The international ENCODE Project (https://www.encodeproject.org/) was established to define and catalog the functional DNA elements in our genome. Operationally, a functional DNA element was defined as a discrete genome segment that makes a defined product (protein or noncoding RNA) or displays a reproducible biochemical signature (such as a protein-binding capacity or a specific chromatin structure). Begun in September 2003 as a pilot project, the ENCODE Project provided a full report in 2012 in a series of papers in Nature and Genome Research.
The ENCODE Project involved global (genome-wide) assays of functional sequences as interpreted by analyzing multiple characteristics that affect gene expression and are dependent on cell type, including: transcript profiles; histone modification states; transcription factor-binding states; chromatin conformation states; and DNA methylation states (Figure 1)—we describe the different types of assay below. The assays were carried out on various cell types from a dedicated resource of over 100 different human cell types, including immortal stem cells, immortalized cell lines, and primary (nontrans formed) cells.
Fig1. Principal methods and data flow in the ENCODE Project. (A) Major methods used to detect functional DNA elements (gray boxes), represented on an idealized model of mammalian chromatin and a mammalian gene. 5C, ChiA-PET and Hi-C are methods used to study chromatin conformation and are described in Box 10.1. The DNAse-Seq, FAIRE-Seq and ChIP-Seq methods seek to identify regions of chromatin where the DNA is readily accessible (and sensitive to DNAse I), and are described in Section 10.1. ChIP-Seq helps define binding sites for DNA-associated proteins (Box 9.3). Whole genome bisulfite sequencing (WGBS), reduced representation bisulfite sequencing (RRBS) and methyl arrays are used to survey DNA methylation profiles across the genome. Methods focused on RNA include RNA-Seq, RNA immunoprecipitation (RIP; using a specific antibody to immunoprecipitate a protein that binds to RNA) and cross-linking immunoprecipitation (CLIP; similar to RIP but with an extra cross-linking step). (B) The overall data flow from the production groups after reproducibility assessment to the Data Coordinating Center (DCC) for public access and to other public databases. (Adapted from The ENCODE Project Consortium [2011] PLoS Biol 9:e1001046; PMID 21526222; and associated website at https://www.encodeproject.org/.) GEO, Gene Expression Omnibus; SRA, Sequence Read Archive; UCSC, University of California at Santa Cruz.
For many assays, only a small subset of the cell types was used, the different cell types being classified into different tiers according to priority. Tier 1 (highest priority) included the H1 human embryonic stem cell line and two well-studied immortalized cell lines: GM12878, a B-lymphoblastoid cell line, and K562, an erythroleukemic cell line. Tier 2 cell types comprised the HepG2 hepatoblastoma cell line, HeLa cells, and primary (non transformed) human umbilical vein endothelial cells. Tier 3 was represented by more than 100 cell types, including different cell lines and cultured cells, plus cells from some primary tissues.
The ENCODE Project comprised a variety of experimental assays plus annotation efforts. The headings below give a brief description of the types of approach used, and also the conclusions that were drawn.
Annotation
An important component of the ENCODE Project was devoted to annotation—descriptive commentary of the characteristics of genes and their products, and of other functional DNA sequences. It encompassed protein-coding genes, pseudogenes, and noncoding transcribed loci across the genome, and was also required in cataloging the products of transcription including alternative splice forms. Some automated annotation, such as by using the Ensembl gene build pipeline, was carried out, but much of the annotation was carried out manually by the HAVANA (human and vertebrate annotation) team at the Wellcome Trust Sanger Institute. GENCODE, the result of ENCODE gene annotation, provides a comprehensive catalog of transcripts and gene models (which continues to be regularly updated), and is available at http://www.gencodegenes.org/.
Transcript analysis
Different methods were used to analyze RNA transcripts as described in Table 1. After surveying 15 different human cell lines, the ENCODE consortium concluded that about 75% of the human genome is transcribed in at least one of the cell types studied (but the percentage is significantly lower in any one cell type). RNA transcripts from both strands were found to be common (both within genes and in intergenic regions close to genes), and a protein-coding gene is now known to produce, on average, about six to seven different RNA transcripts, including some noncoding transcripts—see Figure 2 for a graphical display of transcripts from the CFTR (cystic fibrosis transmembrane regulator) gene. Some genes are known to produce functional noncoding transcripts as well as functional mRNAs, but in general the functional status of noncoding transcripts is underexplored.
Fig2. An example to illustrate using the Ensembl genome browser. Here the October 2016 version of Ensembl was queried with the human CFTR gene (the cystic fibrosis transmembrane regulator gene spans nucleotides 117,465,784–117,715,971 on chromosome 7). The two frames with the title “Genes” show exon–intron organizations of the different transcripts from the two DNA strands: the upper frame shows the CFTR sense transcripts, and the frame below the “Contigs” bar (which gives GenBank accession numbers for indicated DNA sequence contigs) shows CFTR antisense transcripts, plus transcripts from a neighboring, partially overlapping, protein-coding gene, CTTNBP2 (transcribed from the opposite DNA strand). In each case, exons are represented by short vertical bars that are connected by introns (flattened chevrons); protein-coding transcripts are represented by red- and gold-colored lines, noncoding transcripts by blue lines (according to the gene legend at bottom). There are five CFTR protein-coding transcripts (two full-length isoforms and three smaller isoforms), six noncoding sense transcripts, and two antisense transcripts. The “Regulatory Build” shows colored vertical bars representing the positions of indicated regulatory sequences. Clicking over individual items brings up additional information, as shown here by clicking on the CTTNBP2-011 transcript at bottom right (clicking on underlined items in blue font allows access to further information, often presented in further graphical frames).
