Distinct and predictive chromatin signatures of transcriptional promoters and enhancers in the human genome.

Abstract

Eukaryotic gene transcription is accompanied by acetylation and methylation of nucleosomes near promoters, but the locations and roles of histone modifications elsewhere in the genome remain unclear. We determined the chromatin modification states in high resolution along 30 Mb of the human genome and found that active promoters are marked by trimethylation of Lys4 of histone H3 (H3K4), whereas enhancers are marked by monomethylation, but not trimethylation, of H3K4. We developed computational algorithms using these distinct chromatin signatures to identify new regulatory elements, predicting over 200 promoters and 400 enhancers within the 30-Mb region. This approach accurately predicted the location and function of independently identified regulatory elements with high sensitivity and specificity and uncovered a novel functional enhancer for the carnitine transporter SLC22A5 (OCTN2). Our results give insight into the connections between chromatin modifications and transcriptional regulatory activity and provide a new tool for the functional annotation of the human genome.

Full text

Page 1

Distinct and predictive chromatin signatures of
transcriptional promoters and enhancers in the
human genome
Nathaniel D Heintzman1,2, Rhona K Stuart1, Gary Hon1,3, Yutao Fu4, Christina W Ching1, R David Hawkins1,
Leah O Barrera1,3, Sara Van Calcar1, Chunxu Qu1, Keith A Ching1, Wei Wang5, Zhiping Weng4,6,
Roland D Green7, Gregory E Crawford8& Bing Ren1,9
Eukaryotic gene transcription is accompanied by acetylation and methylation of nucleosomes near promoters, but the locations
and roles of histone modifications elsewhere in the genome remain unclear. We determined the chromatin modification states
in high resolution along 30 Mb of the human genome and found that active promoters are marked by trimethylation of Lys4
of histone H3 (H3K4), whereas enhancers are marked by monomethylation, but not trimethylation, of H3K4. We developed
computational algorithms using these distinct chromatin signatures to identify new regulatory elements, predicting over
200 promoters and 400 enhancers within the 30-Mb region. This approach accurately predicted the location and function of
independently identified regulatory elements with high sensitivity and specificity and uncovered a novel functional enhancer for
the carnitine transporter SLC22A5 (OCTN2). Our results give insight into the connections between chromatin modifications and
transcriptional regulatory activity and provide a new tool for the functional annotation of the human genome.
Activation of eukaryotic gene transcription involves the coordination
of a multitude of transcription factors and cofactors on regulatory
DNA sequences such as promoters and enhancers and on the chro-
matin structure containing these elements1–3. Promoters are located at
the 5¢ ends of genes immediately surrounding the transcriptional start
site (TSS) and serve as the point of assembly of the transcriptional
machinery and initiation of transcription4. Enhancers contribute to
the activation of their target genes from positions upstream, down-
stream or within a target or neighboring gene5,6. Deciphering the
regulatory information encoded in the genome will require a thorough
understanding of the relationships between the transcriptional activ-
ities of these different types of cis-regulatory sequence elements and
the epigenetic features of the chromatin surrounding them. Significant
progress in the fields of epigenetics and chromatin biology suggests a
histone code7of ever-increasing complexity with profound implica-
tions for chromatin as both a receptive substrate and a predictive
signal in a variety of biological processes3,8.
Recentinvestigationsusing
(ChIP) and microarray (ChIP-chip) experiments have described the
chromatin architecture of transcriptional promoters in yeast, fly and
mammalian systems9. In a manner largely conserved across species,
chromatinimmunoprecipitation
active promoters are marked by acetylation of various residues of
histones H3 and H4 and methylation of H3K4, particularly trimethy-
lation of this residue. Nucleosome depletion is also a general char-
acteristic of active promoters in yeast and flies, although this feature
remains to be thoroughly examined in mammalian systems. Although
some studies suggest that distal regulatory elements like enhancers
may be marked by similar histone modification patterns10–13, the
distinguishing chromatin features of promoters and enhancers have
yet to be determined, hindering our understanding of a predictive
histone code for different classes of regulatory elements. Here, we
present high-resolution maps of multiple histone modifications and
transcriptional regulators in 30 Mb of the human genome, demon-
strating that active promoters and enhancers are associated with
distinct chromatin signatures that can be used to predict these
regulatory elements in the human genome.
RESULTS
Chromatin architecture and transcription factor localization
We performed ChIP-chip analysis14to determine the chromatin
architecture along 44 human loci selected by the ENCODE consor-
tium as common targets for genomic analysis15, totaling 30 Mb.
Received 27 September 2006; accepted 3 January 2007; published online 4 February 2007; doi:10.1038/ng1966
1Ludwig Institute for Cancer Research,2Biomedical Sciences Graduate Program,3Program in Bioinformatics and9Department of Cellular and Molecular Medicine,
University of California San Diego (UCSD) School of Medicine, 9500 Gilman Drive, La Jolla, California 92093-0653 USA.4Bioinformatics Program, Boston University,
24 Cummington Street, 1002, Boston, Massachusetts 02215 USA.5Department of Chemistry and Biochemistry, UCSD, 9500 Gilman Drive, La Jolla, California
92093 USA.6Biomedical Engineering Department, Boston University, 44 Cummington Street, Boston, MA 022157NimbleGen Systems, Inc., 1 Science Court,
Madison, Wisconsin 53711 USA.8Institute for Genome Sciences & Policy and Department of Pediatrics, Duke University, 101 Science Drive, Durham, North Carolina
27708, USA. Correspondence should be addressed to B.R. (biren@ucsd.edu).
NATURE GENETICS VOLUME 39 [ NUMBER 3 [ MARCH 2007311
ARTICLES
© 2007 Nature Publishing Group http://www.nature.com/naturegenetics

