compared with chromosomal duplication con-
`tent (R2 ⫽ 0.16). The correlation was due to
`intrachromosomal duplications (fig. S5; R2 ⫽
`0.20; P ⫽ 0.04; F test) and was absent for
`interchromosomal duplications (R2 ⫽ 0.002).
`The three most gene-rich chromosomes showed
`high levels of duplication, and the seven most
`gene-poor chromosomes were among the least
`duplicated chromosomes.
`To determine what role recent segmental
`duplications have played in current gene evo-
`lution, we characterized the gene content in
`our filtered set of duplicated genomic se-
`quence. We analyzed a highly curated set of
`13,351 mRNAs assigned to the human ge-
`nome assembly (RefSeq, www.ncbi.nlm.nih-
`.gov/LocusLink/refseq.html). We partitioned
`exons from each gene into a unique or dupli-
`cated sequence on the basis of their map
`position (⬎90% sequence identity). We iden-
`tified a total of 7777 exons as being tran-
`scribed from recently duplicated sequence,
`corresponding to 6.1% of all RefSeq exons
`(128,467). This is slightly greater than the
`genomic representation of segmental duplica-
`tion (5.2%), which confirms that gene-poor
`regions have not been preferentially duplicat-
`ed. In many cases, a complete complement of
`exons was not duplicated. These incomplete
`duplicated genes were often found adjacent to
`other duplicated cassettes that originated
`from elsewhere in the genome. By comparing
`our data with human expressed sequence tag
`databases, we found evidence for “chimeric”
`or fusion transcripts that emerged from the
`physical juxtaposition of incomplete segmen-
`tal duplications. Although the mechanism for
`recent segmental duplications is not under-
`stood, the existing data suggest the process
`may play a role in exon shuffling associated
`with expanding protein diversity. A complete
`list of all genes with one more exons within
`duplicated genomic sequence is available (8).
`To further assess whether specific kinds of
`genes or biological processes have been prefer-
`entially duplicated, we compared all RefSeq
`mRNAs on the basis of their INTERPRO pro-
`tein domain classification (Table 2) (table S7)
`(23). In this analysis, we considered a gene
`duplicated only if all its exons were contained
`within a duplicated genomic region. Our anal-
`ysis suggests a nonrandom distribution of seg-
`mental duplications within the proteome. Genes
`associated with immunity and defense (natural
`killer receptors, defensins, interferons, serine
`proteases, cytokines), membrane surface inter-
`actions (galectins, HLA, lipocalins, carcinoem-
`bryonic antigens), drug detoxification (cyto-
`chrome P450), and growth/development (soma-
`totropins, chorionic gonadotropins, pregnancy-
`specific
`glycoproteins) were
`particularly
`enriched. It should be emphasized that our gene
`analysis is restricted to genomic segments that
`show ⱖ90% sequence identity. On the basis of
`neutral expectation of divergence, this corre-
`
`R E P O R T S
`
`sponds to duplications that have emerged over
`the last ⬃40 million years of human evolution
`(24). Gene duplication followed by functional
`specialization has long been considered a major
`evolutionary force for gene innovation (25).
`Therefore, these genes embedded within recent
`genomic duplications may be considered excel-
`lent candidates for adaptations specific to pri-
`mate evolution.
`
`1.
`
`References and Notes
`International Human Genome Sequencing Consor-
`tium, Nature 409, 860 (2001).
`2. J. A. Bailey, A. M. Yavor, H. F. Massa, B. J. Trask, E. E.
`Eichler, Genome Res. 11, 1005 (2001).
`3. J. C. Venter et al., Science 291, 1304 (2001).
`4. S. Ohno, U. Wolf, N. Atkin, Hereditas 59, 169 (1968).
`5. E. E. Eichler, Trends Genet. 17, 661 (2001).
`6. P. Stankiewicz, J. R. Lupski, Trends Genet. 18, 74
`(2002).
`7. E. E. Eichler, Genome Res. 11, 653 (2001).
`8. See supporting data on Science Online.
`9. V. E. Cheung et al., Nature 409, 953 (2001).
`10. The sequence and underlying test statistics for all
`duplicated regions of the genome are available at
`http://humanparalogy.cwru.edu/SDD. WGAC com-
`parisons of the human genome assembly (UCSC,
`August freeze, 2001; http://genome.ucsc.edu) were
`done as described (2). The WSSD-filtered set of
`WGAC duplications can be interactively searched
`(http://humanparalogy.cwru.edu/SDD). This includes
`extracted sequence files, the actual alignments, the
`location of the alignments within the assembly, and
`whole-chromosomal views comparing WGAC and
`WSSD duplication patterns. An updated WSSD based
`on the analysis of 39,298 clones from April 2002,
`detecting an additional 36 Mb of duplicated se-
`quence, is also available.