Table1. PRINCIPAL TRANSCRIPTION ANALYSES CARRIED OUT IN THE ENCODE PROJECT
(Note: high-resolution spatial expression of transcripts, which is most efficiently carried out in the miniaturized tissues of the embryo, was not carried out by the ENCODE Project. Various other projects have focused on large-scale in situ hybridization studies against primarily mouse embryonic tissue sections, and gene expression data are deposited in resources such as the e-Mouse Atlas at http://www.emouseatlas.org/ and the Allen brain atlas at http://mouse.brain-map.org/.)
DNA methylation and chromatin chemical modification analysis
Both the DNA of our cells and the histone proteins of chromatin are subject to patterns of chemical modification that vary between different cell types; in each case, as detailed in Chapter 10, the chemical modification is a form of epigenetic control of gene expression. Different types of chemical modification occur in the side chains of N-terminal residues of histone proteins—including methylation of certain lysines (K) and arginines (R), and acetylation of certain lysines—but chemical modification of DNA is limited to methylation of certain cytosines.
The patterns of DNA methylation were tracked by comparing DNA samples treated with sodium bisulfite with untreated DNA samples (the bisulfite treatment converts unmethylated cytosines to uracil, but methylated cytosines are not affected). In addition to expensive whole-genome bisulfite sequencing (WGBS), more efficient and high-throughput reduced representation bisulfite sequencing (RRBS) allowed analysis of genome-wide methylation profiles at single nucleotide resolution in gene-rich sequences. (RRBS combines restriction enzymes and bisulfite sequencing to enrich for CpG-rich areas of the genome, and quantitative DNA methylation profiles were obtained for 1.2 million CpGs, on average, in each of 82 cell lines and tissue cells.) Methyl450K bead chips allowed assay of 450,000 methylation sites selected from genes, CpG islands, and enhancers across the genome. Two principal findings were: (1) 96% of CpGs exhibited differential methylation in at least one of the cell types assayed; and (2) the most variably methylated CpGs are found more often in gene bodies (comprising all exons and introns) and intergenic regions rather than in promoters and upstream regulatory regions.
To assay chemical modification of histone side chains, the ChIP-Seq method, combining chromatin immunoprecipitation with DNA sequencing, was used (see Box 9.3 for the method). Various antibodies specific for the histone variants were used. Up to 11 types of histone modification were sampled in 46 cell types, and eight selected variants were assayed across all cell types in tier 1 and tier 2 categories. As expected, the global patterns of histone modification were found to be highly variable across different cell types.
Transcription factor binding, regulatory sequences, and chromatin accessibility
ChIP-Seq was also used to map the 6–25-nucleotide-long genomic positions bound by many different transcription factors across the genomes of 72 different types of living cell. The data, together with that of binding sites for some other DNA-binding proteins (including components of RNA polymerases), are integrated with the ENCODE data for histone modification and nucleosome occupancy profiles, and are available at the Factorbook website at http://www.factorbook.org/human/chipseq/tf/.
Regulatory DNA sequences, including the sequences recognized by transcription factors, must be in regions of chromatin that are accessible to protein factors. Accessible chromatin regions are more sensitive to cleavage by deoxyribonuclease I (DNase I) and the ENCODE Project also mapped nearly 3 million unique, non-overlapping DNase I hypersensitive sites by a technique known as DNase-Seq in 125 cell types. In addition, regulatory elements were isolated by the FAIRE-Seq method: FAIRE (formaldehyde assisted isolation of regulatory elements) isolates nucleosome-depleted genomic regions by exploiting the difference in cross-linking efficiency between nucleosomes (high) and sequence-specific regulatory factors (low). The method involves cross-linking, phenol extraction, and then sequencing the DNA fragments in the aqueous phase.
Chromatin conformation
Cis-acting regulatory elements, such as enhancers, can control genes from long distances (tens to hundreds of kb) on the same DNA molecule through DNA looping: the enhancer and its bound regulatory proteins are brought into close proximity to the gene’s promoter and nearby regulatory elements. To map these long-range interactions, the 5C method, an enhanced version of chromosome conformation capture (3C), was used in the ENCODE Project. We detail these methods in Chapter 10 when we look at aspects of gene regulation.
Overall conclusions and the functional fraction of the genome
Among the big surprises of the ENCODE Project was the finding of pervasive transcription of the human genome. Not only is a large fraction of the genome transcribed (with about 75% of the sequence in the genome being transcribed in one or more of the sampled cell lines), both DNA strands are often transcribed, and the great majority of the transcripts are noncoding RNAs. The analyses also showed evidence for an initial set of nearly 400,000 regions with features resembling enhancers and over 70,000 promoter-like regions. A big surprise from various transcript analyses was that bidirectional transcription commonly proceeds from both promoters and enhancers.
A headline conclusion was that at least 80% of the human genome participates in at least one biochemical RNA and/or chromatin-associated event in at least one cell type. That was interpreted in commentary articles in Nature and Science to mean that most of our genome is functionally important after all, and that it was time to abandon the concept of “junk DNA.” (Previous studies based largely on species comparisons had estimated that a small fraction, perhaps just 3–8%, of the euchromatic genome was functionally significant; much of the rest was sometimes labeled as junk DNA.) However, the idea that the great majority of the genome is functionally important, and that junk DNA is now an outmoded concept, has been strongly resisted by evolutionary geneticists.
