Intrinsic promoter activities of primary DNA sequences in the human genome.

Abstract

In order to understand an overview of promoter activities intrinsic to primary DNA sequences in the human genome within a particular cell type, we carried out systematic quantitative luciferase assays of DNA fragments corresponding to putative promoters for 472 human genes which are expressed in HEK (human embryonic kidney epithelial) 293 cells. We observed the promoter activities of them were distributed in a bimodal manner; putative promoters belonging to the first group (with strong promoter activities) were designated as P1 and the latter (with weak promoter activities) as P2. The frequencies of the TATA-boxes, the CpG islands, and the overall G + C-contents were significantly different between these two populations, indicating there are two separate groups of promoters. Interestingly, similar analysis using 251 randomly isolated genomic DNA fragments showed that P2-type promoter occasionally occurs within the human genome. Furthermore, 35 DNA fragments corresponding to putative promoters of non-protein-coding transcripts (ncRNAs) shared similar features with the P2 in both promoter activities and sequence compositions. At least, a part of ncRNAs, which have been massively identified by full-length cDNA projects with no functional relevance inferred, may have originated from those sporadic promoter activities of primary DNA sequences inherent to the human genome.

Full text

Page 1

Intrinsic Promoter Activities of Primary DNA Sequences
in the Human Genome
Yuta SAKAKIBARA1,3, Takuma IRIE1, Yutaka SUZUKI1,*, Riu YAMASHITA2, Hiroyuki WAKAGURI1, Akinori KANAI1,
Joe CHIBA3, Toshihisa TAKAGI1, Junko MIZUSHIMA-SUGANO1,4, Shin-ichi HASHIMOTO5, Kenta NAKAI2and Sumio SUGANO1
Graduate School of Frontier Sciences, the University of Tokyo, 4-6-1 Shirokanedai, Minatoku, Tokyo 108-8639,
Japan1, Human Genome Center, The Institute of Medical Science, the University of Tokyo, 4-6-1 Shirokanedai,
Minatoku, Tokyo 108-8639, Japan2, Faculty of Industrial Science and Technology, Tokyo University of Science, 2641
Yamazaki, Noda-shi, Chiba 278-8510, Japan3, Laboratory of Viral Infection II, Kitasato Institute for Life Sciences,
Kitasato University, 5-9-1 Sirokane Minato-ku, Tokyo 108-8641, Japan4and School of Medicine, the University of
Tokyo, 7-3-1 Hongo, Bunkyoku, Tokyo 113-0033, Japan5
(Received 14 December 2006; revised 15 March 2007; published online 23 May 2007)
Abstract
In order to understand an overview of promoter activities intrinsic to primary DNA sequences in the human genome
within a particular cell type, we carried out systematic quantitative luciferase assays of DNA fragments corresponding
to putative promoters for 472 human genes which are expressed in HEK (human embryonic kidney epithelial) 293 cells.
We observed the promoter activities of them were distributed in a bimodal manner; putative promoters belonging to
the first group (with strong promoter activities) were designated as P1 and the latter (with weak promoter activities) as
P2. The frequencies of the TATA-boxes, the CpG islands, and the overall G þ C-contents were significantly different
between these two populations, indicating there are two separate groups of promoters. Interestingly, similar analysis
using 251 randomly isolated genomic DNA fragments showed that P2-type promoter occasionally occurs within the
human genome. Furthermore, 35 DNA fragments corresponding to putative promoters of non-protein-coding tran-
scripts (ncRNAs) shared similar features with the P2 in both promoter activities and sequence compositions. At
least, a part of ncRNAs, which have been massively identified by full-length cDNA projects with no functional
relevance inferred, may have originated from those sporadic promoter activities of primary DNA sequences inherent
to the human genome.
Key words: human genome; promoter; transcriptional start site
1.Introduction
With the unprecedented amount of data produced from
both human genome1and transcriptome2–6projects, a
new challenge of genome science is to decode how the
code of the genomic DNA collectively realizes the tran-
scriptome.7To this end, it should be the first step to
understand the code of DNA sequence to control the tran-
scriptional initiation. Although, levels of transcripts are
also controlled at post-transcription-initiation steps such
as transcription elongation, splicing, nuclear export,
degradation steps, and so on,8transcription initiation is
the first step in the gene expression of every gene and is
supported to play fundamental roles.9,10
It is known that, in many cases, the genomic regions
proximal to the transcriptional start sites (TSSs), which
are called promoters, play pivotal roles in determining
the rate of transcription initiation by serving as direct
docking platforms for the RNA polymerase II complex.9
However, the comprehensive view is still obscure regard-
ing which part of the human genome the RNA polymer-
ase II is recruited to and at what frequency the
transcription is initiated. There are currently only 428
Communicated by Minoru Ko
To whom correspondence should be addressed. Tel/Fax. þ81 4-
7136-3607. E-mail: ysuzuki@k.u-tokyo.ac.jp
*
#The Author 2007. Kazusa DNA Research Institute.
The online version of this article has been published under an open access model. Users are entitled to use, reproduce, disseminate, or display the open access
version of this article for non-commercial purposes provided that: the original authorship is properly and fully attributed; the Journal and Oxford University
Press are attributed as the original place of publication with the correct citation details given; if an article is subsequently reproduced or disseminated not in its
entirety but only in part or as a derivative work this must be clearly indicated. For commercial re-use, please contact journals.permissions@oxfordjournals.org
DNA RESEARCH 14, 71–77, (2007) doi:10.1093/dnares/dsm006

