Oncogene (2007) 26, 2071–2081 & 2007 Nature Publishing Group All rights reserved 0950-9232/07 $30.00 www.nature.com/onc ORIGINAL ARTICLE STAT1-independent inhibition of cyclooxygenase-2 expression by IFNc;a common pathway of IFNc-mediated repression but not gene activation

L Klampfer1, J Huang1, P Kaler1, T Sasazuki2, S Shirasawa2 and L Augenlicht1

1Department of Oncology, Albert Einstein Cancer Center, Montefiore Medical Center, Bronx, NY, USA and 2Research Institute, International Medical Center of Japan, Toyama, Shinjuku-ku, Tokyo, Japan

Cyclooxygenase-2 (COX-2), the rate-limiting enzyme in Introduction the synthesis of prostaglandins, promotes the development of colorectal cancer, and is a key molecular target of non- such as transforming growth factor b steroidal anti-inflammatory drugs, compounds that reduce (TGFb), interleukin 6 (IL-6), tumor necrosis factor the relative risk of developing colon cancer. In this study, (TNF) and g (IFNg), produced by intrae- we showed that interferon c (IFNc) inhibits the expression pithelial lymphocytes in the small intestine and by of COX-2 in intestinal epithelial cells (IECs) inflammatory cells present in the lamina propria, play an through a pathway that requires Janus-activated kinase important role in the normal process of intestinal (JAK) activity. In contrast, we demonstrated that renewal and in a variety of immune-mediated bowel transcriptional inhibition of COX-2 by IFNb orIFN c disorders. In addition, they are likely to control the occurs in cells with silenced signal transducer and progression from chronic inflammation to the develop- activator of transcription 1 (STAT1) expression and that ment of cancer. IFNs retained the ability to inhibit COX-2 transcription in The JAK/STAT pathway is the principal signaling cells with activated RasV12, in which IFNc failed to pathway utilized by several cytokines, including IFNg,a induce STAT1. Thus, unlike the activity of JAK, STAT1 proinflammatory that has been shown to is not required for the inhibition of COX-2 expression by regulate intestinal renewal and to modulate epithelial IFNc. In contrast to COX-2, the activation of in barrier function (Guy-Grand et al., 1998; Bruewer et al., response to IFNc, such as interferon regulatory factor-1, 2003). Upon binding to its cell-surface receptor, IFNg was severely impaired by both STAT1 silencing and by triggers receptor oligomerization and activation of two constitutive Ras signaling. To determine whether there is a receptor-associated kinases, JAK1 and JAK2, which in general differential requirement for STAT1 in gene turn phosphorylate the latent cytoplasmic transcription activation and gene repression in response to IFNc in factor, STAT1, on a conserved tyrosine residue (re- intestinal cells, we performed genome-wide analysis of viewed by Levy and Darnell, 2002). Phosphorylation of IFNc target genes in an IEC line in which STAT1 STAT1 results in its dimerization and the subsequent expression was silenced by small interfering RNA. The translocation of STAT1 to the nucleus, where it binds to results confirmed that the activation of the majority of a specific DNA recognition sequence (Schindler et al., genes by IFNc required STAT1. In contrast, the 1992; Shuai et al., 1992; Vinkemeier et al., 1996). STAT1 repression of several genes, as we showed for COX-2 deficiency significantly alters the biological activity of specifically, was largely unaffected in cells with silenced IFNg, attesting to its prominent role in signaling by STAT1. Our results therefore demonstrate that in general IFNg. STAT1 null mice display enhanced susceptibility gene activation by IFNc is more sensitive to STAT1 to viral infections (Durbin et al., 1996; Meraz et al., deficiency than gene repression, and suggest that IFNc 1996) and have a tumor-prone phenotype (Kaplan et al., activates and represses gene expression via distinct 1998; Shankaran et al., 2001), underscoring the sig- pathways that can be distinguished, at least in part, by nificance of the JAK/STAT pathway in vivo. their requirement for STAT1. However, recent reports have established that IFNg Oncogene (2007) 26, 2071–2081. doi:10.1038/sj.onc.1210015; can modulate the expression of several of its target genes published online 2 October 2006 in the absence of STAT1 (Ramana et al., 2000, 2001, 2002; Gil et al., 2001). For example, microarray analysis Keywords: STAT1; COX-2; JAK; IFN; colon cancer revealed that a comparable number of genes are regulated both in wild-type and STAT1-deficient bone marrow-derived (Gil et al., 2001), con- firming that regulation of a subset of IFN target genes Correspondence: Dr L Klampfer, Department of Oncology, Albert does not require STAT1. The nature of signaling by Einstein Cancer Center, Montefiore Medical Center, 111 E 210th IFNg that bypasses the JAK/STAT pathway is not yet Street, Bronx, NY 10467, USA. E-mail: [email protected] well understood, and signaling pathways that may play Received 23 May 2006; revised 27 July 2006; accepted 18 August 2006; a role include candidates such as extracellular signal- published online 2 October 2006 regulated kinase (ERK1/ERK2), phosphatidylinositol IFNs inhibit COX-2 expression L Klampfer et al 2072 3-kinase (PI3K), nuclear factor-kB, AKT, and poten- oncogenic activation of Ras (Sheng et al., 2001). In tially other members of the STAT family (Deb et al., addition, inflammatory cytokines are known to mod- 2001; Hu et al., 2001; Ramsauer et al., 2002; Navarro ulate the expression of COX-2 (Gupta and Dubois, et al., 2003; Qing and Stark, 2004). 2001). IFNg has been shown to upregulate the expres- It has become evident that chronic inflammation is an sion of COX-2 in murine macrophages (Vila-del Sol and important contributing factor in tumorigenesis, and Fresno, 2005); however, its effects on COX-2 expression tumors have been considered to be ‘wounds that never in IECs have not been addressed. We therefore heal’ (Dvorak, 1986). For example, chronic intestinal examined whether IFNg regulates COX-2 expression inflammation has been shown to predispose to the in IECs. IEC-6 cells, which express COX-2 constitu- development of colon cancer (Biasco et al., 2002). In tively, were stimulated with 10ng/ml of IFN g for 24, 48 addition, the majority of precursor lesions in sporadic or 72 h. Treatment of IEC-6 cells with IFNg induced the colorectal cancers are also frequently inflammatory in expression of STAT1, a major nature (Higaki et al., 1999). Cytokines secreted by in IFN signaling, and interferon regulatory factor 1 inflammatory cells can directly promote cancer cell (IRF-1), a downstream target of STAT1 (Figure 1a). In proliferation, cell survival and angiogenesis and thereby contrast, IFNg in a time-dependent manner strongly regulate tumor progression. inhibited the expression of COX-2. To determine One of the key links between inflammation and whether JAKs, kinases participating in signaling by cellular transformation appears to be activation of IFNg, are essential for the inhibition of COX-2 cyclooxygenase-2 (COX-2), an inducible gene with a expression by IFNg, we pretreated cells with a JAK role in both inflammation and tumorigenesis. Its inhibitor 1 (Ji) (Calbiochem, La Jolla, CA, USA) (5 mM) expression is elevated in a variety of human malignan- for 2 h before treatment with IFNg. As shown in cies and their precursor lesions (Eberhart et al., 1994), Figure 1b, inhibition of JAK activity completely establishing COX-2 as an important target for chemo- prevented the inhibition of COX-2 expression by IFNg. preventive agents. Indeed, anti-inflammatory agents Likewise, the induction of IRF-1 and STAT1 was such as non-steroidal anti-inflammatory drugs, which inhibited in cells treated with Ji (data not shown). These target the activity of COX-2, significantly reduce the risk data therefore demonstrated that the activity of JAKs is of colorectal cancer (Brown and DuBois, 2005). required for the inhibition of COX-2 expression by Among known stimuli that induce COX-2 expression IFNg in IECs. in intestinal epithelial cells (IECs) are activation of Ras To determine whether STAT1, a transcription factor (Sheng et al., 2001) or inactivation of the tumor downstream of JAK, mediates inhibition of COX-2 by suppressor gene adenomatous polyposis coli (Apc) IFNs, we examined the effect of IFNg on COX-2 (Araki et al., 2003). Here, we demonstrate that IFNg expression in IEC-6 cells that were transfected with is a potent inhibitor of COX-2 expression in IECs, and STAT1 siRNA. We achieved good silencing of both present data which show that IFNs modulate COX-2 transcription through a pathway that requires the a -244872 activity of JAKs, but bypasses STAT1. Furthermore, (hours) global analysis of IFN-responsive genes in IECs in STAT1 which STAT1 was specifically silenced by small inter- fering RNA (siRNA) confirmed that gene activation in response to IFNg generally requires STAT1, but that IRF-1 gene repression often involves pathways that circumvent STAT1 entirely, or include both STAT1-dependent and STAT1-independent signaling. Our results therefore COX-2 suggest that IFNg activates and represses gene expres- sion via distinct pathways that can be separated, at least in part, by their requirement for STAT1. Finally, we showed that Ras transformation of epithelial cells alters βactin the balance between STAT1-dependent and STAT1- independent signaling owing to impaired expression of STAT1 in Ras-transformed cells. This is likely to have important consequences for the local inflammatory b 0 IFN LPS Ji IFN/Ji LPS/Ji response and, therefore, the progression of intestinal COX-2 tumors.