`
`11. T. H. Shaikh et al., Hum. Mol. Genet. 9, 489 (2000).
`12. J. A. Bailey et al., Am. J. Hum. Genet. 70, 83 (2002).
`13. L. Edelmann, R. K. Pandita, B. E. Morrow, Am. J. Hum.
`Genet. 64, 1076 (1999).
`14. R. Mazzarella, D. Schlessinger, Genome Res. 8, 1007
`(1998).
`15. J. R. Lupski, Trends Genet. 14, 417 (1998).
`16. S. T. Sherry et al., Nucleic Acids Res. 29, 308 (2001).
`17. K. Chen et al., Nature Genet. 17, 154 (1997).
`18. S. L. Christian, J. A. Fantes, S. K. Mewborn, B. Huang,
`D. H. Ledbetter, Hum. Mol. Genet. 8, 1025 (1999).
`19. D. E. Jenne et al., Am. J. Hum. Genet. 69, 516 (2001).
`20. T. Kuroda-Kawaguchi et al., Nature Genet. 29, 279
`(2001).
`21. H. C. Mefford, B. J. Trask, Nature Rev. Genet. 3, 91
`(2002).
`22. J. Guy et al., Hum. Mol. Genet. 9, 2029 (2000).
`23. M. Ashburner et al., Nature Genet. 25, 25 (2000).
`24. W. Li, Molecular Evolution (Sinauer Associates, Sun-
`derland, MA, 1997).
`25. S. Ohno, Evolution by Gene Duplication (Springer-
`Verlag, Berlin, 1970).
`26. We thank L. Christ, M. Eichler, and U. Neuss for
`technical assistance, and H. Willard, J. Nadeau, T.
`Hassold, D. Locke, and J. Horvath for helpful com-
`ments. Supported by NIH grants GM58815 and
`HG002318 and U.S. Department of Energy grant
`ER62862 (E.E.E.), NIH Career Development Program
`in Genomic Epidemiology of Cancer (CA094816)
`and Medical Scientist Training Grant ( J.A.B.), the
`W. M. Keck Foundation, and the Charles B. Wang
`Foundation.
`
`Supporting Online Material
`www.sciencemag.org/cgi/content/full/297/5583/1003/
`DC1
`Materials and Methods
`Tables S1 to S7
`Figs. S1 to S5
`
`20 March 2002; accepted 7 June 2002
`
`Predictive Identification of
`Exonic Splicing Enhancers in
`Human Genes
`William G. Fairbrother,1,2* Ru-Fang Yeh,1* Phillip A. Sharp,1,2
`Christopher B. Burge1†
`
`Specific short oligonucleotide sequences that enhance pre-mRNA splicing when
`present in exons, termed exonic splicing enhancers (ESEs), play important roles
`in constitutive and alternative splicing. A computational method, RESCUE-ESE,
`was developed that predicts which sequences have ESE activity by statistical
`analysis of exon-intron and splice site composition. When large data sets of
`human gene sequences were used, this method identified 10 predicted ESE
`motifs. Representatives of all 10 motifs were found to display enhancer activity
`in vivo, whereas point mutants of these sequences exhibited sharply reduced
`activity. The motifs identified enable prediction of the splicing phenotypes of
`exonic mutations in human genes.
`
`Human genes are generally transcribed as
`much longer precursors, typically tens of ki-
`lobases in length, from which large introns
`
`1Department of Biology, 2Center for Cancer Research,
`Massachusetts Institute of Technology, Cambridge,
`MA 02139, USA.
`
`*These authors contributed equally to this work.
`†To whom correspondence should be addressed. E-
`mail: cburge@mit.edu
`
`must be precisely removed and flanking ex-
`ons precisely ligated to create the mRNA that
`will direct protein synthesis. Sequences
`around the splice junctions—the 5⬘ and 3⬘
`splice sites (5⬘ss and 3⬘ss)—are clearly im-
`portant for splice site recognition. However,
`these signals appear to contain only about
`half of the information required for exon and
`intron recognition in human transcripts (1).
