Several current strategies have elucidated novel gene activities through directing cellular expression of arrayed genes in discrete physical loci (12–16). Although these approaches may enable the examination of gene function at a genome scale, cDNAs used in these recent studies were derived from a restricted or undefined set of genes. Systematic study of mammalian gene function at a genome level, however, would require the incorporation of both comprehensive sequence-annotated nucleic acid arrays and highly parallel experimental strategies.

High-Throughput (Retro) Transfection and Genome Functionalization Through Arrayed cDNA Transduction Analysis (GFAcT). Arrayed 384-well matrices containing 62.5 ng of cDNA per well were incubated with 20 μl of serum-free medium containing FuGENE 6 (Roche, Indianapolis) and indicated reporter plasmids, obtained from a commercial source (Clontech) unless otherwise noted, for ≈20 min. Approximately 37 ng per well of reporters containing AP-1 [phorbol 12-myristate 13-acetate (PMA)] (tandem 6× tgactaa), p53 (tandem 14× tgcctggacttgcctggcc), and 60 ng per well of erythropoietin (Epo) (18) responsive elements were used. Trypsinized cells [human embryonic kidney (HEK)293, HCT116, and HepG₂, respectively] were delivered to wells in 20 μl of serum-containing medium, and 384-well plates were placed in a humidified chamber (5% CO₂, 37°C). After ≈48 h, 40 μl of the luciferase assay reagent Bright-Glo (Promega) was added to each well, and luminescence was read within 10 min by using an Acquest Plate Reader (LJL Biosystems, Sunnyvale, CA). All fluid delivery was done by using Multidrop (Titertek, Huntsville, AL). Relative light activity was normalized on a per-plate basis, and genes were ranked according to median activation. Detailed specifications of this process can be found at http://function.gnf.org.

Data Analysis and Clustering of Activators. Readings for putative modulators across various AP-1-related reporters (Clontech) in HEK293 cells were determined as described above. Data were subsequently normalized to a negative control [pCMV-AsRed (Clontech)], baseline-subtracted for activation of a parental reporter vector pTAL (Clontech), and filtered for background activity. Resulting data were analyzed by linkage across arrays (reporters) and genes by using cluster software (Stanford University, Stanford, CA) and were visualized by using tree-view (Stanford University). All results shown depict the average of at least three experimental data points.

Small Inhibitor RNA (siRNA) Suppression Analysis. Approximately 8,000 HEK293 cells were plated into 96-well tissue culture plates and transfected with 25 ng of indicated siRNA per well (Fig. 3A; see Table 1, which is published as supporting information on the PNAS web site, www.pnas.org) by using Lipofectamine2000 reagent (Invitrogen). After 24 h of incubation, medium was replaced, and cells were transfected by using FuGENE 6 (Roche) with indicated cDNA along with AP-1(PMA)-luciferase (148 ng per well) and constitutive β-galactosidase (5 ng per well) reporter vectors. After 48 h of incubation, single-well luciferase and β-galactosidase activities were determined by using Dual-Lite (Applied Biosystems) luminescence detection reagents. Data were normalized for transfection efficiencies, and activities for each series of AP-1 activators were compared with baseline values in wells with negative control siRNAs (pGL2, scrambled). Data were analyzed by using a hierarchical clustering algorithm (cluster, Stanford University) and visualized in treeview (Stanford University). Sequences of siRNAs are listed in Table 1.

Short Hairpin RNA (shRNA) and cDNA Subcloning. shRNA oligonucleotides were designed and cloned as described (19). After sequence verification, constructs were used in transient transfection analysis. Sequences of shRNAs are listed in Table 1. Down-regulation of endogenous transcript levels was verified by semiquantitative RT-PCR analysis. Transfer of cDNAs into replication-competent avian sarcoma leukosis virus LTR with splice acceptor (RCAS) retroviral vector (20) was mediated by Gateway LR Clonase (Invitrogen). Recombinant viral vectors were sequence verified and used to generate viral supernatants.