βactin Results Figure 1 IFNg inhibits the expression of COX-2 in a JAK- IFNg inhibits COX-2 expression through a pathway that dependent manner. (a) IEC-6 cells were treated with IFNg (10ng/ ml) for 24, 48 or 72 h. The levels of STAT1, IRF-1, COX-2 and b- requires the activity of JAKs, but not STAT1 actin were determined by immunoblotting. (b) IEC-6 cells were COX-2 expression is frequently elevated in colon cancer treated with IFNg or LPS alone, or 2 h after pre-treatment with as a result of Apc mutations (Araki et al., 2003) or 5 mM of Ji (Calbiochem).

Oncogene IFNs inhibit COX-2 expression L Klampfer et al 2073 basal and inducible STAT1 expression (Figure 2a). et al., 2001), we showed that signaling by oncogenic Ras Consistently, the induction of IRF-1, a STAT1- leads to the activation of COX-2 expression. However, dependent transcription factor (Klampfer et al., 2004), despite the impaired STAT1 expression in cells with was significantly inhibited in cells with silenced STAT1 oncogenic Ras, IFNg was able to repress COX-2 expression. In contrast, the results of several experi- expression efficiently in the presence of IPTG ments revealed that IFNg retained most of its ability to (Figure 3b). In contrast, induction of IRF-1 by IFNg inhibit the expression of COX-2 in STAT1-deficient cells was inhibited by the activation of mutant Ras (data not (Figure 2a). Densitometric analysis of three independent shown), consistent with the requirement of STAT1 for experiments, shown in Figure 2b, confirmed that IFNg the expression of IRF-1 (Figure 2a). Altogether, these downregulated the expression of COX-2, and estab- data confirmed that STAT1 is not a major transcription lished that the extent of COX-2 inhibition by IFNg was factor whereby IFNg inhibits the expression of COX-2. not affected by STAT1 deficiency. Altogether, these To determine whether IFNg inhibits COX-2 expres- data strongly suggested that STAT1 is not essential for sion at the level of transcription, we used IEC-ikRas the inhibition of COX-2 expression by IFNg.Aswe cells stably transfected with a COX-2 promoter reporter demonstrated that the expression of IRF-1, a transcrip- (containing the 50-flanking region from þ 59 to À1432) tion factor that has been shown to be required for the (Sheng et al., 2001). IEC-ikRas-COX-2 50 cells were activation of COX-2 expression by IFNg in macro- either left untreated or were pretreated with IPTG for phages (Blanco et al., 2000), is inhibited in STAT1- 24 h and were stimulated with IFNg (1 or 5 ng/ml) or deficient cells (Figure 2a), these data also demonstrated IFNb (100 U/ml) for 24 h. Induction of mutant Ras with that IRF-1 does not mediate the inhibition of COX-2 by IPTG, as described before (Sheng et al., 2001), modestly IFNg in IECs. enhanced COX-2 promoter activity (Figure 3c). We We previously reported that the expression of STAT1 demonstrated that treatment of cells with IFNg or IFNb is reduced in cells that harbor mutant Ras (Klampfer inhibits COX-2 transcriptional activity in the absence or et al., 2003a). Therefore, we examined whether IFNg the presence of IPTG (Figure 3c). These data established preserves the ability to inhibit COX-2 expression upon that IFNs inhibit COX-2 expression at least in part at induction of mutant Ras. To answer this question, we the level of transcription and confirmed that STAT1 used IECs with isopropyl-1-thio-b-D-galactopyranoside does not play a major role in transcriptional inhibition (IPTG)-inducible expression of oncogenic RasV12, IEC- of COX-2 by IFNs. In contrast, inhibition of JAK ikRas cells (Sheng et al., 2001). Activated Ras is induced activity completely prevented the ability of IFNg to to high levels in these cells upon addition of 5 mM IPTG inhibit COX-2 transcriptional activity, both in the (Figure 3a), and we showed that upon induction of absence and in the presence of IPTG (Figure 3d). Thus, oncogenic Ras by IPTG, both basal and IFNg-inducible IFNs inhibit COX-2 expression at a transcriptional level expressions of STAT1 were significantly reduced in a JAK-dependent, STAT1-independent manner. (Figure 3b). Consistent with published data (Sheng Silencing of STAT1 expression in Hke-3 cells Our experiments with COX-2 suggested that the path- a NSP STAT1 (siRNA) ways whereby IFNg inhibits and induces gene expres- sion may be distinct. To determine whether there is 0 24 48 0 24 48 indeed a general differential requirement for STAT1 in STAT1 gene activation and gene repression in response to IFNg, we performed genome-wide analysis of IFNg-responsive

