in Drosophila^{115, 116} and mammals¹¹⁰. Our data support these hypotheses, and we have generalized this idea over many different functional elements. The presence of conserved function encoded by conserved orthologous bases is a commonplace assumption in comparative genomics; our findings suggest that there could be a sizable set of functionally-conserved but non-orthologous elements in the human genome, and that these seem unconstrained across mammals. Functional data akin to the ENCODE Project on other related species, such as mouse, would be critical to understanding the rate of such functionally-conserved but non-orthologous elements.

The generation and analyses of over 200 experimental datasets from studies examining the 44 ENCODE regions provide a rich source of functional information for 30 Mb of the human genome. The first conclusion of these efforts is that these data are remarkably informative. Although there will be on going work to enhance existing assays, invent new techniques, and develop new data-analysis methods, the generation of genome-wide experimental datasets akin to the ENCODE pilot phase would provide an impressive platform for future genome-exploration efforts. This now seems feasible in light of throughput improvements of many of the assays and the ever-declining costs of whole-genome tiling arrays and DNA sequencing. Such genome-wide functional data should be acquired and released openly, as has been done with other large-scale genome projects, to ensure its availability to as a new foundation for all biologists studying the human genome. It is these biologists who will often provide the critical link from biochemical function to biological role for the identified elements.

The scale of the pilot phase of the ENCODE Project was also sufficiently large and unbiased to reveal important principles about the organisation of functional elements in the human genome. In many cases, these principles agree with current mechanistic models. For example, trimethylation of H3K4 is enriched near active genes, which we have further refined to the ability to accurately predict gene activity based on histone modifications. However, we also uncovered some surprises that challenge the current dogma on biological mechanisms. The generation of numerous intercalated transcripts spanning the majority of the genome has been repeatedly suggested^{13, 14}, but this phenomenon has been met with mixed opinions about the biological importance of these transcripts. Our analyses of numerous orthogonal datasets firmly establish the presence of these transcripts, and thus the simple view of the genome as having a defined set of isolated loci transcribed independently does not appear to be accurate. Perhaps the genome encodes a network of transcripts, many of which are linked to protein-coding transcripts and the majority of which we cannot (yet) assign a biological role. Our perspective of transcription and genes may have to evolve and also poses some interesting mechanistic questions. For example, how are splicing signals coordinated and used when there are so many overlapping primary transcripts? Similarly, to what extent does this reflect neutral turnover of reproducible transcripts with no biological role?

We gained subtler but equally important mechanistic findings relating to transcription, replication, and chromatin modification. Transcription factors previously thought to primarily bind promoters are more general, and those which do bind to promoters are equally likely to be downstream of a TSS as upstream. Interestingly, many elements that previously were classified as distal enhancers are, in fact, close to one of the newly-identified TSSs; only about 35% of sites showing evidence of binding by multiple transcription factors are actually distal to a TSS. This need not imply that most regulatory information is confined to classic promoters, but rather it does suggest that transcription and regulation are coordinated actions beyond just the traditional promoter sequences. Meanwhile, while distal regulatory elements could be identified in the ENCODE regions, they are currently difficult to classify, in part due to the lack of a broad set of transcription factors to use in analyzing such elements. Finally, we now have a much better appreciation about how DNA replication is coordinated with histone modifications.

At the outset of the ENCODE Project, many believed that the broad collection of experimental data would nicely dovetail with the detailed evolutionary information derived from comparing multiple mammalian sequences to provide a neat ‘dictionary’ of conserved genomic elements, each with a growing annotation about their biochemical function(s). In one sense, this was achieved; the majority of constrained bases in the ENCODE regions are now associated with at least some experimentally-derived information about function. However, we have also encountered a remarkable excess of unconstrained experimentally-identified functional elements, and these cannot be dismissed for technical reasons. This is perhaps the biggest surprise of the pilot phase of the ENCODE Project, and suggests that we take a more ‘neutral’ view of many of the functions conferred by the genome.

