Logo of pnasPNASInfo for AuthorsSubscriptionsAboutThis Article
Proc Natl Acad Sci U S A. Mar 31, 2009; 106(13): 5117–5122.
Published online Mar 12, 2009. doi:  10.1073/pnas.0900473106
PMCID: PMC2664024

Genome-wide discovery of functional transcription factor binding sites by comparative genomics: The case of Stat3


The identification of direct targets of transcription factors is a key problem in the study of gene regulatory networks. However, the use of high throughput experimental methods, such as ChIP-chip and ChIP-sequencing, is limited by their high cost and strong dependence on cellular type and context. We developed a computational method for the genome-wide identification of functional transcription factor binding sites based on positional weight matrices, comparative genomics, and gene expression profiling. The method was applied to Stat3, a transcription factor playing crucial roles in inflammation, immunity and oncogenesis, and able to induce distinct subsets of target genes in different cell types or conditions. A newly generated positional weight matrix enabled us to assign affinity scores of high specificity, as measured by EMSA competition assays. Phylogenetic conservation with 7 vertebrate species was used to select the binding sites most likely to be functional. Validation was carried out on predicted sites within genes identified as differentially expressed in the presence or absence of Stat3 by microarray analysis. Twelve of the fourteen sites tested were bound by Stat3 in vivo, as assessed by Chromatin Immunoprecipitation, allowing us to identify 9 Stat3 transcriptional targets. Given its high validation rate, and the availability of large transcription factor-dependent gene expression datasets obtained under diverse experimental conditions, our approach appears to be a valid alternative to high-throughput experimental assays for the discovery of novel direct targets of transcription factors.

Keywords: chromatin immunoprecipitation, phylogenetic footprint, positional weight matrix, Stat3 binding sites, Stat3 target genes

Functional transcription factor binding sites (TFBSs) can be identified on a genomic scale either by computational approaches or through elaborated procedures such as chromatin immunoprecipitation followed by either genomic microchip hybridization (ChIP on Chip) or deep sequencing (ChIP and Sequencing) (1). These have the advantage of directly measuring the in vivo occupancy of genomic sites. By definition however, each experiment will only be able to identify sites bound under the specific conditions analyzed, i.e., separate experiments will have to be performed for each condition/tissue type of interest, and this will be particularly true for the many transcription factors (TF) that are known to induce distinct sets of genes in different tissues. Indeed, sets of TFBSs identified with these techniques in different conditions often show limited overlap. The predictions based on computational sequence analysis (2), however, are in principle independent of the cellular context. The ample collection of candidate BSs thus produced will then be available to identify transcriptional targets either as such or within lists of differentially expressed genes generated by microarray experiments, in many cases already available through public databases.

The standard way to describe degenerate cis-regulatory elements takes advantage of positional weight matrices (PWM) constructed using multiple alignment algorithms (3). Genomic sequences can then be scanned for sites showing significant similarity to the PWM compared with a background nucleotide distribution. However, because of both the sequence degeneration of TFBSs and the regulatory features of chromatin, restricting the access to DNA, the vast majority of candidate TFBSs thus identified are not functionally relevant. Because functional TFBSs are in most cases under selective pressure, evolutionary conservation provides a powerful mean to filter the results (4).

Here, we present a computational pipeline to predict TFBSs based on PWMs, large scale comparative genomics and gene expression profiling. The method was applied and experimentally validated on the transcription factor Stat3, whose transcriptional targets are well known to be strongly dependent on cellular context (5, 6). Signal transducers and activators of transcription (STAT) factors play a major role downstream of most cytokine receptors (7). The family member Stat3 is activated by a wide variety of cytokines, growth factors and oncogenes (5). Accordingly, a Stat3 null mutation leads to early embryonic lethality, and conditional inactivation has confirmed pleiotropic functions linked to inflammation, regeneration, proliferation and energy homeostasis (6). In addition, Stat3 is constitutively active in as many as 70% of the primary human tumors and is considered an oncogene (8, 9). The multifaceted functions of Stat3 are partly related to its peculiar ability to activate different sets of genes in different cell types and conditions (5). Despite the efforts to identify Stat3-regulated genes that could be responsible of its many functions and represent potential disease-related targets, only a limited number of bona fide transcriptional Stat3 target genes have been identified so far.