IRF-1 genes in intestinal cells with silenced STAT1 expression. Although large-scale analysis of IFN target genes has been performed in fibroblasts and mononuclear phago- COX-2 cytes isolated from STAT1-deficient mice (Gil et al., et al β 2001; Ramana ., 2001), the responsiveness of actin epithelial cells to IFNg has not been analysed. We performed this experiment in the Hke-3 colon cancer cell b 0 1.00 24 line, because we showed before that these cells are, 48 owing to genetic inactivation of the mutant Ras allele 0.75 actin

β (Shirasawa et al., 1993), highly responsive to IFNg 0.50 (Klampfer et al., 2003b). Cells were transfected with a

COX-2/ 0.25 pool of four different siRNAs directed against the coding region of the STAT1 gene (Dharmacon, Lafay- 0.00 NSP STAT1 NSP STAT1 ette, CO, USA) or with a non-targeting, nonspecific 10 ng/ml30 ng/ml (IFNγ) (NSP) siRNA as we recently described (Klampfer et al., Figure 2 Inhibition of COX-2 expression by IFNg does not 2004). It is important to emphasize that we transfected require STAT1. (a) Rat IEC cells were transfected with either NSP cells with very low concentrations (25 nM) of siRNA to siRNA or siRNA that specifically targets STAT1. Cells were treated with IFNg (10ng/ml) for 24 or 48 h and the levels of STAT, avoid potential NSP effects of IFN production triggered COX-2, IRF-1 and b-actin were determined by immunoblotting. by transfection. Twenty-four hours after transfection, (b) Densitometric analysis of three independent experiments. cells were split into three flasks to minimize differences

Oncogene IFNs inhibit COX-2 expression L Klampfer et al 2074 a IPTG - + b IPTG - - - + + + IFNγ -15-15

Ras STAT1

βactin COX2

c 50 -IPTG +IPTG 40

30

20 LUC activity 10

0

γ 1 γ 5 γ 1 γ 5 β 100 β 100 CTRL IFN IFN CTRL IFN IFN IFN IFN

d 100 -IPTG +IPTG 75

50

LUC activity 25

0 γ β γ β IFN IFN CTRL IFN CTRL IFN γ/JAKinhβ/JAKinh γ/JAKinhβ/JAKinh IFN IFN IFN IFN Figure 3 IFN inhibits elevated expression of COX-2 in cells that harbor mutant Ras. (a) The inducible expression of Ras upon treatment of cells with IPTG (5 mM, 24 h). (b) IEC cells with inducible Ras (Sheng et al., 2001) were treated with IFNg in the absence or the presence of IPTG, as indicated. The levels of STAT1 and COX-2 were determined by immunoblotting. (c and d) IEC-ikRas-COX-2 50 cells were treated with IFNg or IFNb in the absence or the presence of IPTG and Ji (5 mM), as indicated and the LUC activity was determined 24 h after treatment.

in transfection efficiency among samples. Cells were NSP IFN response and trigger the induction of IFN- serum starved overnight, and were then either left responsive genes. untreated or were treated with IFNg (10ng/ml) for 1 An aliquot of transfected cells was maintained for 24 or 3 h. IFNg-responsive genes that we identified by our or 30h after treatment with IFN to determine the effect analysis are therefore likely to represent primary target of silencing also at the protein level by immunoblotting. genes of IFNg. Microarray experiments were performed As shown in Figure 4b, both the basal (‘0’) and inducible using the Affymetrix (Affymetrix, Santa Clara, CA, levels of STAT1 protein were significantly reduced in USA) U133 plus 2.0 gene chip, encompassing 47 000 cells transfected with STAT1 siRNA. The inducible sequences, as described in Materials and methods. expression of STAT1 in cells transfected with STAT1 Analysis of the microarray data confirmed that both siRNA was consistently lower than the basal expression the basal (0) and the inducible expression of STAT1 of STAT1 in cells transfected with control, NSP siRNA were significantly reduced in cells transfected with (Figure 4a and b). The basal expression of STAT3 and STAT1 siRNA (Figure 4a). The basal level of STAT1 STAT5, two STATs that are also known to be involved in non-transfected cells and in cells transfected with NSP in IFN signaling, was not affected by STAT1 silencing siRNA were comparable (not shown), demonstrating (data not shown and Supplementary Figure 1), demon- that transfection of cells with siRNA did not induce a strating that neither STAT3 nor STAT5 compensated

Oncogene IFNs inhibit COX-2 expression L Klampfer et al 2075 for STAT1 deficiency, and establishing specificity of treatment with IFNg (Figure 5a and b). Approximately STAT1 silencing. 40% of those genes were induced twofold within 1 h of treatment (Figure 5a). Among the genes that were Activation of genes in response to IFNg requires STAT1 Genome-wide expression analysis in cells with silenced STAT1 expression revealed that silencing of STAT1 affected the basal expression of a number of genes, as well as the inducible expression in response to IFNg. The basal expression of 257 genes was reduced by 50%, and the expression of 439 genes was enhanced twofold in cells with silenced STAT1 expression. Among genes that required STAT1 for their basal expression were keratin- associated protein 3-1, IFN-induced transmembrane protein 1, stromal antigen 2, Hsp 70and sialyltransfer- ase 7 (see Supplementary Figure 2a). Among genes whose basal expression was elevated in the absence of STAT1 were metallothionein 1, MALAT1, H2aa, H2bd and prothymosin a (see Supplementary Figure 2b). These genes play a significant role in tumorigenesis, and regulation of their expression by STAT1 warrants further investigation. However, in this study we focused on the role of STAT1 in IFNg- inducible gene expression. The inducibility by IFNg was expressed as the ratio of the signals between IFNg-treated cells and control, untreated cells. In cells transfected with nonspecific (NSP), siRNA 236 genes were induced twofold 3 h after

a NSP siRNA NSP STAT1 (siRNA) STAT1 siRNA 3000 0 1 3 0 1 3 (hours) STAT1 STAT1 STAT1 2000 (STAT1) 1000 Intensity of the signal