The methods are described in the Supplementary Information, with more technical details for each experiment often found in the references provided in Table 1. The Supplement sections are arranged in the same order as the manuscript (with similar headings to facilitate cross-referencing). The first page of the Supplement also has an index to aid navigation. Raw data are available in ArrayExpress, GEO, or EMBL/GenBank archive as appropriate, as detailed in Supplementary Information section S1.1. Processed data are also presented in a user-friendly manner at the UCSC Genome Browser’s ENCODE portal (http://genome.ucsc.edu/ENCODE/).

We thank D. Leja for providing graphical expertise and support. Funding support is acknowledged from the following sources: National Institutes of Health, The European Union BioSapiens NoE, Affymetrix, Swiss National Science Foundation, the Spanish Ministerio de Educación y Ciencia, Spanish Ministry of Education and Science, CIBERESP, Genome Spain and Generalitat de Catalunya, Ministry of Education, Culture, Sports, Science and Technology of Japan, the NCCR Frontiers in Genetics, the Jésrôme Lejeune Foundation, the Childcare Foundation, the Novartis Foundations, the Danish Research Council, the Swedish Research Council, the Knut and Alice Wallenberg Foundation, the Wellcome Trust, the Howard Hughes Medical Institute, the Bio-X Institute, the RIKEN Institute, the US Army, National Science Foundation, the Deutsche Forschungsgemeinschaft, the Austrian Gen-AU program, the BBSRC and The European Molecular Biology Laboratory. We thank the Barcelona SuperComputing Center and the NIH Biowulf cluster for computer facilities. The Consortium thanks the ENCODE Scientific Advisory Panel for their advice on the project: G. Weinstock, M. Cherry, G. Churchill, M. Eisen, S. Elgin, J. Lis, J. Rine, M. Vidal and P. Zamore.

Analysis Coordination

Ewan Birney*,¹, John A. Stamatoyannopoulos*,², Anindya Dutta*,³, Roderic Guigó*,^{4, 5}, Thomas R. Gingeras*,⁶, Elliott H. Margulies*,⁷, Zhiping Weng*,^{8, 9}, Michael Snyder*,^{10, 11}, Emmanouil T. Dermitzakis*,¹²;

Chromatin and Replication

John A. Stamatoyannopoulos*,², Robert E. Thurman^{2, 13}, Michael S. Kuehn^{2, 13}, Christopher M. Taylor³, Shane Neph², Christoph M. Koch¹², Saurabh Asthana¹⁴, Ankit Malhotra³, Ivan Adzhubei¹⁴, Jason A. Greenbaum¹⁵, Robert M. Andrews¹², Paul Flicek¹, Patrick J. Boyle³, Hua Cao¹³, Nigel P. Carter¹², Gayle K. Clelland¹², Sean Davis¹⁶, Nathan Day², Pawandeep Dhami¹², Shane C. Dillon¹², Michael O. Dorschner², Heike Fiegler¹², Paul G. Giresi¹⁷, Jeff Goldy², Michael Hawrylycz¹⁸, Andrew Haydock², Richard Humbert², Keith D. James¹², Brett E. Johnson¹³, Ericka M. Johnson¹³, Tristan T. Frum¹³, Elizabeth R. Rosenzweig¹³, Neerja Karnani³, Kirsten Lee², Gregory C. Lefebvre¹², Patrick A. Navas¹³, Fidencio Neri², Stephen C. J. Parker¹⁵, Peter J. Sabo², Richard Sandstrom², Anthony Shafer², David Vetrie¹², Molly Weaver², Sarah Wilcox¹², Man Yu¹³, Francis S. Collins⁷, Job Dekker¹⁹, Jason D. Lieb¹⁷, Thomas D. Tullius¹⁵, Gregory E. Crawford²⁰, Shamil Sunayev¹⁴, William S. Noble², Ian Dunham¹², Anindya Dutta*,³;