We applied our pipeline and a newly constructed PWM to predict Stat3 binding sites on the mouse genome (Stat3-BSs). Highly scoring sites were filtered by evolutionary conservation with 7 vertebrate species followed by integration with microarray gene expression data. This method displays a high predictive power and has allowed us to identify several Stat3 transcriptional targets.


Creation of a Stat3 PWM Model to Identify Potential Stat3 Binding Sites.

A positional weight matrix (PWM) was generated from a pool of 54 Stat3 binding sites characterized for in vitro binding and Stat3 responsiveness (Table S1), using the program MEME (10) as described in Materials and Methods. This PWM differs from the STAT consensus sequence particularly in the information content of the nucleotides in positions 3 and 4 or 6 and 7 (Fig. 1B and Table S2). Our PWM is remarkably similar to the one experimentally determined in Horvath et al. (11).

Fig. 1.
Sequence logo and predicted/experimental affinity of Stat3-BSs. (A) EMSA competion assays. EMSA were carried out using liver nuclear extracts from LPS-treated mice and a radiolabeled HA-SIE probe, using the indicated double stranded unlabeled oligonucleotides ...

We set up a computational strategy able to identify potential Stat3-BSs, each identified by a score given by the logarithmic ratio of the likelihoods computed using the PWM and the background nucleotide frequencies (see Materials and Methods). To asses the quality of our model we selected a representative Stat3-BS identified in the Icam1 promoter (12), showing the highest possible score of 14.53, and generated 9 mutant Icam1 sites by introducing arbitrary mono- and dinucleotide mutations leading to variable differences in score (Table S3). The binding affinities of Icam1 and of the derived mutant sites were assessed by EMSA competition assays, analyzing their ability to compete with the binding of a labeled HA-SIE probe to Stat3 (Fig. 1A). To extend the study, we also analyzed a series of putative Stat3-BSs identified by our program on candidate target genes characterized in the laboratory (Gadd45b, Hmga2, Mkp1, and Egr1) (Fig. 1A and Table S3). All predicted BSs showed strong in vitro binding activity with the exception of Egr1_b, located at position −214 of the mouse Egr1 gene (Fig. 1A). To confirm the good predictive power of our method, we determined the Spearman rank correlation coefficient between scores and binding affinities. The latter were estimated as the percentage of competition obtained with each Stat3-BS as compared with that obtained by competing with an unrelated site. To each binding site we associated the higher of the log-likelihood scores computed on the 2 strands. For 50-fold competitor:probe concentration the Spearman correlation coefficient was 0.680 (P = 0.0014), suggesting that the score computed from our PWM has strong positive correlation to the in vitro binding affinity of the corresponding sequence (Fig. 1C).

Genome-Wide Discovery of Stat3 Binding Sites.

We then performed a genome-wide search to characterize novel Stat3-BSs and their regulated genes. We applied a stringent score threshold of 9.6, determined as described in Materials and Methods, and we scanned the whole mouse genome sequence (sequence assembly NCBI m36) finding a total of 1,355,858 putative binding sites. Such a high number is not unexpected taking into account the loose sequence requirements for Stat3 binding. To these we applied comparative genomics between Mus musculus and 7 different vertebrate species, as detailed in Materials and Methods.