0 0 1 3 IFNγ (hours)

b NSPsiRNA STAT1siRNA 0 24 30 0 24 30 (IFNγ /hours)

STAT1

NSB

Figure 4 Silencing of STAT1 in a colon cancer cell line. (a and b) Expression of STAT1 mRNA in Hke-3 cells transfected with NSP Figure 5 Induction of genes in response to IFNg depends on the or STAT1 siRNA (the results shown represent three different spots presence of STAT1. (a and b) Cells were transfected with NSP on the chip). Cells were either left untreated (0) or were treated with siRNA or STAT1 siRNA and were either left untreated or were IFNg for 1 or 3 h as indicated. (c) Expression of STAT1 protein in treated with IFNg for 1 or 3 h as indicated. The results are cells transfected with NSP siRNA or STAT1 siRNA as determined expressed as the ratio of gene expression between induced and 24 or 30h after treatment with IFN g. NSB: nonspecific band. control cells.

Oncogene IFNs inhibit COX-2 expression L Klampfer et al 2076 induced in control cells at the highest level by IFNg at of IFNg-inducible genes in IECs requires STAT1 1 h were STAT1, STAT3, IRF-1, nuclear antigen 100 (Figure 5a and b). Among several other known IFNg and apolipoprotein L. Three hours after treatment with targets, we showed that a cluster of ISGs, MyD88, IFNg, the expression of programmed cell death 1 ligand phospholipid scramblase1 and Fas all require STAT1 1, apolipoprotein L, GBP1, GBP2, GBP3, STAT1, for their induced expression by IFNg. Moreover, a NOD 27, TAP1, IL-15 and IL-15 receptor reached the subset of genes that were upregulated by IFN in cells highest induction (Figure 5b). transfected with NSP siRNA were actually down- In cells in which STAT1 expression was inhibited by regulated by IFNg in STAT1-deficient cells, demonstrat- transfection with STAT1 siRNA, only 23 of the 236 ing that the absence of STAT1 resulted in perturbed genes remained induced twofold 3 h after treatment. response of cells to IFNs (see Figure 5a and b). Among those, 23 genes were GBP2, GBP3, IFI44, We validated the findings from microarray analysis PDCD1LG and IFRG28. Therefore, silencing of for four induced genes, IRF-9, b2M, KLF4 and IFIT1, STAT1 reduced the number of induced genes by 90%. by quantitative reverse transcriptase–polymerase chain However, this is a minimal estimate, as the analysis of reaction (RT–PCR). As shown in Figure 6a, we the microarray data showed that the basal expression of confirmed that the inducibility of all four genes was the majority of these 23 genes was also significantly significantly affected by STAT1 deficiency. This experi- reduced in STAT1-deficient cells. This low basal ment also demonstrated the excellent correlation expression following STAT1 silencing contributed to between data obtained by microarray analysis and those high (and probably exaggerated) calculated inducibility obtained by quantitative RT–PCR (compare Figure 6a of these 23 genes. In reality, the expression of all and b). these genes was severely affected by STAT1 silencing. In the group of 236 genes, we found only a few genes whose Repression of genes by IFNg occurs through both STAT1- expression was truly STAT1 independent, such as solute dependent and STAT1-independent pathways carrier family 30(zinc transporter), cytochrome P450 The level of gene repression in response to IFNg was and metallothionein 1X (see Supplementary Figure 1). expressed as the ratio of the signals between IFNg- Although several other genes remained partially treated cells and control, untreated cells. Therefore, only responsive to IFNg in the absence of STAT1, these genes whose intensity of expression reached a threshold results demonstrated that the activation of the majority level of expression in unstimulated cells were analysed

a Quantitative RT-PCR data

4 0 1h 3 3h

2

FOLD INDUCTION 1

0 IRF-9 β2M KLF-4 IFIT1 IRF-9 β2M KLF-4 IFIT1

NSP siRNA STAT1 siRNA b Microarray data

3 0 1h 3h 2

1 FOLD INDUCTION

0 IRF-9 β2M KLF-4 IFIT1 IRF-9 β2M KLF-4 IFIT1

NSP siRNA STAT1 siRNA Figure 6 Expression of IRF-9, b2M, KLF4, IFIT1 in cells transfected with NSP siRNA or STAT1 siRNA, as determined by quantitative RT–PCR (a), or by analysis of microarray data (b). Expression of IRF-9, b2M and IFIT1 was determined twice by RT– PCR and the mean of two experiments each done in triplicates, is shown in (a).

Oncogene IFNs inhibit COX-2 expression L Klampfer et al 2077 further. Here, we will present data on genes whose twofold in Hke-3 cells 3 h after treatment with IFNg, expression reached 50( T ¼ 50) in untreated cells. demonstrating that the number of genes that were Among those, 120genes were repressed by IFN g repressed by IFNg was lower than the number of induced genes. Twenty-five percent of genes that were reduced twofold in their expression in cells transfected with NSP siRNA were reduced to the same extent by IFN also in cells in which STAT1 had been silenced through transfection with STAT1 siRNA. In addition, the majority of the remaining genes were also reduced in expression in the absence of STAT1, although to a somewhat smaller extent, or with a delay (Figure 7a). Thus, unlike the activation of gene expression, gene repression in response to IFNg appears to be less dependent on the presence of STAT1 and is more likely to involve STAT1-independent pathways. We found that a cluster of sequences that encode ribosomal (Figure 7b) as well as several eucaryotic translation initiation and elongation factors (Supple- mentary Figure 1, not shown) were downregulated upon IFNg treatment; in most cases, the inhibition of ribosomal proteins, as in the case of COX-2 (Figure 2), occurred in a STAT1-independent manner (Figure 7b). Among genes downregulated by IFN treatment were ID1, ID2 and ID3, a family of proteins that has been shown to inhibit differentiation and to stimulate proliferation (Perk et al., 2005) and SOX-9, a transcrip- tion factor important for differentiation of IECs (Blache et al., 2004). A fourth member of the family, ID4, was poorly expressed in Hke-3 cells and its expression was not regulated by IFNg. The inhibition of ID1, ID2 and ID3 by IFNg may contribute to the prodifferentiation and antiproliferative activities of IFNg. We confirmed by quantitative RT–PCR that inhibi- tion of ID1, ID3 and SOX-9 by IFNg does not depend on the presence of STAT1, as it occurred both in cells transfected with NSP and STAT1 siRNA (Figure 8a). In contrast, inhibition of ID2 expression appears to depend partially on STAT1, because it was diminished in cells with deficient STAT1 expression. These RT–PCR data again correlated well with results obtained by micro- array analysis (Figure 8b). We confirmed the quantita- tive RT–PCR experiment using RNA isolated from an independent experiment (data not shown) and have obtained similar results.