`The sequence or structure context in the vi-
`
`www.sciencemag.org SCIENCE VOL 297 9 AUGUST 2002
`
`1007
`
`Triplet EX1069
`
`1
`
`

`

`cinity of the 5⬘ss and 3⬘ss motifs is known to
`play an important role in splice site recogni-
`tion (2– 4). ESE sequences, which enhance
`
`splicing at nearby sites (5), are an important
`component of this context.
`Exonic enhancers have been identified
`
`R E P O R T S
`
`Fig. 1. Schematic of RESCUE-ESE approach. Exon-intron structures of human genes are derived by
`spliced alignment of cDNAs to the assembled genomic sequence, and splice sites are scored as
`described (17). Values of ⌬EI (scaled difference in frequency between exons and introns) and ⌬WS
`(scaled difference in frequency between weak and strong exons) are calculated as described for
`each of the 4096 possible hexanucleotides (17). Each hexamer is then represented by a colored
`letter at the point (⌬EI, ⌬WS) in the scatterplot. The letters are chosen to reflect the base
`composition of the hexamer according to IUPAC nomenclature (e.g., hexamers containing only A
`and G are represented by the letter “r”). Hexamers containing homonucleotide runs of three or
`more bases (e.g., AAA) are represented by capital letters, all other hexamers by lowercase letters.
`Each letter is colored proportional to the relative content of A (red), C (green), G (blue), and T
`(black) of the hexamer. Hexamers (6mers) satisfying ⌬EI ⬎ 2.5 and ⌬WS ⬎ 2.5 (upper right portion
`of first quadrant) are predicted to have ESE activity. As a test of ESE activity, a 19-base “extended
`exemplar” sequence containing the hexamer in its natural context in a weak exon is chosen and
`inserted into the SXN splicing reporter construct as indicated. SXN is a ␤-globin– derived minigene
`with deleted translation start codon. A point mutant predicted to disrupt ESE activity is also chosen,
`generally the single-base mutant that is farthest to the left and below the predicted ESE hexamer
`in the scatterplot. Transient transfection of the reporter construct followed by quantitative RT-PCR
`with flanking primers is used to assay inclusion of the test exon for the candidate ESE and its
`mutant.
`
`through the analysis of disease alleles (6), by
`site-directed mutagenesis of minigene con-
`structs, and by protocols based on SELEX
`(Systematic Evolution of Ligands by EXpo-
`nential enrichment)
`to identify sequences
`with enhancer activity from a pool of random
`sequences (7–11). These methods initially
`characterized ESEs as purine-rich sequences,
`but additional classes of AC-rich motifs and
`pyrimidine-rich motifs have since emerged
`(7, 10).
`Our strategy for identifying human ESE
`sequences was to first develop a statistical/
`computational method to predict the ESE activ-
`ity of oligonucleotide sequence motifs, to apply
`this method to large data sets of human genom-
`ic sequences, and then to test representatives of
`each predicted motif by means of an in vivo
`splicing assay. At the heart of this approach is a
`sequence
`analysis method that we
`call
`RESCUE (Relative Enhancer and Silencer
`Classification by Unanimous Enrichment).
`RESCUE identifies the set of oligonucleotide
`motifs that enhance or repress a particular bio-
`chemical process; it consists of four steps: (i)
`Identify two or more statistical “attributes” that
`should be manifested by sequences that en-
`hance (or, alternatively, repress) the biochemi-
`cal activity of interest. (ii) Use a statistical
`power calculation to determine an oligonucleo-
`tide “word” size k appropriate for the amount of
`data available. Then represent all possible oli-
`gonucleotides of size k by points in a multidi-
`mensional space, the axes of which represent
`the attributes chosen in the previous step. (iii)
`Define a region in this space corresponding to
`“unanimous enrichment” (i.e.,
`significantly
`high values of all of the chosen attributes) and
`identify clusters of similar sequences that fall in
`this region. (iv) Align the sequences in each
`cluster to produce motifs, and test representa-
`tive sequence(s) from each motif and appropri-
`ate point mutants with the use of a suitable
`functional assay.
`A large body of work suggests that ESEs
`are located in the general vicinity of splice
`sites (12). Unlike transcriptional enhancers,
`ESEs function in a strongly position-depen-
`dent manner,
`enhancing splicing when
`present downstream of a 3⬘ss and/or upstream
`of a 5⬘ss (13), but often repressing splicing
`when present in intronic locations (14, 15).
`These observations suggest that, as one at-
`tribute, ESE sequences should be strongly
`selected for in constitutively spliced exons
`and generally avoided in intronic sequences
`near splice sites.