To assess the validity of our method compared with the more direct ChIP and sequencing approach, we compared the evolutionary conserved BSs identified above with 2 recently reported experimental datasets obtained by measuring Stat3 occupancy in vivo in ES cells and 3T3 fibroblasts, respectively (13, 14). We identified the genes associated to the in vivo-bound genomic sequences reported by Chen et al. (13) as described in SI Material and Methods, obtaining a list of 1,575 genes. This was compared with the 146 target genes reported by Snyder et al. (14). The overlap resulted in only 17 genes. In contrast, the overlap between either experimental set and the genes associated to Stat3-BSs conserved between mouse and at least one other species resulted in 1,155 (73%) and 118 (81%) of the genes found by Chen et al. (13) or Snyder et al. (14), respectively. When considering only sites conserved in at least 2 other species, we retrieved 811 (51%) and 83 (57%) of the targets identified by Chen et al. (13) and Snyder et al. (14), respectively. This supports the idea that data obtained in different biological systems display very limited overlap, in agreement with the knowledge that Stat3 can induce distinct target genes in different tissues/conditions. Conversely, our unbiased strategy allows an ample recovery of in vivo occupied promoters in different contexts without performing additional experiments.

To select promising candidates for direct experimental validation we then considered only the sites located up to 10 kb upstream of the Transcription Start Site (TSS) or in the first intron or first noncoding exon, obtaining a total of 9,648 target genes containing BSs conserved in at least one species (Fig. 2A). This number was reduced to 4,339 if only sites conserved with at least 2 species were selected. The enrichment in confirmed Stat3 targets was evaluated according to Fisher's exact test. Twenty-one of thirty-five confirmed target genes were associated with BSs conserved with at least one species (1.74-fold enrichment compared with chance, P = 1.78 × 10−3). In contrast, genes associated with BSs conserved with at least 2 species yielded 16 confirmed targets with the more significant P value of 2.29 × 10−5 and a 2.95-fold enrichment. Higher levels of stringency in site conservation did not further improve the statistical significance. Therefore, we decided to focus on the 4,339 genes with BSs conserved with at least 2 species, to which we will refer in the following as “conserved binding sites” (CBSs). It should be noted that the use of several organisms improved the results with respect to the simple human-mouse comparison, which would select 7,815 genes including 20 confirmed targets, with a 2.04-fold enrichment (P = 2.71 × 10−4).

Fig. 2.
Phylogenetic conservation and distribution of the conserved Stat3-BSs. (A) The number of genes with at least 1 site conserved in n or more species is plotted as a function of n. (B) Distribution of conserved Stat3-BSs located within 5,000 base pairs from ...

Functional analysis of the genes carrying CBSs revealed very strong over-representation, among others, of genes involved in development, transcription factor activity, intracellular signaling cascades, cell–cell signaling, cell motility and adhesion (Table 1). Interestingly, analysis of the position of CBSs relative to the transcriptional start site (TSS) shows a strong over-representation in the 200 base pairs immediately upstream of the TSS (Fig. 2B).

Table 1.
The GO terms most significantly enriched among the genes with at least 1 CBS

Validation of Candidate Binding Sites by Measuring in Vivo Occupancy.

To test the predictive power of our computational approach we intersected the list of genes carrying Stat3 CBSs with a set of genes found to be differentially expressed in untreated or cytokine-treated Stat3−/− or +/+ mouse embryonal fibroblasts (MEFs) (15). This was generated by microarray analysis of untreated or OSM-treated MEFs of either genotype (see Materials and Methods). After ranking the genes common to the 2 lists based first on the degree of conservation and then on the calculated affinity score, we selected for validation the top 10 scoring Stat3-BSs (Table 2) and 5 other candidate BSs chosen on the basis of their biological functions plus 2 already known sites as positive controls, on the c-Fos and Socs3 promoters (Table 3). This latter set includes 1 gene (Flt4) with a Stat3-BS conserved with 1 species only.

Table 2.
The names and gene IDs of the top-10 predicted Stat3-BSs ranked according to phylogenetic conservation and calculated binding score
Table 3.
The names and gene IDs of 7 additional predicted Stat3-BSs manually selected for validation

In vivo occupancy of each selected Stat3-BS was assessed by Chromatin Immunoprecipitation (ChIP) in the same Stat3+/+ or Stat3−/− MEFs used for the microarray analysis, either untreated or treated with OSM for 30 min to activate Stat3 (Figs. 3A and and44A). The score of the validated BSs is compared with the genome-wide score distribution in Fig. S1.