Discussion

IECs are in close contact with T cells from within the epithelial compartment (intraepithelial lymphocytes) as well as with lymphocytes from the underlying lymphoid tissue (lamina propria lymphocytes). These cells are a rich source of IFNg, a cytokine that, directly or through

Figure 7 Repression of genes by IFNg frequently occurs in a STAT1-independent manner. (a) Expression of 120genes, whose expression was reduced twofold by IFNg.(b) Repression of ribosomal proteins by IFNg. Results are expressed as the ratio between induced and uninduced cells in cells transfected with nonspecific (NSP) SiRNA or with STAT1 SiRNA.

Oncogene IFNs inhibit COX-2 expression L Klampfer et al 2078 a Quantitative RT-PCR data

1.2 0

1.0 1h 3h 0.8

0.6

0.4 FOLD INCREASE 0.2

0.0 ID1 ID2 ID3 SOX9 ID1 ID2 ID3 SOX9 NSP siRNA STAT1 siRNA

b Microarray data 1.25 0 1h 1.00 3h 0.75

0.50 FOLD INCREASE 0.25

0.00 ID1 ID2 ID3 SOX9 ID1 ID2 ID3 SOX9 NSP siRNA STAT1 siRNA Figure 8 Expression of ID1, ID2, ID3 and SOX9 as determined by quantitative RT–PCR (a), or obtained by analysis of microarray data (b). Expression of ID1, ID2 and SOX-9 was determined in two independent experiments by RT–PCR and the mean of two experiments each done in triplicates, is shown in (a).

stimulation of other growth factors, plays a significant is likely to be blunted. We showed that in cells harboring role in intestinal epithelial renewal (Guy-Grand et al., mutant Ras, IFNg failed to induce the expression of 1998; Bruewer et al., 2003). Here, we present data which genes such as IRF-1, a transcription factor required for demonstrate that IFNg inhibits the expression of COX- IFNg-mediated induction of COX-2 in macrophages 2, an important promoter of intestinal tumorigenesis (Blanco et al., 2000). In contrast, IFNg retained the and a target of pharmacological chemopreventive ability to downregulate the expression of COX-2 in cells agents. Although IFNg activates COX-2 expression in with activated Ras, despite the impaired expression of macrophages (Blanco et al., 2000) and epidermal both STAT1 and IRF-1 in these cells. These results keratinocytes (Matsuura et al., 1999), it has been shown indicated a potential divergence in IFNg-mediated gene to downregulate COX-2 in glioma cells (Janabi et al., activation and gene repression. 2004), in the placenta (Hanna et al., 2004) and to inhibit To evaluate the significance of STAT1-independent TNF-induced COX-2 expression in colonic epithelial signaling by IFNg in intestinal cells further, we cells (Wright et al., 2004). In this report, we established performed genome-wide analysis of IFN-responsive that in IECs both IFNg and IFNb inhibit basal COX-2 genes in cells transfected with a non-targeting siRNA expression at the transcriptional level in a JAK- or STAT1-specific siRNA. dependent, but STAT1-independent pathway. Inhibi- Analysis of microarray data revealed a cluster of tion of COX-2 expression by IFNs is likely to contribute genes that had basal expression significantly upregulated to the strong antiangiogenic activity of IFNs (Lindner, in cells with silenced STAT1 expression. Among those 2002). genes were metallothionein 1, glutathione S-transferase STAT1 deficiency has been shown to affect signifi- pi, dihydrofolate reductase, prothymosina (proteins that cantly the biological activity of IFNg in vitro (Ramana play a role in response of cells to stress) and MALAT-1, et al., 2000); however, the biological circumstances a novel non-coding RNA, whose expression has been where STAT1-independent signaling may prevail re- shown to predict metastasis and survival in early-stage mained elusive. We demonstrated that in IECs signaling non-small-cell lung cancer (Ji et al., 2003; Muller-Tidow by oncogenic Ras interferes with STAT1 function et al., 2004). Because we demonstrated that cells (Figure 3b; Klampfer et al., 2003a), thus identifying a harboring mutant Ras have reduced expression of physiological state in which STAT1-dependent signaling STAT1 (Klampfer et al., 2003a), we predicted that the

