High-resolution and genome-wide profile of transcriptional regulatory input and output in the differentiation time course

We performed a wide range of genome-wide experiments for the THP-1 cells (Table 1). To ensure the consistency and comparability of the data taken, all experiments were based on cells that were cultured and cultivated from one initial cell batch (Additional data file 1). The differentiation time course was analyzed in three independent experiments (biological replicates). The expression of all genes was monitored over the time course with microarrays to ensure that the known THP-1 specific marker genes responded correctly. For example, a monocytic marker gene, CD11b (ITGAM), is up regulated in all three biological replicates (Figure 1).

We also profiled TSSs by sequencing of 5'-end tags of mRNAs, CAGE tags, employing the 454 next-generation sequencer [8]. For six time points after PMA stimulation of THP-1 cells, we profiled all three biological replicates independently, resulting in 24 million CAGE tags. We defined 1.9 million individual TSSs (level 1 promoters) giving single base pair resolution of transcriptional initiation; 30,000 promoters (level 2 promoters) contained neighboring TSSs with similar expression profiles over the time course (TSSs were merged to form level 2 promoters); 15,000 promoter regions (level 3 promoters) contained continuous promoters separated by no more than 400 bp distance on the genome (level 3 promoters). We assigned expression profiles according to all of the three levels of transcription initiation based on CAGE data so that promoter activities between independent promoters as well as between the biological replicates can be compared [7]. In general, there was a strong correlation between data from microarray analysis and CAGE profiling. For example, the profiled promoter activity of the marker gene CD11b shows up-regulation in all three biological replicates, which is consistent with the observed microarray-based gene expression (Figure 1). However, in some genes, such as PU.1, there is divergent regulation of independent promoters [7].

For two time points, in the beginning and at the end of the time course, whole genome ChIP-chip experiments for the epigenetic mark of H3K9 acetylation and for the interaction of RNA polymerase II with DNA were performed. Both show substantial (approximately 80%) overlap with the promoter sets defined by deepCAGE [7], while there is no complete overlap as shown in a previous study [9]. Reliable expression over the biological replicates as well as the reasonable coincidence of TSSs and the chromatin status demonstrate the reproducibility and reliability of the experimental data. Additional ChIP-chip experiments for two transcription factors known to be important in macrophage differentiation, SP1 and PU.1, were performed using promoter tiling arrays for two biological replicates. Both factors bind to the promoter region of CD11b, which is consistent with previous studies [10, 11].

Transcriptional regulatory network analysis and 52 systematic siRNA perturbation experiments

We predicted transcriptional regulatory interactions employing a hybrid approach of: transcription factor binding site (TFBS) predictions within evolutionarily conserved regions that are proximal to the CAGE defined promoter regions (-300 bp to +100 bp) using TFBS matrices; and microarray derived gene expression. To validate the predicted interactions, we performed large-scale siRNA perturbation knockdown experiments for 52 transcription factors that are expressed in the undifferentiated state, including the two transcription factors used in the ChIP-chip experiments, SP1 and PU.1. We monitored the effects on gene expression with microarrays at 48 h after transfection. All transfection and subsequent microarray experiments were conducted in biological triplicates. We observed down-regulation of the CD11b gene upon knockdown of SP1 and PU.1, which is consistent with their binding to the CD11b promoter region as well as with previous studies [10, 11].

Interestingly, our large-scale perturbation study revealed that knocking down of Myb induced the expression of CD11b and many other genes that are up-regulated during THP-1 differentiation, indicating that Myb can work directly or indirectly as a transcriptional repressor [7]. The expression of Myb is itself rapidly repressed in response to PMA.

Quantitative RT-PCR for transcription factors, small RNA sequencing, and update of CAGE data

Additional data, in-depth sequencing of small RNA and large-scale quantitative RT-PCR (qRT-PCR) of 2,000 transcription factors were performed subsequently to complement the analysis. The deep sequencing of small RNAs, ranging from 10 to 82 nucleotides, uncovered a novel class of transcription initiation-associated short RNAs with a length around 18 nucleotides [12]. Moreover, four novel microRNAs were found together with general aspects of microRNA expression during THP-1 differentiation (manuscript under review). The viewer also incorporates large-scale qRT-PCR expression profiling of 2,000 transcription factors along the time course for two biological replicates and gives detailed expression levels for these transcription factors, which are generally difficult to capture with conventional microarrays [13]. Additional ongoing experiments, including data obtained from primary monocytes and monocyte-derived macrophages, will be continuously incorporated into this web resource as soon as they are published.

