We showed that DNA damage-induced ATF3 binds to nearly one third of the human RefSeq promoters. Identification of such a large number of targets is not exceptional, since E2F1 and MYC are known to be recruited to >20,000 and >17,000 targets, respectively, in the human genome
, and yeast INO4 has been reported to bind to 1,078 targets upon MMS stimulation
. The number of targets of cancer-associated ATF3 was smaller than that of DNA damage-induced ATF3, perhaps reflecting both higher levels and greater fold-induction of ATF3 in MMS response than those achieved by knockdown of ATF3 in LNCaP cells (data not shown). Nevertheless, the assay was remarkably consistent since 89% of targets in LNCaP cells were also found in MMS-treated HCT116 cells, and eleven out of 41 previously known ATF3 targets including EGR1
could be identified. A survey of ATF3 ChIP-seq data of the ENCODE database (http://hgdownload.cse.ucsc.edu/goldenPath/hg19/encodeDCC/wgEncodeHaibTfbs/
) also indicates that significant proportions of ATF3 targets, i.e. 37.6% (768/2041) in K562 (leukaemia), 71.9% (742/1032) in HepG2 (liver), and 76.2% (809/1062) in H1-hESC (embryonic stem cells), are shared with 2711 targets in GM12878 (lymphoblastoid), which might be considered conservative figures given variations of ChIP-seq data quality (which affects the sensitivity of peak detection) and difference in tissue types.
Notably, bindings sites of several other transcription factors induced by DNA damage (NF-Y, E2F, YY1/NF-E1, FOXJ2, Sp1), UV irradiation (USF2, RFX1), and oxidative stress (c-Ets) were significantly over-represented in the ATF3 targets () indicating an extensive network of these stress response factors (Fig. S6
). A possible explanation for such coincidence of stress-related transcription factors is that ATF3 might synergize with these transcription factors to modulate target gene expression. Alternatively, ATF3 targets identified by ChIP analysis might include indirect binding of ATF3 through interaction with other stress-induced transcription factors. Indeed, ATF3 is a member of a large family of bZip type transcription factors, including those identified in the current study (i.e. AP-1, DDIT3, and USF2), which form homo- and hetero-dimers with different affinities to variants of the consensus ATF/CRE motif
. In addition, Sp1 found in the current analysis is known to physically interacts with ATF3
. Further studies will be required to determine the role of stress-inducible transcription factors, especially those which has not been known to interact with ATF3 before, in binding of ATF3 to target genes.
The current study indicates that the impact of ATF3 on each target gene expression is not all-or-none effects. Rather, effects of ATF3 on many but select genes, such as those involved in cell cycle and cell death, collectively amounted to biologically significant outcome as cell death for ATF3 in stress response () or cell growth for cancer-associated ATF3(). Moreover, our knockdown study shows that only 6–30% of the potential targets of ATF3 were directly regulated by ATF3 in line with previous reports showing that only 25% of p53 targets
, 26% of STAT1 targets
, and 11% of yeast transcription factor targets are affected by gene knockdown
. Several mechanisms have been proposed for p53 to explain why only a subset of potential targets respond to manipulations of p53 including epigenetic states of target genes, promoter occupancy with RNA polymerases, and recruitment of essential co-factors
. Similar mechanisms might be responsible for gene-selective effects of ATF3. Alternatively, multiple forms of ATF3 complexes, such as hetero-dimers with different bZip proteins, might underlie differential effects of ATF3 on target genes.
In the past several years, an increasing number of studies have indicated that ATF3 has pleiotropic functions depending on cell context. Importantly, our study revealed that stress-induced ATF3 and cancer-associated ATF3 have opposing effects on p53 and Wnt pathways: p53 pathway is activated in stress response (Fig. S5
) consistent with the function of ATF3 in p53-mediated cell death 
, whereas Wnt pathway is activated in cancer as previously reported in a study on ATF3 transgenic mice developing mammary tumours 
. One might argue that the difference in genetic backgrounds and/or tissue types between HCT116 cells and LNCaP cells could have influenced the function of ATF3. Indeed, ATF3 appears to play distinct roles in different tissues: ATF3 acts as a tumour suppressor in colorectal cancer
while it is oncogenic in prostate cancer
, mammary cancer
, skin cancer
, and Hodgkin's lymphoma
. Our findings in HCT116 cells (colon) and LNCaP cells (prostate) are consistent with such a hypothesis.
Alterations of the regulatory function of ATF3 is highlighted by our finding that ATF3 is required for both activation and repression of pro-apoptotic genes such as BBC3/PUMA and TNFRSF10B/DR5 in stress response and cancer. Of note, opposing effects of ATF3 on cyclin D1 expression have been previously documented: ATF3 binds to AP-1 motif and activates cyclin D1 in mitogen-stimulated mouse hepatoma cells
, whereas it binds to ATF/CRE site and represses cyclin D1 in mouse fibroblasts stimulated by serum
. In the literature, there are ample precedents of transcription factors which have context-dependent opposing functions on cancer development (
and reference therein). KLF4, for instance, activates p21 and represses p53 both of which are suppressed by activated Ras resulting in growth arrest in the absence of Ras or transforming phenotype in the presence of Ras. Alternatively, a recent study has shown that Kruppel-like factor 5 (KLF5) is required for MYC transcription in proliferating epithelial cells but is essential for TGFb-mediated repression of MYC
. Differential binding of KLF5 to TGFβ inhibitory element in the presence or absence of TGFβ was proposed as a mechanism of the opposite effects. Further studies will be required to determine if combinations of ATF3 and other transcription factors can switch the function of ATF3 on specific targets.
Resistance of tumour cells to various stress conditions remains an important issue. However, our knowledge about the role of cellular stress response machineries in cancer resistance to hypoxia
or chemotherapeutic agents
is still limited. Our finding that ATF3 has opposing effects on pro-apoptotic genes suggests that care must be taken to either enhance or block the function of ATF3 in a novel approach to cancer therapy. Further understanding of the switching mechanism of ATF3 function as a transcriptional activator or repressor might help develop strategies to selectively manipulate a subset of ATF3 target genes to assist the treatment of chemotherapy-resistant cancers.