Oncogene IFNs inhibit COX-2 expression L Klampfer et al 2079 expression of these genes should be increased in cells data clearly demonstrate that activation of the majority that harbor mutant Ras. Indeed, through analysis of of IFNg target genes requires STAT1, whereas gene genome-wide expression of isogenic cell lines that differ repression occurs through both STAT1-dependent and only by the presence of the mutant Ras, we showed that STAT1-independent pathways, or bypasses STAT1 the expression of MT1 and prothymosin a is signifi- altogether. The identity of pathways whereby IFNg cantly increased in HCT116 cells (which contain mutant inhibits the expression of genes, such as COX-2, in the Ras), compared to two isogenic clones with targeted absence of STAT1 and the biological significance of deletion of the mutant Ras, Hke-3 and Hkh2 cells STAT1-independent signaling remain, for now, un- (Klampfer et al., unpublished). Therefore, reduced known. Experiments to address these questions are expression of STAT1 in HCT116 cells is likely to underway in the laboratory. Our preliminary data have contribute to increased expression of MT1 and prothy- excluded the involvement of ERK and PI3K signaling in mosin a in these cells. These data also demonstrate that the inhibition of COX-2 expression by IFNg (not STAT1 is a downstream target of signaling by oncogenic shown). Ras, which is likely to contribute to transformation of Our findings have important implications for the epithelial cells and to modulate their responsiveness to understanding of perturbed biological activity of IFNg therapy. in cancer cells, where STAT1 function is often impaired Our data revealed that induction of genes by IFNg, (Wong et al., 1997; Abril et al., 1998; Landolfo et al., with rare exceptions, requires STAT1. In contrast, we 2000; Clifford et al., 2002), such as in cells harboring demonstrated that IFNg retains the ability to inhibit the Ras mutations (Klampfer et al., 2003a). Our data expression of several of its target genes in cells with predict that gene induction, but not gene repression in silenced STAT1 expression, demonstrating that gene response to IFNg would be affected in cells that harbor activation by IFNg is more sensitive to STAT1 mutant Ras. Indeed, we demonstrated that induction of deficiency than gene repression, and suggesting that oncogenic Ras prevented the induction of STAT1 and the pathways of gene activation and gene repression by IRF-1 in response to IFNg, but did not interfere with IFNg diverge downstream of JAK. the ability of IFNs to downregulate COX-2 expression. Among genes that were found to be downregulated by Therefore, although Ras mutations and consequent IFNg in a STAT1-independent manner were ID1, ID2, reduction of STAT1 expression do not eliminate the ID3 and SOX9, genes that play an important role in responsiveness of cells to IFNg, they are likely to alter maintaining homeostasis in intestinal epithelium. This the biological response to IFN significantly. Consistent finding raises the interesting possibility that, like another with our data, it has been recently shown that the anti- antiproliferative factor, TGFb (Kowanetz et al., 2004), viral activity of IFN is abrogated in cells that carry IFNs inhibit proliferation and promote differentiation oncogenic Ras mutation (Battcock et al., 2006). of epithelial cells through downregulation of the ID family members. Our ongoing experiments using ID1 and ID2 promoters indicate that IFNg inhibits ID1 and Materials and methods ID2 expression at the transcriptional level (Klampfer et al., unpublished). We also found that a cluster of Cells and transfections several ribosomal proteins and eucaryotic translation Hke-3 cells were derived from HCT116 cells through targeted initiation and elongation factors were downregulated deletion of the activated Ras allele (Shirasawa et al., 1993) and, upon IFNg treatment in a STAT1-independent manner. as we have reported (Klampfer et al., 2003a), respond to IFNs Inhibition of these genes is likely to contribute to the with STAT1 activation. Cells were transfected with a pool of antiproliferative and anti-viral activities of IFNs. We four siRNAs (Dharmacon) specific for the human STAT1 gene using the Profection Mammalian Transfection Systems (Pro- recently demonstrated that in vivo the expression of mega, Madison, WA, USA) as we described recently these genes declines as the IECs exit the proliferative (Klampfer et al., 2004). NSP, non-targeting siRNA, was used compartment along the crypt villus axis (Mariadason as control. Both STAT1 and NSP siRNAs were delivered at a et al., 2005). concentration of 25 nM. Twenty-four hours after transfection, Although our results suggest that, generally, IFNg- cells were trypsinized and divided into three T25 flasks to mediated gene activation, but not gene repression minimize differences in transfection efficiency among samples. requires STAT1, we certainly found exceptions. For After reaching 80% confluency, cells were serum starved example, MT1, keratin-associated protein 3-1, IFN- overnight and were either left untreated or were treated with induced transmembrane protein 1, stromal antigen 2, 10ng/ml of recombinant human IFN g (Biosource Interna- Hsp 70and sialyltransferase 7 were all fully or partially tional, Camarillo, CA, USA) for 1 or 3 h. Rat IECs with an inducible k-RasV12 (IEC-ikRas) and IEC-ikRas cells trans- induced by IFNg in cells transfected with STAT1 siRNA fected with the COX-2 promoter region (IEC-ikRas/COX-2 50- (see Supplementary Figure 1). Likewise, genes such as LUC) were a generous gift from Raymond Dubois and were perlecan, a heparin sulfate proteoglycan, were inhibited transfected with siRNA specific for rat STAT1 (Dharmacon) by IFNg in cells transfected with NSP siRNA, but failed using Lipofectamine 2000 (Invitrogen, Carlsbad, CA, USA). to be inhibited by IFNg in cells transfected with STAT1 siRNA (see Supplementary Figure 1). This is consistent RNA isolation with the report that STAT1-deficient fibrosarcoma cells RNA was isolated using TRIZOL Reagent (GIBCO BRL, do not respond to IFNg with inhibition of perlecan Rockville, MD, USA) as suggested by the manufacturer and expression (Sharma and Iozzo, 1998). However, our was further purified using the RNeasy Midi Kit (Qiagen,

Oncogene IFNs inhibit COX-2 expression L Klampfer et al 2080 Valencia, CA, USA). The RNA was then precipitated at b2M F: 50-TGTGCTCGCGCTACTCTCTCT-30, b2M R: 50- À201C overnight by adding 2.5 V of ethanol and 0.1 V of 0.3 M GTCAACTTCAATGTCGGATGGA-30; IRF-9 F: 50-CCGT sodium-acetate (pH 5.3). The integrity and size distribution of GATAATCGTGTCCTGAAA-30, IRF-9 R: 50-CCTGGGTT total RNA was monitored by denaturing agarose gel electro- CACACCATTTGG-30; PKR F: 50-ATGATGGAAAGCG phoresis with detection by ethidium bromide staining. AACAAGGA-30, PKR R: 50-GTTTCAAAAGCAGTGTCA CATACATG-30; KLF4 F: 50-TGCTGATTGTCTATTTTTG Microarray experiments CGTTTA-30, KLF4 R-50-GAGAAGAAACGAAGCCAAAA We used U133 plus 2.0 gene chips produced by CC-30; ID1 F: 50-CCCTTAACTGCATCCAGC-30, ID1 R: 50- Affymetrix for all experiments, which contain 47 000 gene TCAGAAATCTGAGAAGCACC; ID2 F: 50-CATGAAA sequences. Among the sequences on the chip, 27 160tran- GCCTTCAGTCC-30, ID2 R:50-CAAGATGTAGTCGATG scripts were expressed above background (‘P’, as determined ACG-30; ID3 F-50-ACTTGACTTCACCAAATCC-30, ID3 by Affymetrix) in at least one of the samples; only these were R-50-GCAACAGAACCTTTCTCC-30; SOX9 F: 50-CTACGA analysed further. CTGGACGCTGGTG-30, SOX9 R:50-GTAATCCGGGTGG Labeled cRNA was prepared from 5 mg of RNA using The TCCTTCT-30 and GAPDH F: 50-GCACCGTCAAGGCTG One Cycle Target Labeling Assay kit (Affymetrix) as suggested AGAAC-30; GAPDH R-50-GCCTTCCATGGTGGTGAA-30. by the manufacturer. Hybridization and scanning were carried Experiments were carried out in triplicate and the expression out by the Albert Einstein Cancer Center DNA facility. The of each gene was standardized using GAPDH as a reference. results were expressed as the ratio of signals between IFNg- Amplification reactions were analysed using a 7900HT real- treated and control, untreated cells. time PCR instrument (Applied Biosystems, Foster City, CA, USA). The results were expressed as the ratio between the Quantitative RT–PCR analysis expression of the IFNg target genes in treated and untreated Expression levels of selected genes that were shown to be cells. regulated by IFNg by microarray analysis were confirmed by quantitative real-time RT–PCR analysis. RNA was isolated as Acknowledgements described above and 1 mg of RNA was reverse transcribed using TaqMan Reverse Transcription Reagents (Applied We are indebted to Raymond DuBois for his generous gift of Biosystems, Foster City, CA, USA). cDNA was amplified IEC-ikRas and IEC-ikRas-COX-2 5’ cells. We are grateful to using specific primers for glyceraldehyde-3-phosphate dehy- John Mariadason for his advice in analysing the microarray drogenase (GAPDH), IRF-9, b2M and IFN/TETRA1, KLF- data and suggestions regarding the manuscript, and to Laura 4, ID1, ID2, ID3 and SOX9. Primer sequences were as follows Bancroft for technical assistance. This study was supported by IFN/TETRA1 F: 50-TTCCACTATGGTCGGTTTCA 30, RO1 CA111361 (to LK), U54 CA100926 (to LA) and P30- IFN/TETRA1 R: 50-GAGGCTCAAGCTTTCCAGAT-30; 13330from the NCI.