`Moreover, ESEs can compensate for the
`presence of “weak” (nonconsensus) 5⬘ or 3⬘
`splice signals in exons, and strengthening of
`the splice sites of an enhancer-dependent
`exon generally eliminates enhancer depen-
`dence (16). Therefore, we conjecture that ex-
`ons with nonconsensus splice sites (“weak
`exons”) are under much stronger selective
`
`1008
`
`9 AUGUST 2002 VOL 297 SCIENCE www.sciencemag.org
`
`2
`
`

`

`pressure to retain ESEs than are exons with
`consensus splice sites (“strong exons”), re-
`sulting in a significantly higher frequency of
`ESEs in weak exons than in strong exons.
`Available full-length cDNA sequences
`were aligned to the assembled human ge-
`nome by means of the spliced alignment al-
`gorithm that
`is part of the “Genoa” gene
`annotation script (17). Reliable full-length
`alignments were obtained with this approach
`for 4817 human genes containing 31,463 in-
`trons and 28,933 internal exons. Position-
`specific log-odds score matrices were then
`used to score the 5⬘ss and 3⬘ss of these exons,
`and the distributions of 5⬘ss and 3⬘ss scores
`were used to partition exons into categories
`on the basis of the strength of their splice
`sites: “weak 5⬘ exons” (bottom 25% of 5⬘ss
`scores), “strong 5⬘ exons” (top 25% of 5⬘ss
`scores), with “weak 3⬘ exons” and “strong 3⬘
`exons” defined analogously.
`Application of the RESCUE-ESE method
`to this set of human genes is illustrated in Fig.
`1. A power calculation dictated the use of a
`word size of six nucleotides, which is com-
`parable in size to the binding sites of many
`known RNA binding factors (17). In step
`two, each of the 4096 oligonucleotides of
`length six was assigned two scores: ⌬EI, the
`scaled difference between the frequency of
`occurrence of the hexamer in exons and the
`frequency of occurrence near splice sites in
`introns (scaled in standard deviation units);
`and ⌬5WS, the scaled difference between the
`frequency of occurrence of the hexamer in
`weak 5⬘ exons and its frequency in strong 5⬘
`exons (SD units), with ⌬3WS defined analo-
`gously for weak 3⬘ exons versus strong 3⬘
`exons. Each hexamer was then represented
`by a point in the plane with coordinates (⌬EI,
`⌬5WS) for identification of sequences that
`enhance 5⬘ss recognition (5⬘ESEs) (Fig. 2A).
`Alternatively, each hexamer was represented
`by the point (⌬EI, ⌬3WS) for identification
`of sequences that enhance 3⬘ss recognition
`(3⬘ESEs) (Fig. 2B). A statistical significance
`threshold of 2.5 standard deviations above
`the mean (corresponding to a P value of
`⬃0.01) was then applied to each axis inde-
`pendently; that is, any hexamer for which
`both ⌬EI ⬎ 2.5 and ⌬5WS ⬎ 2.5 is predicted
`to be a 5⬘ESE, and any hexamer with both
`⌬EI ⬎ 2.5 and ⌬3WS ⬎ 2.5 is predicted to be
`a 3⬘ESE (hexamers in the upper right portion
`of the first quadrant in the scatterplots). The
`requirement
`that
`each hexamer
`exceed
`thresholds in two separate dimensions, both
`with P ⬃ 0.01, represents essentially a Bon-
`ferroni-type correction for multiple compari-
`sons: Because 4096 different tests are being
`performed, the combined P value is set to
`⬃(0.01)2 ⫽ 10⫺4, giving an expectation of
`less than one false positive hexamer.
`These criteria identified 103 different hex-
`amers as candidate 5⬘ESEs and 198 hexamers
`
`R E P O R T S
`as candidate 3⬘ESEs. These two sets overlap
`fairly extensively, with 63 of the 103 predict-
`ed 5⬘ESEs also contained in the set of pre-
`dicted 3⬘ESEs, which suggests that many en-
`hancers may be capable of acting at both
`splice sites [e.g., (13)]. The total number of
`hexamers predicted to display either 5⬘ or 3⬘
`ESE activity was 238 out of the 4096 possible
`hexamers, about 6% of the total, consistent
`with the notion that ESEs are quite common.
`In step three of the RESCUE procedure,
`predicted 5⬘ESE and 3⬘ESE hexamers were
`clustered on the basis of sequence similarity,
`and the hexamers in each cluster were multiply
`aligned using CLUSTALW (18) to identify
`candidate enhancer motifs (fig. S3) (17). This
`procedure yielded a total of five 5⬘ESE motifs
`(Fig. 2A) and eight 3⬘ESE motifs (Fig. 2B).