Fig. 3.
Validation of the top 10 candidate Stat3-BSs (shown in Table 2) by ChIP and mRNA expression analysis. (A) ChIP assays were performed with Stat3+/+ and Stat3−/− MEFs, either untreated (NT) or treated with Oncostatin M for 30 min (OSM). ...
Fig. 4.
Validation of selected candidate Stat3-BSs by ChIP and mRNA expression analysis. Seven additional predicted Stat3-BSs (shown in Table 3) were manually chosen for validation on the basis of the associated gene's biological functions. Sites information, ...

Among the top 10 candidate sites selected, 2 were associated to Irf1, already known to be a Stat3/Stat1 target, whereas the other 8 belonged to genes never reported before to be regulated by Stat3 and/or OSM. Remarkably, we could demonstrate in vivo Stat3 binding for all tested sequences with the only exception of the Il18rap −17 site (Fig. 3A), suggesting that our approach allows the identification of functional, in vivo bound, Stat3-BSs with a high degree of confidence. In most cases, binding was detected only after cytokine stimulation. Only 3 BSs (Tspan, Irf1 −7089 and Chd8 −1803) showed comparable Stat3 binding both before and after OSM stimulation (Fig. 3A). The specificity of our ChIP assay is confirmed by the observation that Stat3 binding was never detected with chromatin from the Stat3−/− MEFs used as a negative control (Fig. 3A).

Expression analysis in Stat3+/+ or −/− MEFs plus or minus OSM stimulation showed a good degree of correlation between Stat3 binding and Stat3-dependent mRNA regulation. Indeed, most genes with at least one cytokine-dependent Stat3-BS (i.e., Sipa1, Nme3, Irf1, and Uqcr) also showed a significant increase of their expression levels after OSM stimulation, with the exception of Sdc1. This gene was apparently repressed by Stat3 because its expression was significantly higher in the absence of Stat3 under basal conditions and slightly but significantly induced upon OSM stimulation (Fig. 3B). Although Sipa1, Nme3, and Uqcr were all defectively induced in the Stat3−/− MEFs, Irf1 induction was instead comparable in cells of the 2 genotypes, likely because of its previously reported Stat1-dependent regulation in cells lacking Stat3 (15). Mrps34/Nme3 are adjacent genes both located within <6 kb of the second best ranking Stat3-BS (Table 2). Interestingly, only the expression of Nme3 was induced by OSM, in a Stat3-dependent way (Fig. 3B), suggesting differential regulation of the 2 genes despite their shared 5′ sequences. The Stat3-BS associated to the Tspan7 gene, located in the first intron 73,000 bp downstream of the TSS, was constitutively bound by Stat3. Accordingly, Tspan7 expression levels were significantly reduced in the Stat3−/− MEFs already under untreated conditions, suggesting that Stat3 may be required for Tspan7 basal transcription. Despite in vivo binding on 2 Stat3-BSs, Chd8 was the only gene where Stat3 binding did not correlate with Stat3-dependent and/or cytokine-inducible regulation (Fig. 3 A and B), although specific modulation under different conditions cannot be excluded. Finally, the lack of Stat3-dependent transcriptional regulation of the Il18rap gene correlated with the absence of Stat3 binding on its promoter.