Genes and Transcripts

Roderic Guigó*,^{4, 5}, France Denoeud⁵, Alexandre Reymond^{21, 22}, Philipp Kapranov⁶, Joel Rozowsky¹¹, Deyou Zheng¹¹, Robert Castelo⁵, Adam Frankish¹², Jennifer Harrow¹², Srinka Ghosh⁶, Albin Sandelin²³, Ivo L. Hofacker²⁴, Robert Baertsch^{25, 26}, Damian Keefe¹, Paul Flicek¹, Sujit Dike⁶, Jill Cheng⁶, Heather A. Hirsch²⁷, Edward A. Sekinger²⁷, Julien Lagarde⁵, Josep F. Abril^{5, 28}, Atif Shahab²⁹, Christoph Flamm^{24, 30}, Claudia Fried³⁰, Jörg Hackermüller³¹, Jana Hertel³⁰, Manja Lindemeyer³⁰, Kristin Missal^{30, 32}, Andrea Tanzer^{24, 30}, Stefan Washietl²⁴, Jan Korbel¹¹, Olof Emanuelsson¹¹, Jakob S. Pedersen²⁶, Nancy Holroyd¹², Ruth Taylor¹², David Swarbreck¹², Nicholas Matthews¹², Mark C. Dickson³³, Daryl J. Thomas^{25, 26}, Matthew T. Weirauch²⁵, James Gilbert¹², Jorg Drenkow⁶, Ian Bell⁶, XiaoDong Zhao³⁴, K.G. Srinivasan³⁴, Wing-Kin Sung³⁴, Hong Sain Ooi³⁴, Kuo Ping Chiu³⁴, Sylvain Foissac⁴, Tyler Alioto⁴, Michael Brent³⁵, Lior Pachter³⁶, Michael L. Tress³⁷, Alfonso Valencia³⁷, Siew Woh Choo³⁴, Chiou Yu Choo³⁴, Catherine Ucla²², Caroline Manzano²², Carine Wyss²², Evelyn Cheung⁶, Taane G. Clark³⁸, James B. Brown³⁹, Madhavan Ganesh⁶, Sandeep Patel⁶, Hari Tammana⁶, Jacqueline Chrast²¹, Charlotte N. Henrichsen²¹, Chikatoshi Kai²³, Jun Kawai^{23, 40}, Ugrappa Nagalakshmi¹⁰, Jiaqian Wu¹⁰, Zheng Lian⁴¹, Jin Lian⁴¹, Peter Newburger⁴², Xueqing Zhang⁴², Peter Bickel⁴³, John S. Mattick⁴⁴, Piero Carninci⁴⁰,Yoshihide Hayashizaki^{23, 40}, Sherman Weissman⁴¹, Emmanouil T. Dermitzakis*,¹², Elliott H. Margulies*,⁷, Tim Hubbard¹², Richard M. Myers³³, Jane Rogers¹², Peter F. Stadler^{24, 30, 45}, Todd M. Lowe²⁵, Chia-Lin Wei³⁴, Yijun Ruan³⁴, Michael Snyder*,^{10, 11}, Ewan Birney*,¹, Kevin Struhl²⁷, Mark Gerstein^{11, 46, 47}, Stylianos E. Antonarakis²², Thomas R. Gingeras*,⁶;

Integrated Analysis and Manuscript Preparation

James B. Brown³⁹, Paul Flicek¹, Yutao Fu⁸, Damian Keefe¹, Ewan Birney*,¹, France Denoeud⁵, Mark Gerstein^{11, 46, 47}, Eric D. Green^{7, 48}, Philipp Kapranov⁶, Ulaf Karaöz⁸, Richard M. Myers³³, William S. Noble², Alexandre Reymond^{21, 22}, Joel Rozowsky¹¹, Kevin Struhl²⁷, Adam Siepel^{25, 26}, $, John A. Stamatoyannopoulos*,², Christopher M. Taylor³, James Taylor^{49, 50}, Robert E. Thurman^{2, 13}, Thomas D. Tullius¹⁵, Stefan Washietl²⁴, Deyou Zheng¹¹;

Management Group

Laura A. Liefer⁵¹, Kris A. Wetterstrand⁵¹, Peter J. Good⁵¹, Elise A. Feingold⁵¹, Mark S. Guyer⁵¹, Francis S. Collins⁵²;