`Three of the five 5⬘ESE motifs—5A, 5B, and
`5C—are significantly similar to 3⬘ESE motifs
`3G, 3A, and 3D, respectively, so the three pairs
`5A/3G, 5B/3A, and 5C/3D were considered to
`
`represent just three distinct classes, each com-
`prising the union of the pair of similar hexamer
`clusters (17). The total number of distinct can-
`didate enhancer motifs identified by RESCUE-
`ESE was therefore 10.
`In the final step of the RESCUE proce-
`dure, representatives of these candidate en-
`hancer motifs were tested for ESE activity in
`a splicing reporter construct. For each cluster
`of predicted ESEs, a representative hexamer
`was chosen—referred to as the “exemplar” of
`the class. To place each exemplar hexamer in
`its natural context, we screened our human
`spliced gene database for an occurrence of
`each exemplar hexamer in a weak 5⬘ exon
`(bottom 10% of 5⬘ss scores) or weak 3⬘ exon
`(bottom 10% of 3⬘ss scores), as appropriate.
`A slightly longer region of sequence centered
`on the exemplar—referred to as the “extend-
`ed exemplar”—was then chosen from this
`exon and inserted into the reporter construct
`described below. The extended exemplar
`
`Fig. 2. RESCUE-ESE prediction of 5⬘ and 3⬘ ESEs in human genes. (A) Scatterplot for prediction of
`5⬘ESE activity. Hexamers are represented by colored letters as described in Fig. 1. Simplified
`dendrogram shows clustering of 5⬘ESE hexamers (total of 103 hexamers with ⌬EI ⬎ 2.5 and
`⌬5WS ⬎ 2.5) into five clusters of four or more hexamers. (B) Scatterplot for prediction of 3⬘ESE
`activity. Simplified dendrogram shows clustering of 3⬘ESE hexamers (total of 198 hexamers with
`⌬EI ⬎ 2.5 and ⌬3WS ⬎ 2.5) into eight clusters of four or more hexamers. Complete dendrograms
`of all hexamers are shown in fig. S3. The aligned sequences in each cluster are represented as
`Pictograms (http://genes.mit.edu/pictogram.html). Cluster labels (e.g., 3B, 5A/3G) are listed to the
`right of each Pictogram, with the total number of hexamers in the cluster indicated in parentheses.
`Clustering and alignment were performed as described (17).
`
`www.sciencemag.org SCIENCE VOL 297 9 AUGUST 2002
`
`1009
`
`3
`
`

`

`sequences comprise the 19-base region ex-
`tending from six bases 5⬘ of the exemplar
`hexamer to seven bases 3⬘ of the exemplar
`hexamer (Fig. 1).
`The splicing enhancer activity of each ex-
`tended exemplar sequence was then assessed by
`measuring its ability to “rescue” splicing of
`exon 2 of the reporter construct, pSXN (7).
`SXN exon 2 is only 32 bases long, including the
`19-base insert. Previously, this exon was ob-
`served to be predominantly skipped for most
`random insert sequences tested. This failure to
`be included is reversed when the exon is length-
`ened, when a splicing enhancer is present, or
`when the 5⬘ss, the branch point, or the poly-
`pyrimidine tract is improved (fig. S1) (19–21).
`Because strengthening either the 5⬘ss or 3⬘ss
`consensus sequence of SXN exon 2 or inserting
`an ESE causes exon inclusion, we reasoned that
`this exon would be a suitable reporter system
`for testing the activity of candidate 5⬘ESEs as
`well as candidate 3⬘ESEs.
`It was of particular interest to assess the
`ability of the RESCUE approach to predict
`ESE-disrupting mutations. Therefore,
`for
`each exemplar hexamer, a single-base mutant
`was chosen that was predicted to lack en-
`hancer activity [i.e., did not fall in the ex-
`treme upper right (“unanimous enrichment”)
`region of the scatterplot]. Typically, the sin-
`gle-point mutant farthest “southwest of ” (to
`the left and below) the exemplar in the scat-
`terplot was chosen. A “mutant” extended ex-
`emplar sequence containing just this single
`base change was then generated for each
`extended exemplar and inserted into the same
`cloning site in the SXN minigene. Constructs
`containing the extended exemplars and mu-
`tants were transiently transfected into HeLa
`cells, and the splicing phenotype was assayed
`by quantitative reverse-transcription poly-
`merase chain reaction (RT-PCR) (see fig. S2
`for protocol and quantitation curves).