In addition to the above data focusing on THP-1, we updated the genome mapping of the 5 million FANTOM3 CAGE tags obtained from 41 human samples [3]. We incorporated other published human RIKEN CAGE tags as well, resulting in a total of 29 million mapped tags from 127 human RNA samples. Genomic mapping of previously published RIKEN mouse CAGE tags were updated to the latest genome assembly, mm9 (NCBI build 37), resulting in 11 million mapped tags from 206 mouse RNA samples. All of these CAGE data were processed in a uniform way in terms of alignment to the genome and aggregation into promoters as 'tag clusters' consistent with previous work [3, 14].

The database and its interface

All data are accumulated into a single web resource consisting of two databases, the Genome Browser and the EdgeExpressDB [15], as well as a repository for data downloads (Figure 2). The genomic data derived from our experiments are displayed via a genome browser interface (generic genome browser, GBrowse) [16] enabling the inspection of any locus of interest. The browser was configured to show multiple experimental results and multiple conditions at the same time in small 'pop-up' windows within one single browser window (Figure 3), which is implemented with the coolTip library [17]. For example, a mouse click on 'CD11b promoter' produces a small window presenting quantitative promoter activity, in particular, up-regulation of this promoter in THP-1 cells in response to PMA. A mouse click on the microarray probe corresponding to this gene produces a small window depicting its expression, which also shows increased abundance of the mRNA of CD11b. The parallel display of these two windows enables the examination of the relationship between promoter activities identified by CAGE and gene expressions measured by the microarray (Figure 3). Tissue-wide promoter activities derived from CAGE tags sequenced before the FANTOM4 project are available in the same interface (Figure 3), making it possible to examine tissue- or cell-specific expression.

The FANTOM4 EdgeExpressDB [15] facilitates the inspection of regulatory interactions and expression profiles in the same context, which is impossible using the genome browser only. It stores predicted and validated regulatory relationships as well as all of the corresponding expression profiles. The EdgeExpressDB and the genome browser are cross-referenced with hyperlinks (Figure 3). The two interfaces visualize complementary aspects of the same complex data.

In addition to the graphical user interface, we have prepared an application interface (API) for direct access using the DAS protocol [18], which is widely used to exchange genome annotation information. The available tracks and their entry points are described in Table 1 and Additional data file 2.

Standardized meta-data describing large-scale multi-dimensional experiments

A wide-range of large-scale experiments with several replicates was performed for the THP-1 differentiation in response to PMA stimulation. Analyses and findings are being reported in a series of separate publications, each of them employing parts of the complex data. Understanding interconnections between experiments and the data as a whole is challenging. To describe the whole set of experiments in a consistent way, we employed the MAGE/ISA-tab file format [19, 20], a standard format to describe experimental details (available in the 'Download' section of the FANTOM web resource). The file connects all experiments on the level of cell cultivation and biological replicates even if they were used in different analyses. The experimental steps described in the file are visualized with SDRF2GRAPH [21] to support intuitive understanding of the complex experimental steps (Additional data file 1). These meta-data files help to document the data structure of the FANTOM4 project and support its use and biological interpretation. While the FANTOM3 data were centered around the CAGE technology [14], we now explicitly describe the relationships between all types of the experiments in a comprehensive way based mainly upon genomic location.

Cell culture and RNA extraction

The THP-1 cell line was subcloned by limit dilution and one clone (clone 5) was selected for its ability to differentiate relatively homogeneously in response to PMA. THP-1 cells were used for all subsequent experiments. THP-1 cells were cultured in RPMI, 10% fetal bovine serum, penicillin/streptomycin, 10 mM HEPES, 1 mM sodium pyruvate and 50 μM 2-mercaptoethanol. THP-1 was treated with 30 ng/ml PMA (Sigma, St Louis, MO, USA) over a time course of 96 h. Total cell lysates were harvested in TRIzol reagent (Invitrogen, Carlsbad, CA, USA) at each time point. Undifferentiated cells were harvested in TRIzol reagent at the beginning of the PMA time course. Total RNA was purified from TRIzol lysates according to the manufacturer's instructions.