Multi-species Sequence Analysis

Elliott H. Margulies*,⁷, Gregory M. Cooper³³,%, George Asimenos⁵³, Daryl J. Thomas^{25, 26}, Colin N. Dewey⁵⁴, Adam Siepel^{25, 26},$, Ewan Birney*,¹ , Damian Keefe¹, Minmei Hou^{49, 50}, James Taylor^{49, 50}, Sergey Nikolaev²², Juan I. Montoya-Burgos⁵⁵, Ari Löytynoja¹, Simon Whelan¹,¶, Fabio Pardi¹, Tim Massingham¹, James B. Brown³⁹, Haiyan Huang⁴³, Nancy R. Zhang^{43, 56}, Peter Bickel⁴³, Ian Holmes⁵⁷, James C. Mullikin^{7, 48}, Abel Ureta-Vidal¹, Benedict Paten¹, Michael Seringhaus¹¹, Deanna Church⁵⁸, Kate Rosenbloom²⁶, W. James Kent^{25, 26}, NISC Comparative Sequencing Program‡, Baylor College of Medicine Human Genome Sequencing Center‡, Washington University Genome Sequencing Center‡, Broad Institute‡, Children’s Hospital Oakland Research Institute‡, Mark Gerstein^{11, 46, 47}, Stylianos E. Antonarakis²², Serafim Batzoglou⁵³, Nick Goldman¹, Ross C. Hardison^{50, 59}, David Haussler^{25, 26, 60}, Webb Miller^{49, 50, 61}, Lior Pachter³⁶, Eric D. Green^{7, 48}, Arend Sidow^{33, 62};

‡ A list of participants and affiliations appears below

NISC Comparative Sequencing Program

Gerard G. Bouffard^{7, 48}, Xiaobin Guan⁴⁸, Nancy F. Hansen⁴⁸, Jacquelyn R. Idol⁷, Valerie V.B. Maduro⁷, Baishali Maskeri⁴⁸, Jennifer C. McDowell⁴⁸, Morgan Park⁴⁸, Pamela J. Thomas⁴⁸, Alice C. Young⁴⁸, and Robert W. Blakesley^{7, 48};

Baylor College of Medicine Human Genome Sequencing Center

Donna M. Muzny⁶³, Erica Sodergren⁶³, David A. Wheeler⁶³, Kim C. Worley⁶³, Huaiyang Jiang⁶³, George M. Weinstock⁶³, and Richard A. Gibbs⁶³;

Washington University Genome Sequencing Center

Tina Graves⁶⁴, Robert Fulton⁶⁴, Elaine R. Mardis⁶⁴, and Richard K. Wilson⁶⁴;

Broad Institute

Michele Clamp⁶⁵, James Cuff⁶⁵, Sante Gnerre⁶⁵, David B. Jaffe⁶⁵, Jean L. Chang⁶⁵, Kerstin Lindblad-Toh⁶⁵, and Eric S. Lander^{65, 66};

Children’s Hospital Oakland Research Institute

Maxim Koriabine⁶⁷, Mikhail Nefedov⁶⁷, Kazutoyo Osoegawa⁶⁷, Yuko Yoshinaga⁶⁷, Baoli Zhu⁶⁷, and Pieter J. de Jong⁶⁷;