Page 2

genes whose transcriptional regulation is well understood
(TRANSFAC7.2).10Besides, even for those genes, the
promoter activities were measured under arbitrary exper-
imental conditions, and are therefore inappropriate for
use in deciphering the transcriptional network taking
place in a particular cellular context or for comprehensive
quantitative analyses of promoter activities encoded by
DNA sequences.
We have developed a method to construct a full-length
cDNA library, which we named the oligo-capping,11and
have accumulated 1.8 million 50-end sequences of putative
full-length cDNAs.12By utilizing the full-length cDNA
information, we have been identifying exact genomic pos-
itions of the TSSs and characterizing the adjacent
sequences as putative promoter regions (PPRs).13So far
we have made the collected information of the TSSs and
PPRs for about 15 000 human genes available through
our database, DBTSS (http://dbtss.hgc.jp/).
In the present study, we attempted to obtain quantitat-
ive experimental data about the promoter activities
within a particular cell, since such a data set should
serve as a firm foundation for exploring the transcrip-
tional network of human genes. We first collected and
analyzed the promoter activities of DNA fragments corre-
sponding to PPRs of genes which are expressed in human
embryonic kidney epithelial 293 cells (HEK293 cells).
Then, we analyzed the promoter activities of genomic
DNA fragments which were randomly isolated from
non-promoter regions. By this, we wished to understand
the promoter activities intrinsic to primary DNA
sequences. Unexpectedly, analysis of those DNA frag-
ments led us to hypothesis that sometimes even average
human genomic sequences come to have significant pro-
moter activities and that a class of non-protein coding
RNAs should be originated from sporadically occurring
promoter activities of DNA sequences, which is inherent
to a long genome such as human’s. Here we report our sys-
tematic experimental characterization of the promoter
activities of primary DNA sequences using a uniform
experimentalprocedure and
conditions.
standardized cellular
2.Material and methods
2.1.Computational procedures
Oligo-cap cDNA library was constructed from HEK293
cells (ATCC number CRL-1573) according to the pro-
cedure described previously. 12 504 one-pass sequences
were produced and mapped onto the human genomic
sequences (hg_17; http://genome.ucsc.edu/) as previously
described.11(We also confirmed that the update to hg_18
would make essentially no influence on the results.) The
DNA sequences proximal (21 kb to þ200 bp) to the 50-
ends of the mapped oligo-capped cDNAs were retrieved
as PPRs. PPR region was defined as 21 kb to þ200 bp,
because, in many cases, TSSs are distributed over
?100 bp wide and there are usually many TSSs located
downstream ofthe representative
Supplementary Fig. 1B). In order to cover most of those
fluctuating TSSs, PCR primers were set using PRIMER3
between þ100 to þ200 bp. As a result, many of the
PPR clones came to contain so-called upstream ATGs.
For detailed discussion on the possible influences of those
ATGs, see Supplementary Information Fig. 8. For details
of the PPR sequences and the designed PCR primers, see
Supplementary Tables 1–3 and 5.
For prediction of the TATA boxes,9a matrix search
program, MATCH, was run using the position weight
matrices of V$TATA_C (http://www.gene-regulation.com/
cgi-bin/pub/databases/transfac/getTF.cgi?AC=M00216)
andV$TATA_01(http://www.gene-regulation.com/cgi-bin/
pub/databases/transfac/getTF.cgi?AC=M00252) as for the
TRANSFAC7.2 database. Hits observed between 290 and
þ27 relative to the TSSs in the plus strands with cut-off
value of 0.77 were scored as positives. For the statistical test,
Wilcoxon test was used.
TSS12
(alsosee
2.2. Cloning of the DNA fragments of PPRs and
luciferase reporter gene assays
Using the PCR primers designed for randomly selected
1000 PPRs (out of 2170 PPRs which were identified to be
active in HEK293 cells; see Supplementary Fig. 1), 50 ng
of human genomic DNA (Clontech) was amplified by
PCR with 10 pmol of the 50- and 30-PCR primers and
with the 35 reaction cycles of 948C, 1 min; 588C, 1 min;
688C, 2 min; using a KOD Plus PCR kit (ToYoBo). For
cloning the random genomic DNAs, the human genomic
DNA was amplified by a similar procedure using PCR
primers solely containing attB1 and attB2 sites with the
following relaxed PCR conditions: 20 reaction cycles of
948C, 1 min; 408C, 1 min; 728C, 1 min; using a ExTaq
PCR kit (TaKaRa). The amplified fragments were size
fractionated by agarose gel electrophoresis and fragments
of 1.0–1.2 kb were recovered. After the amplified frag-
ments’ lengths were confirmed by agarose gel electrophor-
esis, PPR and random genomic DNA fragments were
cloned into a luciferase reporter gene construct using
the Gateway System (Invitrogen; http://www.invitro
gen.com/content.cfm?pageid=4072). We employed this
multi-faceted cloning system, so that the obtained promo-
ter clones could be used in other vector systems in future
experiments. For Gateway cloning, we performed ‘one-
tube reactions’ according to the supplied instructions.
The insert sequences were determined from both ends.
As for ‘random’ genomic sequences, we checked their
genomic positions using UCSC Genome Browser and
found no explicit bias (Supplementary Table 2). We
also found no difference in the distribution patterns of
theG þ C
genomic fragments and computationally ‘randomly’-
contentsbetweentheclones ‘random’
72Promoter Activities in the Human Genome[Vol. 14,