Page 2

We investigated the patterns of core histone H3 and five histone
modifications: acetylated H3K9/14, acetylated H4K5/8/12/16 and
mono-, di-, and trimethylated histone H3K4. We also examined
binding of two components of the basal transcriptional machinery,
RNA polymerase II (RNAPII) and TBP-associated factor 1 (TAF1),
and the transcriptional coactivator p300 to identify active promoters
and enhancers, respectively. We carried out three biological replicate
ChIP-chip experiments for each marker in HeLa cells before and after
treatment with interferon-gamma (IFNg), as p300 is known to be
involved in the cellular response to this cytokine16. We amplified,
labeled and hybridized ChIP samples to tiling oligonucleotide micro-
arrays covering the nonrepetitive sequences of 30 Mb at 38-bp
resolution. We analyzed the microarray data by standard methods to
determine average enrichments for each marker at every probe,
generating high-resolution maps of histone modifications and tran-
scriptional regulator binding for 1% of the human genome. To
validate our ChIP-chip results, we performed conventional ChIP
against RNAPII and tested for enrichment at 121 sites in the ENCODE
regions using quantitative real-time PCR, which indicated an accuracy
of 97%, a specificity of near 100% and a sensitivity of 82% for our
method (see Supplementary Methods and Supplementary Table 1
online). These values are comparable to other ChIP-chip studies12,17,18
and confirm that our ChIP-chip data are very reliable.
Chromatin signatures of promoters
To explore chromatin features at human promoters, we examined
ChIP-chip profiles along 10-kb regions surrounding well-annotated
promoters in the ENCODE regions and performed computational
clustering to classify each promoter on the basis of histone modifica-
tion patterns. We examined only TSSs corresponding to well-anno-
tated RefSeq19transcripts for which we had collected expression data,
and to prevent interference from neighboring genes, we excluded TSSs
within 10 kb of each other, resulting in a pool of 208 TSSs for
clustering; 104 TSSs are defined as active promoters and 104 as
inactive promoters by gene expression profiling experiments. We
observed four distinct classes of promoters (P1–P4) in untreated
HeLa cells, arranged by the proportion of active promoters within
each class (Fig. 1a and Supplementary Table 2 online). Expression of
transcripts within each class generally increased from class P1 to class
P4, as most of the inactive promoters were found in class P1, whereas
classes P2–P4 were increasingly composed of active promoters. Aver-
age enrichment profiles for each marker within each class (Fig. 1b)
showed that occupancy by all five histone modifications, RNAPII and
TAF1 increased at active promoters in a manner related to gene
expression levels. Moderate p300 enrichment was also present at many
active promoters (see Supplementary Fig. 1 for a representative active
promoter), whereas the largely inactive class P1 was devoid of any
markers. The patterns observed in HeLa cells treated with IFNg were
almost identical (Supplementary Fig. 2). The transition from tri-
methylated to dimethylated to monomethylated H3K4 moving down-
stream from active promoters into coding regions echoes the pattern
seen in small-scale studies in human cells20and globally in yeast17,21.
These results confirm previous observations in other organisms that
histone modifications are linked to promoter activity.
We were interested to find a bimodal distribution of all histone
modifications centered around peak binding of RNAPII and TAF1 at
the TSS, implying depletion of nucleosomes at this position. ChIP-chip
data for histone H3 supported this conclusion (Fig. 1a,b). Our findings
indicate that the nucleosome-free region (NFR) observed at promoters
in yeast and fly is indeed characteristic of active human promoters,
supporting an evolutionarily constrained role for this phenomenon in
transcriptional regulation. The degree of nucleosome depletion seems
to be related to the level of gene expression, as we did not observe
depletion in class P1, suggesting that the formation and maintenance of
NFRs at active promoters is a regulated process. Distribution around
the NFR varied among the histone modifications and promoter classes,
but most modifications were found on both sides of the NFR with an
asymmetrical bias toward the region immediately downstream, parti-
cularly for trimethylated H3K4. Acetylated H4 was the only outlier to
this trend. The observed histone acetylation and methylation at
nucleosomes upstream of the TSS may represent signatures of chro-
matin architecture at promoters that are specific to mammals.
Chromatin signatures of enhancers
Next, we investigated the chromatin features of human transcriptional
enhancers. As previous studies have demonstrated that p300 and
related acetyltransferases are present at enhancers (as well as promo-
ters)10,11, we identified genomic regions in HeLa cells enriched in p300
binding (124 binding sites in untreated cells and 182 sites in treated
cells, listed in Supplementary Table 3) and found that the p300
binding sites demonstrated several features of enhancers. First, the
genomic distribution of p300 sites was consistent with the widespread
location of enhancers relative to their target genes, as over 75% of p300
binding occurs more than 2.5 kb from known 5¢ ends of genes (as
determined by GENCODE22) (Supplementary Fig. 3 online). Second,
transcriptional regulatory elements such as enhancers have long been
known to show increased nuclease sensitivity23, so we mapped the
DNaseI hypersensitive sites (DHSs) in HeLa cells along the ENCODE
regions using a recently developed DNase-chip method24(Supple-
mentary Table 4 online) and found that a significant number of distal
p300 sites (69.7%, P o 1 ? 10–16) overlapped with DHSs, representing
B12% of the distal DHSs we identified (Supplementary Fig. 3).
Third, most distal p300 sites were conserved across species; over 60%
(P o 1 ? 10–16) of these sites contained strongly conserved sequence
(Supplementary Methods). Fourth, a significant number of the distal
p300 sites (44.4%, P ¼ 4.6 ? 10–15) contained independently predicted
regulatory modules (PReMods) identified based on clustering of
conserved transcription factor binding motifs25(Supplementary
Methods). These lines of evidence provide strong support that the
distal p300 binding sites represent a subset of enhancers.
Using the distal p300 binding sites to anchor 10-kb regions
surrounding each putative enhancer, we performed computational
clustering as described above to generate three classes of enhancers
(Fig. 1c,d; classes are arbitrarily named E1–E3 to simplify discussion).
We were interested to find that monomethylated H3K4 was strongly
enriched in a broad pattern at nearly all enhancers. Here we also found
depletion of histone H3 at enhancers, suggesting that nucleosome
depletion is a general feature of both promoters and enhancers,
consistent with their DNaseI hypersensitivity. Although most active
promoters are marked by substantial enrichment of trimethylated
H3K4 at the TSS, enhancers generally lack this histone modification.
Furthermore, active promoters showed a marked depletion of mono-
methylated H3K4 at the TSS and enrichment of this modification
more than 1 kb downstream and upstream, whereas enhancers showed
strong enrichment of monomethylated H3K4 at the peak of p300
binding. Acetylated H4, acetylated H3 and dimethylated H3K4 were
present in varying degrees at both promoters and enhancers, although
the bimodal distribution of these modifications observed at active
promoters was less pronounced at enhancers. TAF1 and RNAPII were
also present at some enhancers, albeit more weakly than at promoters
(reminiscent of the weak p300 enrichment at promoters seen in
Fig. 1a,b), suggesting docking of the transcriptional machinery at
312 VOLUME 39 [ NUMBER 3 [ MARCH 2007 NATURE GENETICS
ARTICLES
© 2007 Nature Publishing Group http://www.nature.com/naturegenetics