Transcriptional Regulatory Elements

Zhiping Weng*,^{8, 9}, Nathan D. Trinklein³³,#, Yutao Fu⁸, Zhengdong D. Zhang¹¹, Ulaf Karaöz⁸, Leah Barrera⁶⁸, Rhona Stuart⁶⁸, Deyou Zheng¹¹, Srinka Ghosh⁶, Paul Flicek¹, David C. King^{50, 59}, James Taylor^{49, 50}, Adam Ameur⁶⁹, Stefan Enroth⁶⁹, Mark C. Bieda⁷⁰, Christoph M. Koch¹², Heather A. Hirsch²⁷, Chia-Lin Wei³⁴, Jill Cheng⁶, Jonghwan Kim⁷¹, Akshay A. Bhinge⁷¹, Paul G. Giresi¹⁷, Nan Jiang⁷², Jun Liu³⁴, Fei Yao³⁴, Wing-Kin Sung³⁴, Kuo Ping Chiu³⁴, Vinsensius B. Vega³⁴, Charlie W.H. Lee³⁴, Patrick Ng³⁴, Atif Shahab²⁹, Edward A. Sekinger²⁷, Annie Yang²⁷, Zarmik Moqtaderi²⁷, Zhou Zhu²⁷, Xiaoqin Xu⁷⁰, Sharon Squazzo⁷⁰, Matthew J. Oberley⁷³, David Inman⁷³, Michael A. Singer⁷², Todd A. Richmond⁷², Kyle J. Munn^{72, 74}, Alvaro Rada-Iglesias⁷⁴, Ola Wallerman⁷⁴, Jan Komorowski⁶⁹, Gayle K. Clelland¹², Sarah Wilcox¹², Shane C. Dillon¹², Robert M. Andrews¹², Joanna C. Fowler¹², Phillippe Couttet¹², Keith D. James¹², Gregory C. Lefebvre¹², Alexander W. Bruce¹², Oliver M. Dovey¹², Peter D. Ellis¹², Pawandeep Dhami¹², Cordelia F. Langford¹², Nigel P. Carter¹², David Vetrie¹², Philipp Kapranov⁶, David A. Nix⁶, Ian Bell⁶, Sandeep Patel⁶, Joel Rozowsky¹¹, Ghia Euskirchen¹⁰, Stephen Hartman¹⁰, Jin Lian⁴¹, Jiaqian Wu¹⁰, Alexander E. Urban¹⁰, Peter Kraus¹⁰, Sara Van Calcar⁶⁸, Nate Heintzman⁶⁸, Tae Hoon Kim⁶⁸, Kun Wang⁶⁸, Chunxu Qu⁶⁸, Gary Hon⁶⁸, Rosa Luna⁷⁵, Christopher K. Glass⁷⁵, M. Geoff Rosenfeld⁷⁵, Shelley Force Aldred³³,#, Sara J. Cooper³³, Anason Halees⁸, Jane M. Lin⁹, Hennady P. Shulha⁹, Xiaoling Zhang⁸, Mousheng Xu⁸, Jaafar N. S. Haidar⁹, Yong Yu⁹, Ewan Birney*,¹, Sherman Weissman⁴¹, Yijun Ruan³⁴, Jason D. Lieb¹⁷, Vishwanath R. Iyer⁷¹, Roland D. Green⁷², Thomas R. Gingeras*,⁶, Claes Wadelius⁷⁴, Ian Dunham¹², Kevin Struhl²⁷, Ross C. Hardison^{50, 59}, Mark Gerstein^{11, 46, 47}, Peggy J. Farnham⁷⁰, Richard M. Myers³³, Bing Ren⁶⁸, Michael Snyder*,^{10, 11};

UCSC Genome Browser

Daryl J. Thomas^{25, 26}, Kate Rosenbloom²⁶, Rachel A. Harte²⁶, Angie S. Hinrichs²⁶, Heather Trumbower²⁶, Hiram Clawson²⁶, Jennifer Hillman-Jackson²⁶, Ann S. Zweig²⁶, Kayla Smith²⁶, Archana Thakkapallayil²⁶, Galt Barber²⁶, Robert M. Kuhn²⁶, Donna Karolchik²⁶, David Haussler^{25, 26, 60}, W. James Kent^{25, 26};

Variation

Emmanouil T. Dermitzakis*,¹², Lluis Armengol⁷⁶, Christine P. Bird¹², Taane G. Clark³⁸, Gregory M. Cooper³³,%, Paul I. W. de Bakker⁷⁷, Andrew D. Kern²⁶, Nuria Lopez-Bigas⁵, Joel D. Martin^{50, 59}, Barbara E. Stranger¹², Daryl J. Thomas^{25, 26}, Abigail Woodroffe⁷⁸, Serafim Batzoglou⁵³, Eugene Davydov⁵³, Antigone Dimas¹², Eduardo Eyras⁵, Ingileif B. Hallgrímsdóttir⁷⁹, Ross C. Hardison^{50, 59}, Julian Huppert¹², Arend Sidow^{33, 62}, James Taylor^{49, 50}, Heather Trumbower²⁶, Michael C. Zody⁷⁷, Roderic Guigó*,^{4, 5}, James C. Mullikin⁷, Gonçalo R. Abecasis⁷⁸, Xavier Estivill^{76, 80} and Ewan Birney*,¹.

