Mapping the chromosomal targets of STAT1 by Sequence Tag Analysis of Genomic Enrichment (STAGE)

Abstract

Identifying the genome-wide binding sites of transcription factors is important in deciphering transcriptional regulatory networks. ChIP-chip (Chromatin immunoprecipitation combined with microarrays) has been widely used to map transcription factor binding sites in the human genome. However, whole genome ChIP-chip analysis is still technically challenging in vertebrates. We recently developed STAGE as an unbiased method for identifying transcription factor binding sites in the genome. STAGE is conceptually based on SAGE, except that the input is ChIP-enriched DNA. In this study, we implemented an improved sequencing strategy and analysis methods and applied STAGE to map the genomic binding profile of the transcription factor STAT1 after interferon treatment. STAT1 is mainly responsible for mediating the cellular responses to interferons, such as cell proliferation, apoptosis, immune surveillance, and immune responses. We present novel algorithms for STAGE tag analysis to identify enriched loci with high specificity, as verified by quantitative ChIP. STAGE identified several previously unknown STAT1 target genes, many of which are involved in mediating the response to interferon-gamma signaling. STAGE is thus a viable method for identifying the chromosomal targets of transcription factors and generating meaningful biological hypotheses that further our understanding of transcriptional regulatory networks.

Figure 1.

Comparison of the STAT1 STAGE tag library with a simulated randomly generated background library. A background library was generated to simulate STAGE tag libraries by randomly selecting the same number of tags from the genome as the experimental STAGE library. This procedure was repeated 20 times and the values were averaged. Only tags with a single, unique hit on the genome were used in this analysis. The numbers of single-hit tags (Y-axis) were plotted against the frequencies of those tags in the random (gray bars) and experimental (black bars) tag library (X-axis). For frequencies of 2 and above, the STAGE tag library for STAT1 shows a clear enrichment over a randomly generated tag library.

Figure 2.

Determination of optimal window size used for target identification. Windows of different sizes (300, 500, 1000, and 2000 bp) were scanned across the entire genome. For each window, we defined k as the number of single-hit tags found within the window. The number of windows observed for a given k in the STAGE tag data was compared with the number observed in random simulated data. A window size of 500 bp gave an optimal separation between random and real data. Data shown is for a window size of 500 bp. The gray bars indicate log₁₀ of the number of windows detected based on STAT1 tags, with actual numbers of windows at each k listed at the top of the column. The black line shows the decline in the false discovery rate (FDR) with increasing k. The FDR was calculated as the ratio of the number of windows found in the random simulated library to the number of windows detected in the experimental STAT1 library. The raw data for other window sizes is included in Supplemental Table 1.

Figure 3.

(A) STAT1 binding sites in the ENCODE regions. A portion of the ENCODE region ENm002 is shown as displayed in the UCSC Human Genome Browser. Three out of the seven STAT1 binding sites identified by STAGE matched STAT1 binding sites identified by ChIP-chip analysis performed on NimbleGen ENCODE region tiling arrays. Transcripts identified in this region by the GENCODE project are shown in green. The bottom shows raw ratio data as well as peak calls for STAT1 binding sites from NimbleGen ChIP-chip data. (B) Quantitative ChIP analysis of binding sites identified by STAGE. Nine out of 10 binding sites detected by STAGE were validated as true binding loci by quantitative PCR. Columns show fold enrichment of each locus in the ChIP sample relative to input DNA, normalized to an unrelated control locus. STAGE detected two binding sites separated by >1500 bp in the IRF1 promoter which are indicated in the figure. IRF-D indicates the distal (IRF1-distal) and IRF1-P indicates the proximal site (IRF1-proximal). No genes were found in the proximity of the site indicated as chr22-34786430.

Figure 4.

(A) Overlap of STAT1 target genes identified by STAGE with STAT1 target genes identified by ChIP-chip using a core promoter array. STAGE identified 29 promoters out of the ∼9000 promoters present on the core promoter array as STAT1 target promoters. Eleven out of these 29 overlapped with the 157 promoters identified as STAT1 targets by ChIP-chip analysis at an enrichment ratio greater than threefold. The enrichment ratio refers to the ratio of the fluorescence intensity of ChIP DNA to that of reference DNA at each spot on the core promoter microarray. (B) Motif analysis. The Y-axis shows the percentage counts of the number of sites bearing the given motif(s) out of the 381 STAT1 binding sites detected by STAGE. Almost 60% of the 381 binding sites had the STAT1 motif TTCNNNGAA as compared to 27% in the background. We also detected an enrichment for the co-occurrence of the binding motifs for STAT1 and AP1 (TGAG/CTCA), STAT1 and MYC (CACA/GTG), and STAT1 and NFKB (GGGA/GNNC/TC/TCC) in accordance with the fact that STAT1 exhibits cooperative binding with these factors to regulate downstream promoters.