Page 3

extracted genomic sequences (also see Supplementary Fig.
7). Although it is still possible that undetected bias might
have been introduced by PCR amplifications, we con-
sidered the obtained clone set mostly represented
‘random’ genomic sequences.
For transient transfection of the promoter clones, the
DNAs were purified using QIAwell 96 Ultra (QIAGEN).
HEK293 cells were cultured in 96-well micro-titre plates
with a cell density of 1.0 ? 104per well. For each well,
50 ng of the promoter clones together with the 5 ng of
pTK-Renilla were transiently transfected into HEK293
cells using 0.3 ml of Fugene6 (Roche). Forty-eight hours
after the transfection, dual luciferase assays were per-
formed according to the manufacturer’s instructions.
The above procedures were repeated three times as inde-
pendent cell culture and transfection experiments. For
86% of the data from 472 PPRs and 251 random frag-
ments, estimated signal versus experimental noise (aver-
aged luciferase activity versus the standard deviation of
them calculated from the three experiments) ratio was
less than 0.25. Identification, physical cloning, and lucifer-
ase assays of the PPRs of the ncRNAs were performed
similarly with the cases of the PPRs.
3.Results and discussion
3.1.Luciferase assays of the DNA fragments
corresponding to PPRs
In order to understand the overview of the promoter
activities of primary DNA sequences in the human
genome in a particular cell, we attempted to collect
DNA fragments of PPRs which are active in HEK293
cells. For this purpose, we first collected the TSS infor-
mation of the genes expressed in HEK293 cells by sequen-
cing a 50-end-enriched cDNA library constructed using
our oligo-capping method10(schematic representation of
the workflow is shown in Supplementary Fig. 1A). The
overall full-lengthness of the cDNA library was estimated
to be 85% (Supplementary Fig. 2). By mapping 12504
50-endscDNAsequences
BP870448–BP873619; BP244227–BP249739), we identi-
fied positional information of the PPRs, which are
expressed in this cell line. The genomic DNAs correspond-
ing to 21 kb to þ200 bp were amplified by PCR and 472
PPRs were physically cloned into the luciferase vector
(for examples, see Supplementary Figs 1B and C). In
the present study, we focused on 21 kb to þ200 bp
regions as PPRs (82% of the previously characterized
regulatory elements were shown to be located within
this range; see Supplementary Fig. 3).
We did not take the distal regulatory elements, such as
enhancers and locus control regions, into consideration.
Also, we did not assess the effects of chromatin structure
or epigenetic modulations.8Actually, in the reporter gene
assays used in this study, influences of such factors are
(Genbank accessions:
not reflected in exchange for its advantages in high
throughput and quantitativeness.
focused on each gene would be indispensable for those pur-
poses. Rather, we considered it should be prioritized to
characterize the promoter activities intrinsic to the naked
primaryDNAsequencesoftheveryadjacentoroverlapping
regions of the TSSs, providing actual docking platforms of
transcriptional machineries. (In this issue, extensiveanaly-
sis, comparing the activities of primary DNAs of PPRs
observed using luciferase assays and the eventual mRNA
expression levels observed using SAGE and microarray, is
also presented in Supplementary Fig. 4.)
Using the 472 successfully cloned PPRs, luciferase
assays were systematically carried out under standardized
cell culture conditions (Fig. 1). The assay was repeated at
least three times for each of the PPRs (for raw data, see
Supplementary Tables 1–3). We further selected 445
PPRs that were supported by more than three indepen-
dently isolated oligo-cap cDNAs. This was done to
assure that the frequency of the erroneously identified
PPRs due to errors of the oligo-capping would be mini-
mized (see in what follows). For the later analyses, we
used the 472 PPR data set. Essentially, the same results
were obtained using the selected 445 PPR data set
(Fig. 1D; for the results of each of the following analyses,
see Supplementary Fig. 5).
Intensive analysis
3.2.Two separate populations of PPRs
As shown in Fig. 1C, the observed promoter activities
seemed to show a bimodal pattern of distribution. When
we separated the PPRs with the luciferase activities of
.100.8fold as P1 and those with ,100.8fold as P2
(Table 1), we found that the bimodality of the distri-
butionwasstatistically significant
Table 4). Furthermore, sequence characterization of the
PPRs revealed that there was a qualitative difference
between the sequence features of P1 and P2 (Table 2).
For P1,thefrequencies
(TATA[T/A][T/A]), less-strict TATA boxes (TATA-like
elements; hits from the matrix search with relaxed para-
meters: see Material and methods) and CpG-islands
were 7, 20, and 63%, respectively. The overall G þ C
content was 0.54, which is far more GC-rich than the
average G þ C content of the entire human genome
(0.45). This is in good agreement with our previous stat-
istical analysis of PPRs.13Also, within P1, the presence of
the strict TATA-containing PPRs was enriched in the
PPRs whose promoter activities were in the top 25% com-
pared with those whose activities were in the bottom 25%
(18 and 3%, respectively; P, 0.001). Generally, the
sequence features of P1 were consistent with the previous
view that the promoters are embedded in a relatively
G þ C rich sequence context, often associated with CpG
islands, and the view that because the presence of the
canonical TATA box provides the optimal docking
(Supplementary
ofstrictTATA boxes
No. 2] Y. Sakakibara et al. 73