Page 3

enhancers or physical interaction between enhancers and active
promoters as proposed in various models of enhancer action5,6. In
spite of some similarities between the histone modification profiles of
active promoters and enhancers, the sharp contrasts of their mono-
methylated H3K4 and trimethylated H3K4 profiles represent distinct
chromatin signatures for these different classes of regulatory elements.
Predicting promoters and enhancers via chromatin signatures
Next, we investigated the possibility that the different chromatin
signatures of active promoters and enhancers could be used to predict
these transcriptional regulatory elements in the human genome. We
constructed training sets with histone modification profiles surround-
ing known TSSs and p300 binding sites in untreated HeLa cells and
used them to develop a computational prediction algorithm to locate
promoters and enhancers in the ENCODE regions based on similarity
to the training set chromatin profiles (Fig. 2a and Supplementary
Methods). Our two-stage method of regulatory element identification
consists of a primary descriptive prediction followed by secondary
discriminative filters (Supplementary Methods). To qualify as a high-
confidence predicted regulatory element, a region of chromatin must
unambiguously match one of the training set profiles.
Atotal of 198 active promoters (Supplementary Table 5 online) were
predicted in the ENCODE regions in untreated HeLa cells, clustered, as
described previously, into four classes (named PI–PIV to distinguish
them from the known promoters presented in Fig. 1) (Fig. 2b). In HeLa
cells treated with IFNg, we predicted 208 promoters (Supplementary
Table 5), with greater than 90% overlap between the untreated and
treated prediction sets (Fig. 2c), supporting the accuracy of our method
in identifying promoters in an independent data set. The untreated
prediction set contained 140 (79%) of the 177 active RefSeq promoters
within the ENCODE regions and 32 (21%) of 155 inactive RefSeq
promoters, and 180 predictions (91%) mapped to known GENCODE
5¢ ends of genes (Fig. 2d), indicating a high degree of sensitivity and
accuracy of promoter prediction. Promoter predictions in treated cells
H3K4me1H3K4me2H3K4me3H3 H4acH3acRNAPIITAF1
P1
P2
P3
P4
% active
18/102
24/41
21/23
41/42
logR
2.5
–0.5
0
logR
2.5
–0.5
0
TSS –5 kb
Class P1
+5 kb
Class P2
TSS TSS
Class P4
TSSTSSTSSTSSTSSTSS
Peak –5 kb +5 kb Peak PeakPeakPeakPeak PeakPeakPeak
Class P3
logR
0–33
logR
0–33
p300
Class E1 Class E2
Class E3
E1
E2
E3
H3K4me1H3K4me2H3K4me3H3 H4acH3acRNAPIITAF1 p300
208 promoters (RefSeq TSSs)
a
b
c
d
74 enhancers (distal p300 binding sites)
Figure 1 Features of human transcriptional promoters and enhancers. ChIP-chip was performed against six histone markers and three general transcription
factors in the ENCODE regions, and the data were clustered to reveal patterns at annotated promoters (a) and distal p300 binding sites (c). Promoter
clustering was performed with 10-kb windows centered on RefSeq TSSs; enhancer windows were centered on promoter-distal p300 binding peaks. Average
profiles for each marker within each class are shown below the clusters (b,d); each class is represented by a different color. The proportion of expressed
genes (‘% active’) in each promoter class is presented to the right of the cluster, illustrating the relationship between histone modification patterns and gene
expression. Comparison of the clusters shows that active promoters and enhancers are similarly marked by nucleosome depletion (column H3) but distinctly
marked by mono- and trimethylation of histone H3K4 (columns H3K4me1 and H3K4me3). Note the depletion of H3K4me1 and the peak of H3K4me3 at
the TSS in promoters, compared with the enrichment of H3K4me1 and lack of H3K4me3 at enhancers. The presence of histone methylation and acetylation
upstream and downstream of the TSS at promoters is distinct from the primarily downstream localization of these markers observed at yeast promoters. The
same procedure was performed on data from treated HeLa cells, yielding similar results (Supplementary Fig. 2). H4ac, acetylated H4; H3ac, acetylated H3.
NATURE GENETICS VOLUME 39 [ NUMBER 3 [ MARCH 2007313
ARTICLES
© 2007 Nature Publishing Group http://www.nature.com/naturegenetics