* Co-Chairs of the ENCODE analysis groups, and corresponding authors (E-mail: Ewan Birney, birney/at/ebi.ac.uk; John A. Stamatoyannopoulos, jstam/at/u.washington.edu; Anindya Dutta, ad8q/at/virginia.edu; Roderic Guigó, rguigo/at/imim.es; Thomas R. Gingeras, Tom_Gingeras/at/affymetrix.com; Elliott H. Margulies, elliott/at/nhgri.nih.gov; Zhiping Weng, zhiping/at/bu.edu; Michael Snyder, michael.snyder/at/yale.edu; Emmanouil T. Dermitzakis, md4/at/sanger.ac.uk)

% Current Address: Department of Genome Sciences, University of Washington School of Medicine, Seattle, Washington 98195, USA.

$ Current Address: Department of Biological Statistics & Computational Biology, Cornell University, Ithaca, New York 14853, USA.

¶ Current Address: Faculty of Life Sciences, University of Manchester, Michael Smith Building, Oxford Road, Manchester, M13 9PT, UK.

# Current Address: SwitchGear Genomics, 1455 Adams Drive, Suite 2015, Menlo Park, California 94025, USA.

• EMBL-European Bioinformatics Institute, Wellcome Trust Genome Campus, Hinxton, Cambridge, CB10 1SD, UK.
• Department of Genome Sciences, 1705 NE Pacific Street, Box 357730, University of Washington, Seattle, Washington 98195, USA.
• Department of Biochemistry and Molecular Genetics, Jordan 1240, Box 800733, 1300 Jefferson Park Ave, University of Virginia School of Medicine, Charlottesville, Virginia 22908, USA.
• Genomic Bioinformatics Program, Center for Genomic Regulation, C/Dr. Aiguader 88, Barcelona Biomedical Research Park Building, 08003 Barcelona, Catalonia, Spain.
• Research Group in Biomedical Informatics, Institut Municipal d’Investigació Mèdica/Universitat Pompeu Fabra, C/Dr. Aiguader 88, Barcelona Biomedical Research Park Building, 08003 Barcelona, Catalonia, Spain.
• Affymetrix, Inc., Santa Clara, California 95051, USA.
• Genome Technology Branch, National Human Genome Research Institute, National Institutes of Health, Bethesda, Maryland 20892, USA.
• Bioinformatics Program, Boston University, 24 Cummington St., Boston, Massachusetts 02215, USA.
• Biomedical Engineering Department, Boston University, 44 Cummington St., Boston, Massachusetts 02215, USA.
• Department of Molecular, Cellular, and Developmental Biology, Yale University, New Haven, Connecticut 06520, USA.
• Department of Molecular Biophysics and Biochemistry, Yale University, PO Box 208114, New Haven, Connecticut 06520, USA.
• The Wellcome Trust Sanger Institute, Wellcome Trust Genome Campus, Hinxton, Cambridge, CB10 1SA, UK.
• Division of Medical Genetics, 1705 NE Pacific Street, Box 357720, University of Washington, Seattle, Washington 98195, USA.
• Division of Genetics, Brigham and Women’s Hospital and Harvard Medical School, 77 Avenue Louis Pasteur, Boston, Massachusetts 02115, USA.
• Department of Chemistry and Program in Bioinformatics, Boston University, 590 Commonwealth Ave, Boston, Massachusetts 02215, USA.
• Genetics Branch, Center for Cancer Research, National Cancer Institute, National Institutes of Health, Bethesda, Maryland 20892, USA.
• Department of Biology and Carolina Center for Genome Sciences, CB# 3280, 202 Fordham Hall, The University of North Carolina at Chapel Hill, Chapel Hill, North Carolina 27599, USA.
• Allen Institute for Brain Sciences, 551 N. 34th Street, Seattle, Washington 98103, USA.
• Program in Gene Function and Expression and Department of Biochemistry and Molecular Pharmacology, University of Massachusetts Medical School, 364 Plantation Street, Worcester, Massachusetts 01605, USA.