The second group of candidate Stat3-BSs (Table 3) included those on the promoters of the c-Fos and Socs3 genes, 2 well characterized Stat3 transcriptional targets. ChIP analysis confirmed OSM-inducible Stat3 binding on both sites (Fig. 4A), correlating with OSM-dependent induction of their mRNA expression levels as measured by quantitative RT-PCR (Fig. 4B). As already reported, whereas Socs3 induction was completely Stat3-dependent, c-Fos induction was only partially reduced in the absence of Stat3 (16, 17). Remarkably, the putative Stat3-BSs associated to the Nfil3, Gadd45b, IL4ra, and Flt4 genes were all bound in vivo, the first 3 in an OSM-inducible way while Flt4 was bound constitutively (Fig. 4A). Accordingly, OSM induced a Stat3-dependent increase of the Nfil3, Gadd45b and Il4ra mRNA levels, whereas Flt4 expression was defective in the Stat3−/− MEFs both before and after OSM treatment (Fig. 4B). Within this group, we failed to demonstrate Stat3 binding only to the Selp1 Stat3-BS (Fig. 4A), even though Selp1 mRNA expression was strongly defective in Stat3−/− MEFs already under basal conditions. This could be due either to indirect regulation by Stat3 or to Stat3 binding to other sites within the gene (some of which were identified by our program but not tested in this study because of lower species conservation).

To analyze the correlation between BS scores and gene expression in a more general context, we defined a total gene score as the sum of the scores of all sites associated to each gene and computed its correlation with fold-change as measured in the microarray experiments under the OSM-stimulated conditions. For both replicates we found a positive and statistically significant correlation that increases with the conservation stringency, as shown in Table S4.


The true targets of transcription factors can be identified among lists of differentially expressed genes by selecting those carrying potentially functional TFBSs. This requires 2 elements: (i) a way to assign affinity scores with a good level of predictivity and (ii) a suitable method to identify the most likely in vivo functional BSs. The computational strategy used in this work, based on the generation of a literature-based PWM, log-likelihood scoring of the candidate BSs, comparison with 7 other vertebrate genomes and integration with gene expression data, produced a sizeable number of high-confidence predicted functional BSs for Stat3. Indeed, 12 of the 14 newly predicted Stat3-BSs selected for direct experimental validation, associated to 12 candidate target genes, were able to bind Stat3 in vivo. In addition, most of the related genes turned out to be true Stat3 transcriptional targets, that is genes whose expression is regulated in a Stat3-dependent way and that are associated to functional, in vivo bound, Stat3-BSs. Our computational strategy allowed thus the discovery of 9 direct Stat3 transcriptional targets by testing only 12 candidates.

Elemento et al. (18) showed that conservation of specific BSs can occur outside aligned sequences, using an exact word-matching approach. However, this is unlikely to be effective in the case of TFs with highly degenerate consensus sequences, such as Stat3. Our results show that high level of sensitivity can be achieved by using alignment-based comparative genomics with multiple species: This strategy provided a measurable advantage with respect to simple human-mouse conservation, as shown by a more significant enrichment in confirmed target genes. This could be due to divergent species-specific evolution of conserved BSs, a phenomenon already observed in other systems.

With respect to wet bench approaches such as direct identification of in vivo bound sites by ChIP and Sequencing, our method has the advantage of providing lists of BSs independent of the cellular context. ChIP-based methods, in contrast, will provide sets of data only applicable to the specific system analyzed, and will have to be repeated for each condition of interest. Indeed, even with the stringent score cutoff imposed, there was a high degree of overlap between the target genes associated to our CBSs and those identified by ChIP methods in 2 distinct cellular systems (13, 14), testifying to the validity of our approach. The lists of CBSs generated with our method will be a powerful tool to rapidly identify bona fide TF targets in any given cell system for which relevant gene expression data are available.

Despite the multiplicity of Stat3 biological functions and the central role that this factor plays in tumor biology, a relatively low number of direct transcriptional target genes has been so far functionally identified, making of it an ideal testbed for our method. Interestingly, among the functional Stat3-BSs identified, both Tspan7 +73288 and Flt4 +8136, located far downstream from the TSS within the first intron, displayed constitutive Stat3 binding correlating with completely defective expression in the absence of Stat3, suggesting possible actions at the level of chromatin conformation. Constitutive Stat3 binding to a subset of sites in NIH 3T3 cells is shown in ref. 14, although mRNA expression of the corresponding genes was not tested. At present, we cannot say whether Y705 phosphorylated Stat3, detected in low amounts in growing, unstimulated cells, is involved in constitutive binding, or whether noncanonical mechanisms recruiting unphosphorylated Stat3 are involved. Whatever the mechanism, the ability of Stat3 to transcriptionally regulate a subset of target genes under basal conditions may have important implications for its physiology.