Page 4

were distributed very similarly (Fig. 2e). Comparison with the recent
RIKEN human CAGE data set26showed that the vast majority of the
predicted promoters were supported by multiple CAGE tags (Supple-
mentary Methods). Even predicted promoters that did not map to a
known GENCODE 5¢ end were largely supported by multiple CAGE
tags (50% in untreated cells, 27% in treated cells) or DHSs (83% in
untreated cells, 73% in treated cells). It is possible that the inactive
promoters identified in our analysis correspond to transcripts that are
expressed below the detection threshold, or these promoters might
retain some features of transcriptional competence in the absence of
active transcription. Six promoter predictions in untreated HeLa cells
(nine predictions in treated cells) did not correspond to any known or
putative 5¢ ends and probably represent previously unknown promo-
ters; all of these predicted new promoters overlap with DHSs.
We also predicted 389 enhancers (Supplementary Table 5) in
untreated HeLa cells (Fig. 3a; enhancer predictions are classified
EI–EIV to distinguish them from the p300 binding sites presented
in Fig. 1) and 324 enhancers in treated cells, with 89% overlap
between prediction sets (Fig. 3b). Although the prediction algorithm
was trained on the histone modification profiles of untreated cells, it
accurately identified 77% of the distal p300 binding sites in IFNg-
treated cells, indicating a high degree of sensitivity for the prediction
of enhancers in an independent data set. Several lines of evidence
support the function of these predictions as enhancers. First, over 85%
of predictions were located more than 2.5 kb from known gene 5¢ ends
(Fig. 3c and Supplementary Fig. 4 online), consistent with their
predicted function. Second, they were evolutionarily conserved, with
53.3% (P o 1 ? 10–16) containing a strongly conserved sequence.
Third, many overlapped with predicted transcriptional regulatory
modules (36.3%, P ¼ 1.7 ? 10–4). Fourth, a significant proportion
of the enhancer predictions (55.3%, P o 1 ? 10–16) overlapped with
DHSs, including the well-known HS2 enhancer in the b-globin locus27
(Supplementary Fig. 5 online). Of 587 distal DHSs in HeLa cells, we
predict that 175 (29.8%) are enhancers; the other distal DHSs
probably represent additional regulatory elements such as repressors
or insulators or sequences that contribute to chromatin organization.
H3K4me1H3K4me2H3K4me3H3H4acH3acRNAPIITAF1p300
b
198 high-confidence promoter predictions
PI
PII
PIII
PIV
d
Mapping predictions to GENCODE gene set
Distribution, untreated HeLa
3.0
Novel promoters
Within putative gene
Within known gene
Pseudogene TSS
Putative gene TSS
Known gene TSS
90.9
e
Mapping predictions to GENCODE gene set
Distribution, IFNγ-treated HeLa
4.3
Novel promoters
Within putative gene
Within known gene
Pseudogene TSS
Putative gene TSS
Known gene TSS
87.5
a
Scan genomic regions with profilesPredict regulatory elements
Classify predictions as
promoter or enhancer
based on best correlation
to chromatin signatures
Descriptive prediction Discriminative filter
>
<
Promoter
Enhancer
c
Overlap between prediction sets
187 2111
Untreated HeLa promoter predictions
IFNγ-treated HeLa promoter predictions
Figure 2 Prediction of promoters based on chromatin signatures. (a) A general scheme of the prediction method. Left: features of established transcriptional
promoters (and enhancers) were analyzed to yield descriptive histone modification profiles used in scanning genomic regions for novel regulatory elements
(Supplementary Fig. 6 online). Right: predictions were filtered and classified as promoters or enhancers based on correlation with monomethylated H3K4
(H3K4me1) and trimethylated H3K4 (H3K4me3) chromatin signatures (Supplementary Methods). (b) 198 active promoters were predicted in the ENCODE
regions in untreated HeLa cells and clustered into classes PI–PIV. The predictions contain 140 active RefSeq promoters and 32 inactive RefSeq promoters,
indicating a sensitivity of 79.1% for active promoter detection. (c) The high degree of overlap between untreated and IFNg-treated HeLa promoter prediction
sets supports the applicability of our approach to independent data sets. The majority of predicted promoters map to known GENCODE 5¢ ends in untreated
(d) and treated cells (e), confirming the accuracy of our predictions.
314 VOLUME 39 [ NUMBER 3 [ MARCH 2007 NATURE GENETICS
ARTICLES
© 2007 Nature Publishing Group http://www.nature.com/naturegenetics

Page 5

Finally, 86 enhancer predictions in the untreated set (and 116 in the
treated set) mapped to distal p300 binding sites (Fig. 3d,e) and many
others seemed to be enriched in p300 binding but fell below the
threshold of our target selection.
We also discovered that many predicted enhancers lacked p300
binding. To determine if these genomic regions were occupied by
p300-independent transcriptional coactivator complexes, we performed
additional ChIP-chip experiments to examine binding of TRAP220, a
component of the Mediator complex that has been shown to occupy
enhancers as well as promoters10,11. Of 162 TRAP220 binding sites we
identified in the ENCODE regions (Supplementary Table 6 online), 78
(48.1%) were located far from known 5¢ ends of transcripts and might
represent potential enhancers. Almost 63% of the distal TRAP220 sites
were contained within our enhancer prediction set (Fig. 3d), and 18 of
them were bound by TRAP220 but not p300, confirming the identity of
these predicted enhancers. This result suggests that our chromatin-
based prediction model is not limited only to enhancers marked by
p300. Overall, the majority of predicted enhancers (63.5%) are sup-
ported by DNaseI hypersensitivity, binding of p300, binding of
TRAP220 or a combination of these features (Fig. 3f).
Identification of a novel enhancer for SLC22A5
To confirm the potential of our approach to identify enhancers that
regulate the activity of target human promoters, we next examined a
predicted enhancer located 6 kb upstream of SLC22A5 (also known as
OCTN2) on chromosome 5 (Fig. 4a). SLC22A5 is a widely expressed
gene that encodes a carnitine transporter28–31. Mutations in this gene
have been identified as a cause of systematic carnitine deficiency, a
condition occurring mostly in children that prevents the body from
using fats for energy and can result in symptoms including encephalo-
pathy, cardiomyopathy, hypoglycemia and, in serious complications,
heart failure, liver problems, coma and sudden unexpected death32–35.
Although substantial research has been devoted to the role of SLC22A5
in carnitine transport, fatty acid metabolism and related human
diseases, very little is known about the transcriptional regulation of
this gene. We cloned a region of the SLC22A5 locus (L) containing the
promoter and predicted enhancer (E) into a luciferase reporter
construct and compared its activity to that of the locus without the
predicted enhancer (LDE) in transiently transfected HeLa cells. The
deletion of the predicted enhancer caused a 2.5-fold reduction in
reporter activity (Fig. 4b), supporting the necessity of this site for full
activity of the SLC22A5 promoter. We then cloned the predicted
enhancer downstream of the luciferase gene in a construct containing
the proximal SLC22A5 promoter (PS) and observed 4.2-fold greater
reporter activity from the promoter-enhancer construct (PSE) than the
construct containing only the promoter (Fig. 4b), confirming that the
predicted enhancer is sufficient to increase the activity of this
promoter in a position-independent manner. The predicted enhancer
H3K4me1H3K4me2 H3K4me3H3 H4acH3acRNAPIITAF1 p300
389 high-confidence enhancer predictions
a
cdef
b
EI
EIII
EIV
Mapping predictions to GENCODE gene set
Distribution, untreated HeLa
Intergenic
Within putative gene
Within known gene
Pseudogene TSS
Putative gene TSS
Known gene TSS
50.9
29.0
14.7
Independent experimental support
for predicted enhancers
36.5
38.0
5.7
Overlap between predictions and
p300/TRAP220 binding sites,
untreated HeLa
Overlap between predictions and
p300 binding sites, IFNγ-treated HeLa
EII
Overlap between prediction sets
28935
100
Untreated HeLa enhancer predictions
IFNγ-treated HeLa enhancer predictions
Enhancer predictions
Distal p300 binding sites
Enhancer predictions
Distal p300 binding sites
Distal TRAP220 binding sites
55
12
303
116 35
208
31
18
28
1
DHS & p300 & TRAP220
DHS & p300
DHS & TRAP220
TRAP220 & p300
DHS
p300
TRAP220
None
9.5
Figure 3 Prediction of enhancers based on chromatin signatures. (a) 389 enhancers were predicted in the ENCODE regions in untreated HeLa cells
and clustered into classes EI–EIV. The predicted enhancers show the monomethylated H3K4 (H3K4me1) enrichment and lack of trimethylated H3K4
(H3K4me3) observed at distal p300 binding sites. H4ac, acetylated H4; H3ac, acetylated H3. (b) The high degree of overlap between untreated and IFNg-
treated HeLa enhancer prediction sets supports the applicability of our approach to independent data sets. The majority of enhancer predictions in untreated
(c) and treated cells (see Supplementary Fig. 4) are found away from known GENCODE 5¢ ends, similar to p300 binding site distribution. The enhancer
prediction sets contain the majority of known distal p300 binding sites in untreated (d) and treated cells (e), confirming the sensitivity of our approach, even
though the prediction algorithm was trained only on data from untreated cells. (f) Most enhancers predicted on the basis of their chromatin signatures are
also supported by DNaseI hypersensitivity and/or binding of p300 and/or TRAP220.
NATURE GENETICS VOLUME 39 [ NUMBER 3 [ MARCH 2007315
ARTICLES
© 2007 Nature Publishing Group http://www.nature.com/naturegenetics