References

Abril E, Real LM, Serrano A, Jimenez P, Garcia A, Canton J Clifford JL, Walch E, Yang X, Xu X, Alberts DS, Clayman et al. (1998). Unresponsiveness to interferon associated with GL et al. (2002). Suppression of type I interferon signaling STAT1 protein deficiency in a gastric adenocarcinoma cell proteins is an early event in squamous skin carcinogenesis. line. Cancer Immunol Immunother 47: 113–120. Clin Cancer Res 8: 2067–2072. Araki Y, Okamura S, Hussain SP, Nagashima M, He P, Deb A, Haque SJ, Mogensen T, Silverman RH, Williams BR. Shiseki M et al. (2003). Regulation of cyclooxygenase-2 (2001). RNA-dependent protein kinase PKR is required for expression by the Wnt and ras pathways. Cancer Res 63: activation of NF-kappa B by IFN-gamma in a STAT1- 728–734. independent pathway. J Immunol 166: 6170–6180. Battcock SM, Collier TW, Zu D, Hirasawa K. (2006). Durbin JE, Hackenmiller R, Simon MC, Levy DE. (1996). Negative regulation of the alpha interferon-induced antiviral Targeted disruption of the mouse Stat1 gene results in response by the Ras/Raf/MEK pathway. J Virol 80: compromised innate immunity to viral disease. Cell 84: 4422–4430. 443–450. Biasco G, Rossini FP, Hakim R, Brandi G, Di Battista M, Dvorak HF. (1986). Tumors: wounds that do not heal. Di Febo G et al. (2002). Cancer surveillance in ulcerative Similarities between tumor stroma generation and wound colitis: critical analysis of long-term prospective programme. healing. N Engl J Med 315: 1650–1659. Dig Liver Dis 34: 339–342. Eberhart CE, Coffey RJ, Radhika A, Giardiello FM, Blache P, van de Wetering M, Duluc I, Domon C, Berta P, Ferrenbach S, DuBois RN. (1994). Up-regulation of Freund JN et al. (2004). SOX9 is an intestine crypt cyclooxygenase 2 gene expression in human colorectal transcription factor, is regulated by the Wnt pathway, adenomas and adenocarcinomas. Gastroenterology 107: and represses the CDX2 and MUC2 genes. J Cell Biol 166: 1183–1188. 37–47. Gil MP, Bohn E, O’Guin AK, Ramana CV, Levine B, Stark GR Blanco JC, Contursi C, Salkowski CA, DeWitt DL, Ozato K, et al. (2001). Biologic consequences of Stat1-independent Vogel SN. (2000). Interferon regulatory factor (IRF)-1 and IFN signaling. Proc Natl Acad Sci USA 98: 6680–6685. IRF-2 regulate -dependent cyclooxygenase Gupta RA, Dubois RN. (2001). Colorectal cancer prevention 2 expression. J Exp Med 191: 2131–2144. and treatment by inhibition of cyclooxygenase-2. Nat Rev Brown JR, DuBois RN. (2005). COX-2: a molecular target for Cancer 1: 11–21. colorectal cancer prevention. J Clin Oncol 23: 2840–2855. Guy-Grand D, DiSanto JP, Henchoz P, Malassis-Seris M, Bruewer M, Luegering A, Kucharzik T, Parkos CA, Madara JL, Vassalli P. (1998). Small bowel enteropathy: role of Hopkins AM et al. (2003). Proinflammatory cytokines intraepithelial lymphocytes and of cytokines (IL-12, IFN- disrupt epithelial barrier function by apoptosis-independent gamma, TNF) in the induction of epithelial cell death and mechanisms. J Immunol 171: 6164–6172. renewal. Eur J Immunol 28: 730–744.