• Institute for Genome Sciences & Policy and Department of Pediatrics, 101 Science Drive, Duke University, Durham, North Carolina 27708, USA.
• Center for Integrative Genomics, University of Lausanne, Genopode building, 1015 Lausanne, Switzerland.
• Department of Genetic Medicine and Development, University of Geneva Medical School, 1211 Geneva, Switzerland.
• Genome Exploration Research Group, RIKEN Genomic Sciences Center (GSC), RIKEN Yokohama Institute, 1-7-22, Suehiro-cho, Tsurumi-ku, Yokohama, Kanagawa, 230-0045, Japan.
• Institute for Theoretical Chemistry, University of Vienna, Währingerstraße 17, A-1090 Wien, Austria.
• Department of Biomolecular Engineering, University of California, Santa Cruz, 1156 High Street, Santa Cruz, California 95064, USA.
• Center for Biomolecular Science and Engineering, Engineering 2, Suite 501, Mail Stop CBSE/ITI, University of California, Santa Cruz, California 95064, USA.
• Department of Biological Chemistry & Molecular Pharmacology, Harvard Medical School, 240 Longwood Avenue, Boston, Massachusetts 02115, USA.
• Department of Genetics, Facultat de Biologia, Universitat de Barcelona, Av Diagonal, 645, 08028, Barcelona, Catalonia, Spain.
• Bioinformatics Institute, 30 Biopolis Street, #07-01 Matrix, Singapore, 138671, Singapore.
• Bioinformatics Group, Department of Computer Science, University of Leipzig, Härtelstraße 16-18, D-04107 Leipzig, Germany.
• Fraunhofer Institut für Zelltherapie und Immunologie - IZI, Deutscher Platz 5e, D-04103 Leipzig, Germany.
• Interdisciplinary Center of Bioinformatics, University of Leipzig, Härtelstraße 16-18, D-04107 Leipzig, Germany.
• Department of Genetics, Stanford University School of Medicine, Stanford, California 94305, USA.
• Genome Institute of Singapore, 60 Biopolis Street, Singapore 138672, Singapore.
• Laboratory for Computational Genomics, Washington University, Campus Box 1045, Saint Louis, Missouri 63130, USA.
• Department of Mathematics and Computer Science, University of California, Berkeley, California 94720, USA.
• Spanish National Cancer Research Centre, CNIO, Madrid, Spain, and Biosapiences NoE.
• Department of Epidemiology and Public Health, Imperial College, St Mary’s Campus, Norfolk Place, London W2 1PG, UK.
• Department of Applied Science & Technology, University of California, Berkeley, California 94720, USA.
• Genome Science Laboratory, Discovery and Research Institute, RIKEN Wako Institute, 2-1 Hirosawa, Wako, Saitama, 351-0198, Japan.
• Department of Genetics, Yale University School of Medicine, 333 Cedar Street, New Haven, Connecticut 06510, USA.
• Department of Pediatrics, University of Massachusetts Medical School, 55 Lake Avenue, North Worcester, Massachusetts 01605, USA.
• Department of Statistics, University of California, Berkeley, California 94720, USA.
• Institute for Molecular Bioscience, University of Queensland, St. Lucia, QLD 4072, Australia.
• The Santa Fe Institute, 1399 Hyde Park Road, Santa Fe, New Mexico 87501, USA.
• Department of Computer Science, Yale University, PO Box 208114, New Haven, Connecticut 06520-8114, USA.
• Program in Computational Biology & Bioinformatics, Yale University, PO Box 208114, New Haven, Connecticut 06520-8114, USA.
• NIH Intramural Sequencing Center, National Human Genome Research Institute, National Institutes of Health, Bethesda, Maryland 20892, USA.
• Department of Computer Science and Engineering, The Pennsylvania State University, University Park, Pennsylvania 16802, USA.