Page 6

did not stimulate reporter activity in the absence of the SLC22A5
promoter (data not shown). Our results suggest that the putative
SLC22A5 enhancer identified by our method is indeed critical for
optimal transcriptional activation of this gene.
Functional validation of promoter and enhancer predictions
To further assess the accuracy of our predictions, we compared our
high-confidence prediction sets to a list of in vivo STAT1 binding sites
independently mapped in the ENCODE regions, hypothesizing that
STAT1 sites are likely to occupy both promoters and enhancers. We
performed ChIP-chip in HeLa cells before and after IFNg treatment as
described above and additionally on a PCR-product microarray
platform (Supplementary Methods) and validated the results using
quantitative real-time PCR, generating a list of 13 high-confidence
STAT1 sites in IFNg-treated cells (Supplementary Table 7 online); as
expected, we did not detect any STAT1 binding in cells before
treatment. We compared the STAT1 sites with our prediction lists
and found that seven STAT1 sites map to promoter predictions, four
map to enhancer predictions, and two are not near any predictions,
indicating that our prediction model is capable of detecting the
majority (484%) of an independently generated collection of putative
regulatory elements. Four of the seven promoter predictions map to
known TSSs: IRF1 (a known STAT1 target), RPS9, c21orf59 and
IFNAR2. All of these genes are expressed in HeLa cells, supporting
the accuracy of our active promoter predictions.
To validate the new promoter and enhancer predictions at STAT1
sites, we examined their functional properties using reporter assays. As
two adjacent STAT1 sites on chromosome 5 (STAT1.02-.03) map to
the same promoter prediction, we examined the closer of the two sites
along with the other novel STAT1 promoter prediction (Fig. 5a), four
STAT1 enhancer predictions (Fig. 5b), and the two non-predicted
STAT1 sites (Fig. 5c). To test for promoter activity, regions containing
the STAT1 sites were amplified from genomic DNA and cloned up-
stream of the luciferase gene in vectors lacking a promoter (Fig. 5d);
to test for enhancer activity, the same fragments were cloned down-
stream of the luciferase gene into vectors containing the SV40 minimal
promoter (Fig. 5e), as enhancers are thought to contribute to target
gene activation regardless of their position relative to the gene
promoter. Clones were transiently transfected into HeLa cells and
assayed for reporter activity before and after treatment with IFNg.
Both STAT1 promoter predictions stimulated reporter activity in
the absence of the SV40 promoter when cloned in the upstream
position (Fig. 5d), in accord with their predicted function. Three
STAT1 enhancer predictions (STAT1.08-.10) stimulated strong repor-
ter activity when cloned in the downstream position (Fig. 5e) but
required the presence of the SV40 promoter (Supplementary
Methods), consistent with the position-independence and promo-
ter-dependence of enhancer activity. The fourth enhancer prediction
(STAT1.11) showed only weak enhancer activity, though we noted that
the STAT1 site in this region is further away from the prediction
(710 bp) than any of the other STAT1 sites that we examined (average
B240 bp). The effect of IFNg is variable among the different sites in
both ChIP-chip binding profiles and reporter activity, though there
seems to be a relationship between inducibility of p300 binding and
reporter activity. Notably, one promoter prediction (STAT1.03) also
showed enhancer activity (Fig. 5e). Examination of our prefiltering
prediction lists (Supplementary Methods) uncovered a predicted
enhancer within the STAT1.03 cloned region, explaining the apparent
dual functional activity of this newly discovered promoter. The
nonpredicted sites (STAT1.12-.13) did not show any functional
activity and were not marked by either of the distinctive histone modi-
fication patterns, supporting the specificity of our model. It is still
possible that these sites are actually regulatory elements that cannot be
tested in our system owing to their function or a requirement for
native chromatin context, but it is worth noting that these are the only
two STAT1 sites that did not demonstrate DNaseI hypersensitivity.
These data provide functional validation for our model of distinct
chromatin signatures at promoters and enhancers, confirm that our
computational approach can accurately predict the position and
function of these transcriptional regulatory elements on the basis of
their chromatin signatures and suggest a direct connection between
chromatin signatures and the regulatory potential of the DNA
sequences that they denote.
DISCUSSION
In summary, we mapped five histone modifications, four general
transcription factors and nucleosome density at high resolution in
30 Mb of the human genome, identifying chromatin features that
distinguish promoters from enhancers. Although both kinds of
regulatory elements share some features such as nucleosome depletion
and enrichment of histone acetylation and dimethylated H3K4, the
high-resolution profiles of these markers and the dichotomy of
enrichment for trimethylated H3K4 and monomethylated H3K4 at
active promoters and enhancers define chromatin signatures that can
be used to locate novel regulatory elements in the human genome. The
monomethylated H3K4 enhancer signature is present in HeLa cell
chromatin at multiple loci whose enhancer activity was functionally
validated, including a putative novel enhancer for SLC22A5.
SLC22A5
Distance, kb
01
pGL3-B
0
0.5
1.0
1.5
2.0
0
5
10
15
20
Promoter (PS)
Entire locus (L)
Locus - E (L∆E)
Predicted enhancer (E)
Relative intensity (arbitrary units)
L
L∆E
L∆E
PS
PS
PSE
PSEpGL3-BL
Relative intensity (arbitrary units)
Luciferase reporter activity in untreated HeLa cells
Cloned regions of the SLC22A5 locus (chr5)
a
b
Figure 4 Identification of a putative novel enhancer for SLC22A5. (a) To test
the effect of a predicted enhancer on a nearby promoter, regions of the
SLC22A5 locus were cloned into pGL3 reporter constructs in the direction
indicated. (b) Luciferase activity of the entire 6.5-kb locus (L) was reduced
2.5-fold by the deletion of 700 bp containing the predicted enhancer (LDE),
and the presence of the predicted enhancer downstream of the luciferase
gene in a construct containing the SLC22A5 promoter (PSE) caused a
fourfold increase in activity compared with the promoter alone (PS). Error
bars represent s.d.
316VOLUME 39 [ NUMBER 3 [ MARCH 2007 NATURE GENETICS
ARTICLES
© 2007 Nature Publishing Group http://www.nature.com/naturegenetics