Intriguingly, most identified genes are known to have functions correlated with tumor transformation, metastasis, and growth. For example, Sipa1 was identified as the gene responsible for the activity of a metastasis efficiency locus on mouse chromosome 19 and its levels positively correlate with the metastatic capacity of a mouse breast cancer cell line (19). Nme3 is highly expressed in solid tumor cell lines and may contribute to the differentiative arrest of myeloid leukemia cells (20, 21). Tspan7 was reported to behave either as a metastasis suppressor or as a marker of poor prognosis in AML, depending on the cell system (22, 23). Loss of Sdc1, which appears to be negatively regulated by Stat3, correlates with higher malignancy of ductal breast cancer cells and with epithelial to mesenchimal transition of epithelial cancer cells (24). Gadd45b is a prosurvival factor associated to stress resistance in tumors and its overexpression can transform NIH 3T3 fibroblasts (25). Nfil3 mediates IL-3 prosurvival functions in pro-B cells and its activation by dexamethasone correlates with down-regulation of proinflammatory mediators that are also down-regulated in tumor cells with constitutively active Stat3 (26). Selp (P-selectin) encodes an adhesion receptor expressed on platelets and endothelial cells and plays an important role in tumor metastasis (27, 28). Finally Flt4, the receptor for Vegf-c, is important for tumor angiogenesis and lymphogenous metastasis (29, 30).

The Stat3 targets identified in this work may represent previously unrecognized mediators of Stat3 prooncogenic functions. For the many TFs involved in pathological processes, our method can thus help understanding the molecular mechanisms underlying TF physiological and pathological functions and identifying potential therapeutic targets among the regulated genes.

Materials and Methods

PWM Construction with MEME.

The PWM was derived from the alignment of 54 experimentally validated Stat3-BSs as described in SI Materials and Methods.

Identification of Stat3-BSs.

Stat3-BSs were identified with log likelihood ratios as described in SI Materials and Methods.

EMSA Competition Assays.

EMSA probes and competitors consisted in double stranded DNA oligonucleotides formed by a 9bp long Stat3 binding site flanked by 3bp on both sides and by a GATC protruding sequence on the 5′ end. Labeling and EMSA were performed as described in ref. 31. See SI Materials and Methods for details and oligonucleotide sequences (Table S5).

Comparative Genomics Analysis.

Putative Stat3-BSs above the score cutoff of 9.6 were selected from the mouse reference genome NCBI36M and analyzed as described in SI Materials and Methods.

Cell Lines and Treatments.

Spontaneously immortalized Stat3+/+ and Stat3−/− MEFs (15) were grown in DMEM (Gibco-BRL) supplemented with 10% heat-inactivated FCS (Gibco-BRL), 2 mM l-glutamine, 100 units/mL of penicillin, 100 μg/mL of streptomycin and maintained at 37 °C in a 5% CO2 atmosphere. Cells were treated with Oncostatin M (OSM) (R&D Systems) at a final concentration of 20 ng/mL for 30 min.

Microarray Analysis.

Total RNA was extracted and purified from Stat3+/+ and Stat3−/− MEF cell lines, using the Quiagen RNeasy Mini Kit (Qiagen, Valencia, CA) as suggested by the manufacturer. RNAs were then quantified and inspected with a Bioanalyzer (Agilent Technologies). cRNAs were generated and hybridized on 8 arrays MGU74A v2 Affymetrix DNA chips according to the Affymetrix protocol. The chips were scanned with a specific scanner (Affymetrix) to generate digitized image data files. Data analysis is reported in SI Materials and Methods. The microarray data are available in the GEO database under accession GSE12262.