Page 4

platform for RNA polymerase II, it drives the strongest
promoter activity.8It should be the next step analysis
to further narrow down the observed promoter activities
are realized by what range of the DNA sequences within
the PPRs.
In contrast, P2 was far more AT-rich (G þ C content ¼
0.47) than P1 (P ,6.0 ? 10210; also see Supplementary
Table 6). CpG islands were far less frequent in P2 (13%;
P,1.0 ? 1026). Although the frequency of strict TATA
boxes in P2 was similar to that in P1 (7%), that of the
less-strict TATA-boxes was
P,1.0 ? 1022). These sequence features, distinct from
those of P1, were somewhat different from the features
included in the classical view of promoters. Relative
enrichment of the less-strict TATA boxes may indicate
that sequences favourable for the binding of TATA-
binding protein might be indispensable for P2, whose
members otherwise meet few of the requirements of pro-
moters as conventionally understood. Actually, G þ C
contents of P2 containing strict TATA boxes were lower
than those of P1 similarly containing strict TATA boxes
(Fig. 2). Again, this result indicates that a TATA box
embedded in a relatively G þ C rich sequence should be
necessary for realizing strong promoter activity.
We considered it unlikely that P2 consisted pre-
dominantly of erroneously identified PPRs. First, the fide-
lity of the identified PPRs should have increased, as the
number of supporting oligo-cap cDNAs increases. As
shown in Fig. 1D, among 472 PPRs, 445 (including 47 of
the PPRs belonging to P2) were supported by more than
three independently isolated oligo-cap cDNAs. Also,
when the PPRs with more than three supporting oligo-
cap cDNAs were used for all of the analyses, essentially
much higher(36%;
Figure 1. Luciferase activities of the PPRs. Luciferase activities of the PPRs (A) and the randomly isolated genomic fragments (B) Error bar
indicates the standard deviation of each assay. (C) Distribution of the luciferase activities of the PPRs (gray bars), random genomic
fragments (blank bars) and PPRs of the ‘ncRNAs’ (solid bars). (D) The distribution of the PPRs which are supported by more than three
oligo-cap cDNAs is shown by blank bars. The average luciferase activity of the random genomic fragments was designated as 1 for all of the
analyses. Details of the methods are provided as supporting information.
Table 1. Luciferase activities and the classification of the PPRs
Luciferase activity
.103
103–102
102–101
101–100
100–101
,1021
P P1P2
0
G G1
0
G2 ncRNA
02200
17517500001
21721703303
6317 46433013 16
150151960196 15
0009090
472411 61251 33218 35
Statistical significances of the marked positions are shown in the
margin.
74Promoter Activities in the Human Genome[Vol. 14,