Page 7

Previous studies have identified some histone modification patterns
of promoters and heterochromatin, but our findings expand the
current knowledge of chromatin architecture at human promoters
and present evidence for previously unknown chromatin features of
human enhancers, representing an effective new strategy for identify-
ing and distinguishing promoters and enhancers. The presence of
histone acetylation and methylation that we observe upstream of the
TSSs of active human promoters has not been reported in yeast and
suggests some transcriptional regulatory mechanisms specific to
mammalian gene expression. Additionally, the discovery of mono-
methylated H3K4 at human enhancers may contribute to our under-
standing of how enhancers function in tissue-specific gene regulation.
In recent years, the genome sequences of a growing number of
organisms have been obtained, but extracting functional information
from these nucleotide sequences remains a great challenge, as our
knowledge of transcription factor binding motifs is incomplete, and
current sequence-based computational tools are limited in their ability
to predict the regulatory function of genomic sequences. Here, we
present a strategy to identify transcriptional regulatory elements on
the basis of their epigenetic characteristics, independent of motifs or
other sequence features. Our chromatin-based prediction model
provides a means to locate and distinguish promoters and enhancers
at high resolution and with high degrees of sensitivity and specificity.
Although the prediction model was trained only on data from
untreated HeLa cells, the sensitivity of the model in data from
IFNg-treated cells supports the utility of our approach in analyzing
independent data sets. The results of the functional assays of predicted
STAT1 binding sites confirm the ability of our prediction model for
identifying the location and function of novel promoters and enhan-
cers, even before their activation. Because we used the histone
modification profiles at distal p300 binding sites as the basis for our
enhancer prediction strategy, we were initially concerned that our
predictions might be biased toward only the subset of enhancers
bound by p300. Based on the overlap of our predictions with 63% of
distal TRAP220 sites and 30% of distal DHSs, however, we conclude
that our model is not biased toward p300 binding sites and that the
chromatin signatures we observed are not limited to this subset of
enhancers. Extension of our model to additional cell types and other
components of chromatin architecture will be useful in determining
the mechanisms of enhancer maintenance and function in regulating
tissue-specific gene expression, findings that will be particularly
important to our knowledge of how epigenetic factors and distal
transcriptional regulatory elements contribute to human development
and disease.
Our approach will also be valuable to the functional annotation
of the human genome, as it provides a new, effective means to locate
active transcriptional enhancers that have thus far eluded identifica-
tion on a large scale. Given the degree of structural and functional
conservation of chromatin and histone modifications from yeast
to humans, these predictive chromatin signatures may be useful
in annotating promoters and enhancers in the genomes of a variety
of organisms.
0
10
20
30
40
50
STAT1.03
chr5:131861052
STAT1.04
chr15:41573134
STAT1.08
chr5:131791096
STAT1.12
chr5:132135776
a
H3K4me1
H3K4me2
H3K4me3
H3
H4ac
H3ac
RNAPII
TAF1
p300
logR
–2
4
H3K4me1
H3K4me2
H3K4me3
H3
H4ac
H3ac
RNAPII
TAF1
p300
logR
–2
4
H3K4me1
H3K4me2
H3K4me3
H3
H4ac
H3ac
RNAPII
TAF1
p300
logR
–2
4
Uninduced onlyAfter IFN-γ only
Both time points
01
Distance, kb
STAT1.09
chr7:115940908
STAT1.10
chr9:129176452
bc
STAT1.11
chr21:34091578
STAT1.13
chr21:39297423
Clone upstream,
no promoter
Untreated
24 h IFNγ
0
10
20
30
40
50
pGL3-B
Relative intensity (arbitrary units)
Promoter test
STAT1.03STAT1.04
d
Clone downstream,
SV40 promoter
Relative intensity (arbitrary units)
Enhancer test
Untreated
24 h IFNγ
e
STAT1.08STAT1.11STAT1.09STAT1.10STAT1.12STAT1.13
pGL3-P
STAT1.03STAT1.04STAT1.08STAT1.11STAT1.09STAT1.10STAT1.12STAT1.13
Predicted novel promotersPredicted novel enhancersNon-predicted
Figure 5 Validation of the prediction model by STAT1 binding and reporter assays. Of 13 high-confidence STAT1 binding sites, four are found at known
promoters (not shown), two at predicted novel promoters (a) and four at predicted enhancers (b), and two are not predicted (c). The eight STAT1 sites in a–c
were cloned into pGL3 reporter constructs to examine their regulatory potential as promoters (d) and enhancers (e). Error bars represent s.d. The coverage
and direction of each clone are represented by orange arrows in a–c, and ChIP-chip profiles of each marker are shown at the eight STAT1 binding sites,
before and after IFNg treatment (green and red, respectively, where brown indicates enrichment at both time points; images generated in part at http://
genome.ucsc.edu using hg17). The STAT1 binding sites in a and b function as predicted in the reporter assays, whereas the non-predicted sites in c show no
reporter activity. H4ac, acetylated H4; H3ac, acetylated H3.
NATURE GENETICS VOLUME 39 [ NUMBER 3 [ MARCH 2007317
ARTICLES
© 2007 Nature Publishing Group http://www.nature.com/naturegenetics