`An initial set of experiments evaluated
`the robustness of the approach with respect
`to differences in the local context of the
`exemplar hexamer. For this purpose we
`focused on a
`representative hexamer,
`GAAGAA, chosen from the large purine-
`rich 5C/3D cluster of predicted enhancers
`(Fig. 2). The consensus sequences for these
`clusters and the chosen exemplar are simi-
`lar to the classical “GARGAR” enhancer
`(R represents either purine nucleotide, A or
`G). Occurrences of GAAGAA were identi-
`fied in three exons with weak splice sites,
`generating extended exemplar sequences GAA-
`GAA.1, GAAGAA.2, and GAAGAA.3, which
`lack appreciable similarity other than the
`shared hexamer GAAGAA (Fig. 3). All three
`extended exemplars conferred high levels of
`inclusion on the test exon, ranging from
`⬃50% for GAAGAA.3 to ⬃70% for GAA-
`GAA.1 (see fig. S4 for representative gels).
`All three of these extended exemplars con-
`
`R E P O R T S
`
`tained additional purine-rich hexamers
`overlapping the central GAAGAA hex-
`amer, which were also predicted to have
`enhancer activity by RESCUE-ESE (indi-
`cated by the vertical blue bars in Fig. 3).
`Next, the mutation G4⬎T was introduced
`into each extended exemplar [i.e., each cen-
`tral GAAGAA hexamer was mutated to
`GAATAA, a hexamer that falls far “south-
`west” of GAAGAA in the scatterplots and is
`predicted to lack ESE activity (see Fig. 4,
`motif 5C/3D)]. This mutation also disrupts
`many or all (for GAAGAA.3) of the overlap-
`ping RESCUE-ESE hexamers. As predicted,
`this mutation produced sharply reduced lev-
`els of inclusion in each of the three contexts,
`ranging from ⬃5% to ⬃30% of the wild-type
`level (Fig. 3). Taken together, these data sug-
`gest that different occurrences of the same
`exemplar tend to be qualitatively similar in
`their ability to enhance splicing and in their
`response to specific point mutations, but that
`the precise level of ESE activity depends
`on local sequence context. Another muta-
`tion predicted to disrupt ESE activity of
`GAAGAA, A2⬎T (Fig. 3, M2), also gave
`
`reduced levels of exon inclusion in the
`context of GAAGAA.3. On the other hand,
`the mutation A5⬎C (Fig. 3, M3) is predict-
`ed to preserve ESE activity because it con-
`verts GAAGAA to GAAGCA, another pre-
`dicted ESE hexamer, and this mutation
`slightly increases exon inclusion in the con-
`text of GAAGAA.3 (Fig. 3). These data
`anecdotally suggest that RESCUE-ESE can
`accurately predict which mutations will
`disrupt the enhancing activity of an ESE;
`some evidence for this conclusion is dis-
`cussed below.
`To assess the degree to which different
`exemplar hexamers from the same cluster
`would have similar ESE activity, we chose a
`quite different exemplar, AGAAAC, from the
`same 5C/3D cluster as GAAGAA. The ex-
`tended exemplar AGAAAC.1 also displayed
`ESE activity in the range observed for the
`different extended exemplars of GAAGAA
`(Fig. 3). However, the mutation G2⬎T, pre-
`dicted to disrupt the activity of AGAAAC,
`gave only a moderate (⬃27%) reduction in
`exon inclusion, from ⬃75% to ⬃55%. This
`remaining ESE activity might be attributable
`
`Fig. 3. Analysis of ESE activity for predicted enhancers of class 5C/3D. Upper panel: Extended
`exemplar sequences for three occurrences of the GAAGAA exemplar and one occurrence of the
`AGAAAC exemplar. All extended exemplars derive from arbitrarily selected occurrences of the
`exemplar in human exons with weak splice sites, as described in the text. Gene name and exon
`number are listed above each sequence. GenBank accession numbers for the mRNAs are as follows:
`XM_046769 (GAAGAA.1), XM_010365 (GAAGAA.2), AF212232 (GAAGAA.3), and BC020651
`(AGAAAC.1). Predicted ESE hexamers in each extended exemplar are indicated by blue bars above
`the sequence. Point mutations introduced into these sequences are shown in red, with predicted
`ESE hexamers in the mutant sequence shown by blue bars below the sequence. Each mutant is
`labeled by a red M if the mutation is predicted to disrupt ESE activity, or by a blue M if the mutant
`sequence is predicted to retain ESE activity. Total RNA extracted from HeLa cells was amplified by
`RT-PCR after transient transfection with the SXN reporter containing the indicated insert. Radio-
`labeled products were analyzed by polyacrylamide gel electrophoresis and visualized using a
`phosphorimager. (Representative autoradiographs are shown in fig. S4.) Bottom panel: Percent
`inclusion for each construct was calculated as the ratio of the intensity of the upper band (including
`exon 2) to the sum of the intensities of the upper and lower bands. All transfections were
`performed at least twice. The height of the colored bar indicates the average of all measurements;
`horizontal black lines indicate the minimum and maximum inclusion values observed.