Chromatin Immunoprecipitation (ChIP) Assay.

Stat3+/+ and Stat3−/− MEFs were treated or not with OSM as described above. ChIP assays were performed with the fast ChIP method (32) with some modifications (see SI Materials and Methods). Immunoprecipitations were performed by incubating overnight at 4 °C 1 mL of sheared chromatin with anti-Stat3 serum (R&D Systems; 5 μL), anti-acetyl-Histone H3 (Upstate Cell Signaling Solutions, 2 μg) or negative control IgG (ChIP-IT control kit-mouse, active motif, 2 μg). Primer sequences used in quantitative Real-Time PCR and semiquantitative PCRs are reported in SI Material and Methods and Tables S6 and S7.

Supplementary Material

Supporting Information:


We thank Professors F. Di Cunto and R. D. Mitra for helpful suggestions and Dr. Ivan Molineris for help in sequence analysis. This work was supported by grants from the Fondo per gli Investimenti della Ricerca di Base and the Italian Cancer Research Association (to V.P.).


The authors declare no conflict of interest.

Data deposition: The data reported in this paper have been deposited in the Gene Expression Omnibus (GEO) database, www.ncbi.nlm.nih.gov/geo (accession no. GSE12262).

This article contains supporting information online at www.pnas.org/cgi/content/full/0900473106/DCSupplemental.