Page 8

METHODS
For detailed methods and materials, please refer to Supplementary Methods.
Briefly, HeLa cells were cultured under adherent conditions in DMEM +
10% FBS. Three biological replicates of IFNg-treated and untreated cells were
cross-linked and harvested as previously described18, except that cells were
cross-linked for 20 min at 37 1C. Chromatin preparation, ChIP-chip, DNA
purification and LM-PCR were performed as previously described18using
commercially available antibodies against the following proteins: histone H3
(Abcam ab1791), acetylated H4 (Upstate 06-866), acetylated H3 (Upstate 06-
599), monomethylated H3K4 (Abcam ab8895), dimethylated H3K4 (Upstate
07-030), trimethylated H3K4 (Upstate 07-473), RNAPII (Covance MMS-
126R), TAF1 (Santa Cruz sc-735), p300 (Santa Cruz sc-585) and TRAP220
(Santa Cruz sc-5334). ChIP-DNA samples were labeled and hybridized to
NimbleGen ENCODE HG17 microarrays (NimbleGen Systems). Data were
analyzed using standard methods, and ChIP-chip targets for RNAPII were
selected with the Mpeak program and validated by quantitative real-time PCR
using the iCycler and SYBR-green iQ Supermix (Bio-Rad Laboratories). Gene
expression in treated and untreated HeLa cells was analyzed using HU133
Plus 2.0 microarrays (Affymetrix) as described18. DNase-chip was performed
and the data analyzed as described24. Promoters (TSSs) and putative enhancers
(p300 binding sites) were clustered on the basis of histone modification
patterns using K-means clustering of 10-kb windows centered on each target.
Average profiles were generated for each class of promoter and enhancer and
were used to train a computational prediction model to identify promoters and
enhancers on the basis of histone modification ChIP-chip profiles. Predictions
were further filtered by correlation to chromatin signatures to remove false
positives and ambiguous classifications. Predicted regulatory modules were
obtained from http://genomequebec.mcgill.ca/PReMod/; phastCons and CAGE
data were extracted from the UCSC Genome Browser and binding sites and
predictions were mapped relative to the October 2005 hg17 GENCODE gene
sets. ChIP-chip was performed against STAT1 (using anti-STAT1, Santa Cruz
sc-345) as described above and using PCR microarrays, and the results were
validated by quantitative real-time PCR. Predicted STAT1 sites were cloned into
modified pGL3 reporter constructs (Promega), transiently transfected into
HeLa cells and assayed for luciferase activity before and after IFNg treatment
using the Dual Luciferase Kit (Promega). Raw and processed data for the
microarray experiments can be found at the UCSC genome browser (http://
genome.ucsc.edu) and http://licr-renlab.ucsd.edu/download.html.
Accession codes. GEO: raw and processed data for the microarray experiments
can be found under accession number GSE6273.
Note: Supplementary information is available on the Nature Genetics website.
ACKNOWLEDGMENTS
We thank J. Kadonaga, X. Fu and members of the Ren lab for comments. This
work was supported by funding from the Ludwig Institute for Cancer Research
(B.R.), the National Human Genome Research Institute (B.R., Z.W. and R.D.G.)
and the National Cancer Institute (B.R.). Requests for materials should be
addressed to B.R.
AUTHOR CONTRIBUTIONS
N.D.H., B.R. and R.D.G. designed the transcription factor and histone ChIP-chip
experiments; G.E.C. designed and performed the DNase-chip experiments;
N.D.H., R.K.S., C.W.C., R.D.H. and S.V.C. conducted the ChIP-chip experiments;
N.D.H., G.H., L.O.B., K.A.C. and C.Q. analyzed the microarray data; G.H.,
N.D.H., B.R. and W.W. conceived and developed the promoter and enhancer
prediction method. Independently, Y.F. and Z.W. discovered the promoter-
associated chromatin signatures. N.D.H. and B.R. wrote the manuscript.
COMPETING INTERESTS STATEMENT
The authors declare competing financial interests (see the Nature Genetics website
for details).
Published online at http://www.nature.com/naturegenetics
Reprints and permissions information is available online at http://npg.nature.com/
reprintsandpermissions
1. Lemon, B. & Tjian, R. Orchestrated response: a symphony of transcription factors for
gene control. Genes Dev. 14, 2551–2569 (2000).
2. Orphanides, G. & Reinberg, D. A unified theory of gene expression. Cell 108, 439–451
(2002).
3. Nightingale, K.P., O’Neill, L.P. & Turner, B.M. Histone modifications: signalling
receptors and potential elements of a heritable epigenetic code. Curr. Opin. Genet.
Dev. 16, 125–136 (2006).
4. Smale, S.T. & Kadonaga, J.T. The RNA polymerase II core promoter. Annu. Rev.
Biochem. 72, 449–479 (2003).
5. Blackwood, E.M. & Kadonaga, J.T. Going the distance: a current view of enhancer