Oncogene IFNs inhibit COX-2 expression L Klampfer et al 2081 Hanna N, Bonifacio L, Reddy P, Hanna I, Weinberger B, early-stage non-small-cell lung cancer. Lung Cancer Murphy S et al. (2004). IFN-gamma-mediated inhibition of 45(Suppl 2): S145–S150. COX-2 expression in the placenta from term and preterm Navarro A, Anand-Apte B, Tanabe Y, Feldman G, Larner AC. labor pregnancies. Am J Reprod Immunol 51: 311–318. (2003). A PI-3 kinase-dependent, Stat1-independent signal- Higaki S, Akazawa A, Nakamura H, Yanai H, Yoshida T, ing pathway regulates interferon-stimulated monocyte adhe- Okita K. (1999). Metaplastic polyp of the colon develops sion. J Leukocyte Biol 73: 540–545. in response to inflammation. J Gastroenterol Hepatol 14: Perk J, Iavarone A, Benezra R. (2005). Id family of helix– 709–714. loop–helix proteins in cancer. Nat Rev Cancer 5: 603–614. Hu J, Roy SK, Shapiro PS, Rodig SR, Reddy SP, Platanias Qing Y, Stark GR. (2004). Alternative activation of STAT1 LC et al. (2001). ERK1 and ERK2 activate CCAAAT/ and STAT3 in response to interferon-gamma. J Biol Chem enhancer-binding protein-beta-dependent gene transcription 279: 41679–41685. in response to interferon-gamma. J Biol Chem 276: 287–297. Ramana CV, Gil MP, Han Y, Ransohoff RM, Schreiber RD, Janabi N, Jensen PN, Major EO. (2004). Differential effects of Stark GR. (2001). Stat1-independent regulation of gene interferon-gamma on the expression of cyclooxygenase-2 expression in response to IFN-gamma. Proc Natl Acad Sci in high-grade human gliomas versus primary astrocytes. USA 98: 6674–6679. J Neuroimmunol 156: 113–122. Ramana CV, Gil MP, Schreiber RD, Stark GR. (2002). Stat1- Ji P, Diederichs S, Wang W, Boing S, Metzger R, Schneider PM dependent and -independent pathways in IFN-gamma- et al. (2003). MALAT-1, a novel noncoding RNA, and dependent signaling. Trends Immunol 23: 96–101. thymosin beta4 predict metastasis and survival in early-stage Ramana CV, Grammatikakis N, Chernov M, Nguyen H, non-small cell lung cancer. Oncogene 22: 8031–8041. Goh KC, Williams BR et al. (2000). Regulation of c-myc Kaplan DH, Shankaran V, Dighe AS, Stockert E, Aguet M, expression by IFN-gamma through Stat1-dependent and - Old LJ et al. (1998). Demonstration of an interferon gamma- independent pathways. EMBO J 19: 263–272. dependent tumor surveillance system in immunocompetent Ramsauer K, Sadzak I, Porras A, Pilz A, Nebreda AR, mice. Proc Natl Acad Sci USA 95: 7556–7561. Decker T et al. (2002). p38 MAPK enhances STAT1- Klampfer L, Huang J, Corner G, Mariadason J, Arango D, dependent transcription independently of Ser-727 phosphor- Sasazuki T et al. (2003a). Oncogenic k-ras inhibits ylation. Proc Natl Acad Sci USA 99: 12859–12864. the expression of IFN-responsive genes through inhibition Schindler C, Shuai K, Prezioso VR, Darnell Jr JE. (1992). of STAT1 and STAT2 expression. J Biol Chem 278: 46278– Interferon-dependent tyrosine phosphorylation of a latent 46287. cytoplasmic transcription factor. Science 257: 809–813. Klampfer L, Huang J, Sasazuki T, Shirasawa S, Augenlicht L. Shankaran V, Ikeda H, Bruce AT, White JM, Swanson PE, (2003b). Inhibition of interferon gamma signaling by the Old LJ et al. (2001). IFN gamma and lymphocytes prevent short chain fatty acid butyrate. Mol Cancer Res 1: 855–862. primary tumour development and shape tumour immuno- Klampfer L, Huang J, Swaby LA, Augenlicht L. (2004). genicity. Nature 410: 1107–1111. Requirement of histone deacetylase activity for signaling by Sharma B, Iozzo RV. (1998). Transcriptional silencing of STAT1. J Biol Chem 279: 30358–30368. perlecan gene expression by interferon-gamma. J Biol Chem Kowanetz M, Valcourt U, Bergstrom R, Heldin CH, 273: 4642–4646. Moustakas A. (2004). Id2 and Id3 define the potency of Sheng H, Shao J, Dubois RN. (2001). K-Ras-mediated cell proliferation and differentiation responses to transform- increase in cyclooxygenase 2 mRNA stability involves acti- ing growth factor beta and bone morphogenetic protein. vation of the protein kinase B1. Cancer Res 61: 2670–2675. Mol Cell Biol 24: 4241–4254. Shirasawa S, Furuse M, Yokoyama N, Sasazuki T. (1993). Landolfo S, Guarini A, Riera L, Gariglio M, Gribaudo G, Altered growth of human colon cancer cell lines disrupted at Cignetti A et al. (2000). Chronic myeloid leukemia cells activated Ki-ras. Science 260: 85–88. resistant to interferon-alpha lack STAT1 expression. Hema- Shuai K, Schindler C, Prezioso VR, Darnell Jr JE. (1992). tol J 1: 7–14. Activation of transcription by IFN-gamma: tyrosine phos- Levy DE, Darnell Jr JE. (2002). Stats: transcriptional control phorylation of a 91-kD DNA binding protein. Science 258: and biological impact. Nat Rev Mol Cell Biol 3: 651–662. 1808–1812. Lindner DJ. (2002). as antiangiogenic agents. Curr Vila-del Sol V, Fresno M. (2005). Involvement of TNF and Oncol Rep 4: 510–514. NF-kappa B in the transcriptional control of cyclooxygen- Mariadason JM, Nicholas C, L’Italien KE, Zhuang M, Smartt ase-2 expression by IFN-gamma in macrophages. J Immunol HJ, Heerdt BG et al. (2005). Gene expression profiling of 174: 2825–2833. intestinal epithelial cell maturation along the crypt–villus Vinkemeier U, Cohen SL, Moarefi I, Chait BT, Kuriyan J, axis. Gastroenterology 128: 1081–1088. Darnell Jr JE. (1996). DNA binding of in vitro activated Matsuura H, Sakaue M, Subbaramaiah K, Kamitani H, Eling Stat1 alpha, Stat1 beta and truncated Stat1: interaction TE, Dannenberg AJ et al. (1999). Regulation of cycloox- between NH2-terminal domains stabilizes binding of two ygenase-2 by interferon gamma and transforming growth dimers to tandem DNA sites. EMBO J 15: 5616–5626. factor alpha in normal human epidermal keratinocytes and Wong LH, Krauer KG, Hatzinisiriou I, Estcourt MJ, squamous carcinoma cells. Role of mitogen-activated Hersey P, Tam ND et al. (1997). Interferon-resistant protein kinases. J Biol Chem 274: 29138–29148. human melanoma cells are deficient in ISGF3 components, Meraz MA, White JM, Sheehan KC, Bach EA, Rodig SJ, STAT1, STAT2, and p48-ISGF3gamma. J Biol Chem 272: Dighe AS et al. (1996). Targeted disruption of the Stat1 gene 28779–28785. in mice reveals unexpected physiologic specificity in the Wright KL, Weaver SA, Patel K, Coopman K, Feeney M, JAK-STAT signaling pathway. Cell 84: 431–442. Kolios G et al. (2004). Differential regulation of prostaglan- Muller-Tidow C, Diederichs S, Thomas M, Serve H. (2004). din E biosynthesis by interferon-gamma in colonic epithelial Genome-wide screening for prognosis-predicting genes in cells. Br J Pharmacol 141: 1091–1097.

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc).

Oncogene