DeepCAGE

The preparation of the CAGE library from total RNA was a modification of methods described by Shiraki et al. [26] and Kodzius et al. [6], adapted to work with the 454 Life Sciences sequencer. The sequenced CAGE tags were firstly mapped onto the latest genome assemblies, hg18 (NCBI build 36.1), using an in-house developed program, nexAlign, and post-processed to consider CAGE tags mapped to multiple loci [27]. The mapped tags were then clustered into three different levels of detail: on individual TSSs giving single base pair resolution for each CAGE tag; on the level of promoters joining neighboring TSSs with similar expression profiles over the time course; and on the level of promoter regions containing continuous promoters separated by a distance of 400 bp on the genome. This computation was performed subsequent to expression normalization to enable comparisons between the biological replicates, resulting in a set of TSS positions, promoters and promoter regions with genomic locations and normalized expression for each time point in the differentiation time course.

Illumina microarray analysis

THP-1 samples were identical to those used for deepCAGE libraries, and RNA was purified for expression analysis by RNeasy columns (Qiagen, West Sussex, UK), FastPure RNA Kit or TRIzol (TAKARA BIO, Otsu, Shiga Japan). RNA quality was checked using the Nanodrop (Nanodrop, Wilmington, DE, USA) and Bioanalyser (Agilent, Santa Clara, CA, USA). RNA (500 ng) was amplified using the Illumina TotalPrep RNA Amplification Kit, according to the manufacturer's instructions. cRNA was hybridized to Illumina Human Sentrix-6 bead chips Ver.2, according to standard Illumina protocols. Chip scans were processed using Illumina BeadScan and BeadStudio software packages and summarized data were generated in BeadStudio (version 3.1).

ChIP on chip analysis

THP-1 cells were cross-linked with 1% formaldehyde for 10 minutes and cells were collected by centrifugation and washed twice in cold 1× phosphate-buffered saline. The cells were sonicated for 5-7 minutes with a Branson 450 Sonicator to shear the chromatin (Branson Ultrasonic, Danbury, CT, USA). Complexes containing DNA were immunoprecipitated with antibodies against H3K9Ac (07-352; Upstate Biotechnology, Lake Placid, NY, USA), PU.1 (T-21; Santa Cruz Biotechnology, Santa Cruz, CA), SP1 (07-645; Upstate), and RNA Polymerase II (8WG16; Abcam, Cambridge, MA, USA). The immunoprecipitated sample was incubated with magnetic beads/Protein G (Dynal, Oslo, Norway) for 1 h at 4°C followed by washing. The complexes were eluted from the magnetic beads by addition of 1% SDS and 100 mM NaHCO₃. Beads were vortexed for 60 minutes at room temperature. The supernatants were incubated for 3.5 h at 65°C to reverse the cross-links, and incubated with RNaseA, and then proteinase K, followed by a phenol:chloroform:isoamyl alcohol extraction and ethanol precipitation to recover the DNA. Immunoprecipitated DNA was amplified by either linker-mediated PCR or in vitro transcription followed by synthesis of double-strand cDNA. Amplified DNA was end-labeled with biotin-ddATP and was hybridized to Affymetrix whole genome tiling for Polymerase II (GeneChip Human Tiling 1.0R Array) and H3K9Ac (GeneChip Human Tiling 1.0R Array), and promoter arrays (GeneChip Human Promoter 1.0R Array) for PU.1 and SP1.

siRNA perturbation experiments

THP-1 cells were seeded in 6 cm dishes at a density of 1 × 10⁶ cells/dish for transfection. Transfection was performed with 1.6 μg/ml (final concentration) of Lipofectamine 2000 (Invitrogen) and 20 μM (final concentration) of stealth siRNA (Invitrogen) or 20 μM (final concentration) of pre-microRNA (Ambion, Austin, Tx, USA, or Nihon-shinyaku, Kyoto, Japan) by reverse transfection protocol in accordance with the manufacturers' instructions. Total RNA for Illumina microarray analysis was extracted 48 h after transfection, using the FastPure RNA kit (TAKARA BIO, Ohtsu, Shiga, Japan) in accordance with the manufacturer's instructions. All microarray experiments were conducted in biological triplicate.

Resource availability

The graphical user interface, DAS API, the original GFF files of the genome browser, and all the additional data files are publicly available at [28].