action. Science 281, 60–63 (1998).
6. Bulger, M. & Groudine, M. Looping versus linking: toward a model for long-distance
gene activation. Genes Dev. 13, 2465–2477 (1999).
7. Strahl, B.D. & Allis, C.D. The language of covalent histone modifications. Nature 403,
41–45 (2000).
8. Margueron, R., Trojer, P. & Reinberg, D. The key to development: interpreting the
histone code? Curr. Opin. Genet. Dev. 15, 163–176 (2005).
9. Barrera, L.O. & Ren, B. The transcriptional regulatory code of eukaryotic cells - insights
from genome-wide analysis of chromatin organization and transcription factor binding.
Curr. Opin. Cell. Biol. 18, 291–298 (2006).
10. Hatzis, P. & Talianidis, I. Dynamics of enhancer-promoter communication during
differentiation-induced gene activation. Mol. Cell 10, 1467–1477 (2002).
11. Wang, Q., Carroll, J.S. & Brown, M. Spatial and temporal recruitment of androgen
receptor and its coactivators involves chromosomal looping and polymerase tracking.
Mol. Cell 19, 631–642 (2005).
12. Bernstein, B.E. et al. Genomic maps and comparative analysis of histone modifications
in human and mouse. Cell 120, 169–181 (2005).
13. Roh, T.Y., Cuddapah, S. & Zhao, K. Active chromatin domains are defined by acety-
lation islands revealed by genome-wide mapping. Genes Dev. 19, 542–552 (2005).
14. Kim, T.H. & Ren, B. Genome-wide analysis of protein-DNA interactions. Annu. Rev.
Genomics Hum. Genet. 7, 81–102 (2006).
15. The ENCODE Project Consortium. The ENCODE (ENCyclopedia Of DNA Elements)
Project. Science 306, 636–640 (2004).
16. Horvai, A.E. et al. Nuclear integration of JAK/STATand Ras/AP-1 signaling by CBP and
p300. Proc. Natl. Acad. Sci. USA 94, 1074–1079 (1997).
17. Pokholok, D.K. et al. Genome-wide map of nucleosome acetylation and methylation in
yeast. Cell 122, 517–527 (2005).
18. Kim, T.H. et al. A high-resolution map of active promoters in the human genome.
Nature 436, 876–880 (2005).
19. Pruitt, K.D., Tatusova, T. & Maglott, D.R. NCBI Reference Sequence (RefSeq): a
curated non-redundant sequence database of genomes, transcripts and proteins.
Nucleic Acids Res. 33, D501–D504 (2005).
20. Kouskouti, A. & Talianidis, I. Histone modifications defining active genes persist after
transcriptional and mitotic inactivation. EMBO J. 24, 347–357 (2005).
21. Liu, C.L. et al. Single-nucleosome mapping of histone modifications in S. cerevisiae.
PLoS Biol. 3, e328 (2005).
22. Harrow, J. et al. GENCODE: producing a reference annotation for ENCODE. Genome
Biol. 7(Suppl.), S4.1–S4.9 (2006).
23. Felsenfeld, G. Chromatin unfolds. Cell 86, 13–19 (1996).
24. Crawford, G.E. et al. DNase-chip: a high-resolution method to identify DNase I
hypersensitive sites using tiled microarrays. Nat. Methods 3, 503–509 (2006).
25. Blanchette, M. et al. Genome-wide computational prediction of transcriptional reg-
ulatory modules reveals new insights into human gene expression. Genome Res. 16,
656–668 (2006).
26. Carninci, P. et al. Genome-wide analysis of mammalian promoter architecture and
evolution. Nat. Genet. 38, 626–635 (2006).
27. Li, Q., Peterson, K.R., Fang, X. & Stamatoyannopoulos, G. Locus control regions. Blood
100, 3077–3086 (2002).
28. Schomig, E. et al. Molecular cloning and characterization of two novel transport
proteins from rat kidney. FEBS Lett. 425, 79–86 (1998).
29. Sekine, T. et al. Molecular cloning and characterization of high-affinity carnitine
transporter from rat intestine. Biochem. Biophys. Res. Commun. 251, 586–591
(1998).
30. Tamai, I. et al. Molecular and functional identification of sodium ion-dependent, high
affinity human carnitine transporter OCTN2. J. Biol. Chem. 273, 20378–20382
(1998).
31. Wu, X., Prasad, P.D., Leibach, F.H. & Ganapathy, V. cDNA sequence, transport
function, and genomic organization of human OCTN2, a new member of the
organic cation transporter family. Biochem. Biophys. Res. Commun. 246, 589–595
(1998).
32. Nezu, J. et al. Primary systemic carnitine deficiency is caused by mutations in a
gene encoding sodium ion-dependent carnitine transporter. Nat. Genet. 21, 91–94
(1999).
33. Shoji, Y. et al. Evidence for linkage of human primary systemic carnitine deficiency
with D5S436: a novel gene locus on chromosome 5q. Am. J. Hum. Genet. 63,
101–108 (1998).
34. Stanley, C.A. Carnitine deficiency disorders in children. Ann. NYAcad. Sci. 1033, 42–
51 (2004).
35. Wang, Y., Ye, J., Ganapathy, V. & Longo, N. Mutations in the organic cation/carnitine
transporter OCTN2 in primary carnitine deficiency. Proc. Natl. Acad. Sci. USA 96,
2356–2360 (1999).
318VOLUME 39 [ NUMBER 3 [ MARCH 2007 NATURE GENETICS
ARTICLES
© 2007 Nature Publishing Group http://www.nature.com/naturegenetics