1. Collas P, Dahl JA. Chop it, ChIP it, check it: The current status of chromatin immunoprecipitation. Front Biosci. 2008;13:929–943. [PubMed]
2. Stormo GD. DNA binding sites: Representation and discovery. Bioinformatics. 2000;16:16–23. [PubMed]
3. Tompa M, et al. Assessing computational tools for the discovery of transcription factor binding sites. Nat Biotechnol. 2005;23:137–144. [PubMed]
4. Wasserman WW, Palumbo M, Thompson W, Fickett JW, Lawrence CE. Human-mouse genome comparisons to locate regulatory sites. Nat Gen. 2000;26:225–228. [PubMed]
5. Levy DE, Lee CK. What does Stat3 do? J Clin Invest. 2002;109:1143–1148. [PMC free article] [PubMed]
6. Poli V, Alonzi T. STAT3 function in vivo. In: Sehgal PB, Levy D, Hirano T, editors. Signal Transducers and Activators of Transcription (STATs): Activation and Biology. The Netherlands: Kluwer, Dordrecht; 2003. pp. 493–512.
7. Schindler C, Levy DE, Decker T. JAK-STAT signaling: From interferons to cytokines. J Biol Chem. 2007;282:20059–20063. [PubMed]
8. Bromberg JF, et al. Stat3 as an oncogene. Cell. 1999;98:295–303. [PubMed]
9. Kortylewski M, Jove R, Yu H. Targeting STAT3 affects melanoma on multiple fronts. Cancer Metastasis Rev. 2005;24:315–327. [PubMed]
10. Bailey TL, Williams N, Misleh C, Li WW. MEME: Discovering and analyzing DNA and protein sequence motifs. Nucleic Acids Res. 2006;34:W369–W373. [PMC free article] [PubMed]
11. Horvath CM, Wen Z, Darnell JE., Jr A STAT protein domain that determines DNA sequence recognition suggests a novel DNA-binding domain. Genes Dev. 1995;9:984–994. [PubMed]
12. Caldenhoven E, et al. STAT3beta, a splice variant of transcription factor STAT3, is a dominant negative regulator of transcription. J Biol Chem. 1996;271:13221–13227. [PubMed]
13. Chen X, et al. Integration of external signaling pathways with the core transcriptional network in embryonic stem cells. Cell. 2008;133:1106–1117. [PubMed]
14. Snyder M, Huang XY, Zhang JJ. Identification of novel direct Stat3 target genes for control of growth and differentiation. J Biol Chem. 2008;283:3791–3798. [PubMed]
15. Costa-Pereira AP, et al. Mutational switch of an IL-6 response to an interferon-gamma-like response. Proc Natl Acad Sci USA. 2002;99:8043–8047. [PMC free article] [PubMed]
16. Maritano D, et al. The STAT3 isoforms alpha and beta have unique and specific functions. Nat Immunol. 2004;5:401–409. [PubMed]
17. Yang E, Lerner L, Besser D, Darnell JE., Jr Independent and cooperative activation of chromosomal c-fos promoter by STAT3. J Biol Chem. 2003;278:15794–15799. [PubMed]
18. Elemento O, Tavazoie S. Fast and systematic genome-wide discovery of conserved regulatory elements using a non-alignment based approach. Genome Biol. 2005;6:R18. [PMC free article] [PubMed]
19. Park YG, et al. Sipa1 is a candidate for underlying the metastasis efficiency modifier locus Mtes1. Nat Gen. 2005;37:1055–1062. [PMC free article] [PubMed]
20. Martinez R, et al. Gene structure, promoter activity, and chromosomal location of the DR-nm23 gene, a related member of the nm23 gene family. Cancer Res. 1997;57:1180–1187. [PubMed]
21. Venturelli D, et al. Overexpression of DR-nm23, a protein encoded by a member of the nm23 gene family, inhibits granulocyte differentiation and induces apoptosis in 32Dc13 myeloid cells. Proc Natl Acad Sci USA. 1995;92:7435–7439. [PMC free article] [PubMed]
22. Dunne J, et al. siRNA-mediated AML1/MTG8 depletion affects differentiation and proliferation-associated gene expression in t(8;21)-positive cell lines and primary AML blasts. Oncogene. 2006;25:6067–6078. [PubMed]
23. Tonoli H, Barrett JC. CD82 metastasis suppressor gene: A potential target for new therapeutics? Trends Mol Med. 2005;11:563–570. [PubMed]
24. Loussouarn D, et al. Prognostic impact of syndecan-1 expression in invasive ductal breast carcinomas. Br J Cancer. 2008;98:1993–1998. [PMC free article] [PubMed]
25. Engelmann A, Speidel D, Bornkamm GW, Deppert W, Stocking C. Gadd45 beta is a pro-survival factor associated with stress-resistant tumors. Oncogene. 2008;27:1429–1438. [PubMed]
26. Cowell IG. E4BP4/NFIL3, a PAR-related bZIP factor with many roles. Bioessays. 2002;24:1023–1029. [PubMed]
27. Ding L, et al. In vivo evaluation of the early events associated with liver metastasis of circulating cancer cells. Br J Cancer. 2001;85:431–438. [PMC free article] [PubMed]
28. Garcia J, Callewaert N, Borsig L. P-selectin mediates metastatic progression through binding to sulfatides on tumor cells. Glycobiology. 2007;17:185–196. [PubMed]
29. Grau SJ, et al. Expression of VEGFR3 in glioma endothelium correlates with tumor grade. J Neurooncol. 2007;82:141–150. [PubMed]
30. Lin J, et al. Inhibition of lymphogenous metastasis using adeno-associated virus-mediated gene transfer of a soluble VEGFR-3 decoy receptor. Cancer Res. 2005;65:6901–6909. [PubMed]
31. Alonzi T, et al. Essential role of STAT3 in the control of the acute-phase response as revealed by inducible gene inactivation [correction of activation] in the liver. Mol Cell Biol. 2001;21:1621–1632. [PMC free article] [PubMed]
32. Nelson JD, Denisenko O, Bomsztyk K. Protocol for the fast chromatin immunoprecipitation (ChIP) method. Nat Protocols. 2006;1:179–185. [PubMed]

Articles from Proceedings of the National Academy of Sciences of the United States of America are provided here courtesy of National Academy of Sciences
PubReader format: click here to try


Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...


Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...