Page 5

the same results were obtained (Supplementary Fig. 5).
Secondly, even for the PPRs in P2, it was not the case
that the promoter activities were not observed at all.
Although weak, the activities of these PPRs were clearly
higher than most of the promoter activities observed for
randomly isolated genomic fragments (G2: see in what
follows). Lastly, supporting evidences have been reported.
Although their purpose was different from ours, Trinklein
et al.14also performed luciferase assays for 152 kinds of
PPRs identified from full-length cDNAs in HEK293 cells.
Their results also indicated similar bimodal patterns of
the promoter activities. Moreover, Versteeg et al.15
reported that human genes with high expression levels
tend to be located in GC-rich regions and genes with low
expression levels tend to be located in AT-rich regions
according to their genome-wide ‘human transcriptome
mapping’ analysis, which could be interpreted as the fea-
tures of P1- and P2-driven genes, respectively.
3.3. Possible universal promoter activities of the DNA
sequences in the human genome
We then attempted to examine the promoter activities
of average human genomic DNA fragments in contrast to
the observed PPR activities. We randomly isolated 251
non-genic genomic fragments of approximately the same
length and measured their promoter activities (Fig. 1B,
Supplementary Table 2; also see Material and methods).
As shown in Fig. 1C, unexpectedly, we occasionally
observed promoter activities comparable with those of
P2. The subpopulation of the genomic fragments whose
promoter activities were more than average of the P2
was designated as G1 and the others as G2. As shown
in Table 2, the overall G þ C content of the G1 was
similar to that of P2, and CpG islands were not observed
at all. Interestingly, the frequency of TATA boxes (both
strict and less-strict) in G1 was the highest in all of the
populations. It is possible that TATA-like sequences
present among the relatively AT-rich genomic sequences
may happen to possess the minimal capacity to provide
a sufficient docking platform for the RNA polymerase II
complex (including the TATA binding protein), allowing
the corresponding genomic regions to serve as ‘pseudo-’
promoters. Among them, some might have been fixed as
P2 as a result of the fact that the downstream sequences
became subjected to functional constraints as genes
during the course of evolution.
3.4.Possible origin of a class of ‘ncRNAs’
From the analyses of promoter activities of primary
DNA sequences of PPRs and random genomic sequences,
we unexpectedly observed that the randomly isolated
genomic fragments occasionally resembled P2 PPRs and
displayed some promoter activities. This observation led
us to hypothesize that this type of promoter activity
could explain the origin of a population of a novel class
of transcripts, non-protein-coding RNAs (ncRNAs).
Recently, both human and mouse full-length cDNA pro-
jects have demonstrated that there is an unexpectedly
large number of transcript species that are unlikely to
encode any proteins. For example, our FLJ project
isolated 768 human full-length cDNAs that should be
categorized as ‘ncRNAs’. Similarly, the FANTOM
project also identified 4280 mouse ‘ncRNAs’ (‘ncRNA
core’).5,16,17The biological significance of this emerging
class of ‘ncRNAs’ is currently of great interest, especially
for those which are sometimes called, ‘mRNA-like long
ncRNAs’ or ‘Transcripts of Unknown Functions’18
(we simply call them as ‘ncRNA’ hereafter).
We first compared those so-called ‘ncRNAs’ identified
from human and mouse full-length cDNA projects with
Figure 2. G þ C content of the PPRs. Box plot chart of the G þ C
content of the indicated population of the PPRs is shown.
Table 2. Sequence features of the PPRs and genomic fragments
P (%)
267 (57)
P1 (%)
259 (63)*
P2 (%)
8 (13)
G (%)
0 (0)
G1 (%)
0 (0)
G2 (%)
0 (0)CpG island
TATA box: strict34 (7) 30 (7)4 (7) 47 (19) 7 (21) 40 (18)
TATA box: less strict 103 (22) 81 (20)**22 (36) 140 (56)20 (61) 120 (55)
Average G þ C content
Total
0.530.54***0.47***0.450.430.45
472 41161251 33218
No. 2]Y. Sakakibara et al. 75