• Center for Comparative Genomics and Bioinformatics, Huck Institutes for Life Sciences, The Pennsylvania State University, University Park, Pennsylvania 16802, USA.
• Division of Extramural Research, National Human Genome Research Institute, National Institute of Health, 5635 Fishers Lane, Suite 4076, Bethesda, Maryland 20892-9305, USA.
• Office of the Director, National Human Genome Research Institute, 31 Center Drive, Suite 4B09, Bethesda, Maryland 20892-2152, USA.
• Department of Computer Science, Stanford University, Stanford, California 94305, USA.
• Department of Biostatistics and Medical Informatics, University of Wisconsin-Madison, 6720 MSC, 1300 University Ave, Madison, Wisconsin 53706, USA.
• Department of Zoology and Animal Biology, Faculty of Sciences, University of Geneva, Switzerland.
• Department of Statistics, Stanford University, Stanford, California 94305, USA.
• Department of Bioengineering, University of California, Berkeley, California 94720-1762, USA.
• National Center for Biotechnology Information, National Institutes of Health, Bethesda, Maryland 20894, USA.
• Department of Biochemistry and Molecular Biology, Huck Institutes of Life Sciences, The Pennsylvania State University, University Park, Pennsylvania 16802, USA.
• Howard Hughes Medical Institute, University of California, Santa Cruz, California 95064, USA.
• Department of Biology, The Pennsylvania State University, University Park, Pennsylvania 16802, USA.
• Department of Pathology, Stanford University School of Medicine, Stanford, California 94305, USA.
• Human Genome Sequencing Center and Department of Molecular and Human Genetics, Baylor College of Medicine, Houston, Texas 77030, USA.
• Genome Sequencing Center, Washington University School of Medicine, Campus Box 8501, 4444 Forest Park Avenue, Saint Louis, Missouri 63108, USA.
• Broad Institute of Harvard University and Massachusetts Institute of Technology, 320 Charles Street, Cambridge, Massachusetts 02141, USA.
• Whitehead Institute for Biomedical Research, 9 Cambridge Center, Cambridge, Massachusetts 02142, USA.
• Children’s Hospital Oakland Research Institute, BACPAC Resources, 747 52nd Street, Oakland, California 94609, USA.
• Ludwig Institute for Cancer Research, 9500 Gilman Drive, La Jolla, California 92093-0653, USA.
• The Linnaeus Centre for Bioinformatics, Uppsala University, BMC, Box 598, SE-75124 Uppsala, Sweden.
• Department of Pharmacology and the Genome Center, University of California, Davis, California 95616, USA.
• Institute for Cellular & Molecular Biology, The University of Texas at Austin, 1 University Station A4800, Austin, Texas 78712, USA.
• NimbleGen Systems, Inc., 1 Science Court, Madison, Wisconsin 53711, USA.
• University of Wisconsin Medical School, Madison, Wisconsin 53706, USA.
• Department of Genetics and Pathology, Rudbeck Laboratory, Uppsala University, SE-75185 Uppsala, Sweden.
• University of California, San Diego School of Medicine, 9500 Gilman Drive, La Jolla, California 92093, USA.
• Genes and Disease Program, Center for Genomic Regulation, C/Dr. Aiguader 88, Barcelona Biomedical Research Park Building, 08003 Barcelona, Catalonia, Spain.
• Broad Institute of MIT and Harvard, 7 Cambridge Center, Cambridge, Massachusetts 02142, USA.
• Center for Statistical Genetics, Department of Biostatistics, SPH II, 1420 Washington Heights, Ann Arbor, Michigan 48109-2029, USA.
• Department of Statistics, University of Oxford, Oxford, UK.
• Universitat Pompeu Fabra, C/Dr. Aiguader 88, Barcelona Biomedical Research Park Building, 08003 Barcelona, Catalonia, Spain.