`
`1010
`
`9 AUGUST 2002 VOL 297 SCIENCE www.sciencemag.org
`
`4
`
`

`

`to the retention of two predicted ESE hexa-
`mers in the mutated AGAAAC.1 sequence
`(Fig. 3), although this was not tested. This
`example underscores the difficulty in inter-
`preting the results of mutations in sequences
`containing additional predicted ESE hexa-
`mers. To rigorously test the predictions of the
`RESCUE-ESE method, we used sequences
`specifically chosen to contain exactly one
`predicted ESE hexamer (or zero, in the case
`of mutant sequences) in all other experiments
`reported here.
`One exemplar and a corresponding ex-
`tended exemplar were chosen from each of
`the 10 motifs for testing in the reporter con-
`struct. Only 19-nucleotide oligomers contain-
`ing a single RESCUE-ESE–predicted hex-
`amer in the middle were considered (table
`S1). Although this restriction might result in
`a bias toward selection of weaker enhancers,
`it was considered essential in order to avoid
`complications in interpreting the splicing
`phenotypes of sequences containing overlap-
`ping or adjacent predicted ESEs. For each
`motif, a single-base mutant predicted to dis-
`rupt the ESE activity of the exemplar hex-
`amer was chosen as described above, intro-
`duced into the extended exemplar sequence,
`and cloned into the reporter construct. The 10
`predicted enhancer and mutant constructs
`were transiently transfected and assayed for
`splicing as before (Fig. 4).
`All 10 of the predicted enhancers (blue
`bars) displayed ESE activity in the reporter
`system,
`ranging from weakly enhancing
`(⬃20% inclusion for 5B/3A and 3B)
`to
`strongly enhancing (⬃60 to 80% inclusion,
`for 5D and 3E). In addition, for 9 of 10
`classes of enhancer tested, the predicted ESE
`
`R E P O R T S
`
`sequence gave a significantly higher level of
`inclusion than the mutant (blue bar higher
`than red bar), motif 3F being the only excep-
`tion (see fig. S5 for representative gels).
`These results demonstrate the effectiveness
`of RESCUE-ESE for prediction of the effects
`of single base changes on ESE activity. The
`different point mutant sequences exhibited
`varying levels of inclusion. Mutant 3F gave
`comparable inclusion to wild-type 3F en-
`hancer. Three other mutants, 3C, 3E, and 3H,
`gave about two-thirds the level of inclusion
`of the wild-type sequence,
`indicating that
`ESE activity had been only partially im-
`paired. On the other hand, the remaining six
`mutants all had 10 to 50% of the wild-type
`level of inclusion. In absolute terms, these six
`mutants had inclusion levels below 20% and
`often less than 10%, comparable to that seen
`by others for typical random inserts in this
`context (7).
`In a large set of human exons, slightly
`more than 10% of all the hexanucleotides
`were found to match RESCUE-ESE hexa-
`mers (22), often in overlapping clumps as in
`Fig. 3, suggesting that ESEs are very com-
`mon in human genes. Counting each overlap-
`ping clump as a single enhancer, we found an
`average of 5.2 predicted enhancers per exon,
`with most exons containing between three
`and seven ESEs (20th and 80th percentiles,
`respectively). The hexamers in each cluster
`typically occurred more frequently in exons
`than in introns by a factor of 1.5 to 2 and
`more frequently in weak exons than in strong
`exons by a factor of 1.3 to 1.4. The average
`frequencies of hexamers in each of the 10
`RESCUE-ESE motif clusters were compara-
`ble or slightly lower in a database of more
`
`than 2000 alternative (skipped) exons than in
`our database of constitutively spliced exons
`(22); this finding suggests that the motifs we
`have identified are involved in recognition of
`both constitutively and alternatively spliced
`exons.
`Some sequences that display ESE activity
`were missed by the RESCUE method in its
`current form. For example, mutant 3F and
`one of the three predicted “neutral” sequenc-
`es tested (fig. S5) displayed enhancer activ-
`ity but did not contain any RESCUE-pre-
`dicted ESE hexamers. Analysis of the three
`predicted neutral sequences—19-base seg-
`ments that
`lack predicted ESE hexamers
`chosen from exons with weak splice sites—
`suggests a possible modification of the cut-
`offs used in the RESCUE-ESE protocol.
`Specifically, it was found that neutral se-
`quence N3 contained a hexamer that was
`close to the cutoff for ESEs: The hexamer
`CTACGC had ⌬EI ⫽ 16.9 (⬎⬎ 2.5) and
`⌬5WS ⫽ 2.2, just below the cutoff. By
`contrast, no hexamer in neutral sequence
`N1 or N2 had both ⌬EI ⬎ 2.5 and ⌬5WS or
`⌬3WS ⬎ 1.5, and no hexamer in any of the
`neutral sequences had ⌬5WS or ⌬3WS ⬎
`2.5. These and other data (22) suggest that
`altering the cutoffs used in RESCUE-ESE,
`perhaps by increasing the ⌬EI cutoff while
`simultaneously reducing the ⌬WS cutoff to
`1.5 or 2, might result in improved detection
`of ESEs.
`A database of published mutationally char-
`acterized natural ESE sequences was construct-
`ed, and these sequences were searched for
`occurrences of the hexamers in each cluster
`(tables S2 to S4). Five of the RESCUE-ESE
`clusters (5C/3D, 5E, 3C, 3E, and 3F) resemble
`
`Fig. 4. Analysis of ESE activity for 10 classes of predicted enhancers. HeLa
`cells were transfected with SXN splicing reporter construct containing
`inserts representing all 10 classes of predicted ESEs and point mutants of
`these sequences. The extended exemplar sequences used are listed in
`table S1. Upper panel: schematic representing ⌬EI and ⌬WS values for
`each tested exemplar hexamer (blue E) and point mutant hexamer (red
`M) from Fig. 2A or 2B, as appropriate. The label of the predicted ESE
`cluster from Fig. 2 is indicated above. Lower panel: Percent inclusion for
`
`each construct, calculated as in Fig. 3. (Representative autoradiographs
`are shown in fig. S5.) All transfections were performed at least twice. The
`height of the colored bar (blue for predicted ESE, red for mutant
`predicted to disrupt ESE activity) indicates the average of all measure-
`ments; horizontal black lines indicate the minimum and maximum in-
`clusion values observed. The predicted ESE hexamer and point mutant
`sequence are shown below the blue and red bars, respectively, with the
`mutated base shown in the corresponding color.
`
`www.sciencemag.org SCIENCE VOL 297 9 AUGUST 2002
`
`1011
`
`5
`
`

`

`R E P O R T S
`
`Fig. 5. Correlation between predicted
`ESEs and exon skipping mutations in
`human HPRT gene. (A) Exon skipping
`mutations were analyzed in terms of
`the set of hexamers affected; mutation
`2 (G88⬎T) is shown here as an exam-
`ple. All hexamers affected by the mu-
`tation are shown, with matches to
`RESCUE-predicted ESEs shown in blue
`and represented by a plus sign. The
`location of the mutation is indicated by
`a red arrow. (B) Summary of human
`HPRT gene mutations known to cause
`exon skipping [from (23, 24)]. Base
`changes that occurred within five nu-
`cleotides of a splice junction were ex-
`cluded, as they may alter the splice site
`signals. The first column lists the na-
`ture of the mutation (del ⫽ deletion,
`X⬎Y ⫽ substitution of base Y for base
`X), with coordinates listed relative to
`the translation start site of the HPRT
`cDNA (GenBank accession number
`NM_000194). The last two columns list
`the number of predicted ESE hexamers
`in the affected region of the wild-type
`and mutant sequences,
`respectively.
`(C) Locations of all mutations listed in
`(B) are indicated relative to the exon-
`intron structure of the HPRT gene by
`red arrows. Exon sizes are to scale;
`intron sizes are not. Mutations that
`alter RESCUE-predicted ESEs are shown
`below the
`exon-intron schematic,
`numbered according to (B). Mutations
`3
`labeled E
`X disrupt predicted ESE
`3
`hexamers; those labeled X
`E create pre