Western Blot and Antibody Reagents. Isolated cellular extracts were prepared, separated, and transferred onto nitrocellulose membranes as described (21). Antibodies used in these studies were purchased directly from suppliers and include c-Fos (Santa Cruz Biotechnology), c-Jun (Pharmingen), anti-rabbit IgG(H+L) horseradish peroxidase (HRP) (Promega), anti-mouse IgG HRP (Promega), Erk 1 and 2 (Santa Cruz Biotechnology), and γ tubulin (Santa Cruz Biotechnology).

Cell Culture and Growth Curves. HCT116 cells were cultured in McCoy 5A (GIBCO/BRL) media with 10% FBS, and 5 × 10⁵/ml were plated in 384-well plates. HepG₂ cells were cultured in DMEM (GIBCO/BRL) supplemented with 3% FBS at 37°C, 5% CO₂, and plating densities of 1.8 × 10⁶/ml were used for screening. HEK293 cells plated at a final density of 3.75 × 10⁵ cells per ml were cultured in DMEM (GIBCO/BRL), supplemented with 10% FBS, at 37°C, 5% CO₂. Primary chicken embryo fibroblast (CEF) cells were maintained and infected with AP-1 activator clones in subtype A RCAS virus and passaged three times, as described (22). Individual AP-1 activator-expressing CEF lines were then seeded in six-well plates at a density of 0.5 × 10⁶ cells per well. The next day (day 0), cells were counted in triplicate for each respective clone, and counts were performed thereafter on days indicated.

In addition to elucidating on the order of 10² gene activities, our analysis also, as expected, identified modulators that have a known association with their respective pathways. For instance, c-FOS, c-JUN, GADD45 β and γ, TRAF6, ERB-B-2, FOSL1, ATF3, MAP3K11, RICK, and PKCζ were among the top 0.5% of activators in the AP-1 screen (Table 3 and data not shown). Furthermore, of the canonical FOS and JUN family members represented in the GFAcT cDNA matrix (FOS, FRA-1, FRA-2, JUNB, and JUND), only JUNB did not score in the top 3% of screened activators (Table 3 and data not shown). Surprisingly, this relatively unbiased survey also revealed a high proportion of functionally novel activities within the set of identified Epo, p53, and AP-1 effectors (Fig. 1D), suggesting that a significant portion of the cellular machinery that governs these biologically important pathways remains uncharacterized.

AP-1 activity is regulated by a variety of extracellular stimuli, including growth factors, cytokines, cell–matrix interactions, and genotoxic stress, among others (24). Although all of the activating genes were identified through their abilities to modulate AP-1(PMA)-dependent reporter gene transcription, we used a response-pathway mapping strategy to further classify novel AP-1 inducers on the basis of comparative activation of related reporter constructs. Specifically, we profiled relative activities of AP-1 modulators on reporter elements responsive to growth and proliferation signals {serum response element (SRE), NFATc response elements (NFAT), and PMA response elements [AP-1(PMA)]} (25, 26), and also those that preferentially monitor inflammation, IFN, or stress stimuli [cAMP response element-binding protein (CREB) response elements (CRE), IFN response elements (ISRE), and AP-1 response elements (AP-1)] (27, 28). Interestingly, hierarchical cluster analysis (29) of these data resulted in the co-segregation of reporter activities in relation to specific facets of AP-1 signaling (Fig. 2A, Table 5, which is published as supporting information on the PNAS web site). This allowed for distinctions between encoded proteins likely involved with genotoxic stress and (proinflammatory) cytokine signaling (ISRE, AP-1, CRE), and, conversely, those affiliated with growth factor and serum response [SRE, AP-1(PMA), NFATc response elements].

We further investigated the mechanisms by which the latter subset of genes potentially regulates AP-1 growth and proliferation responsive pathways. Consistent with their postulated roles, transfection of most cDNAs from these predicted subsets resulted in the coordinate induction of c-Fos and/or c-Jun protein levels, as demonstrated by Western blot analysis (Fig. 2B). However, expression of a few encoded proteins in this grouping, such as TYRO3, P2RY2, and UBB, did not result in detectable elevated levels of the prototypic AP-1 transcription factors (data not shown), which may indicate that they modulate AP-1 activity through alternate family members or through as-yet-unknown mechanisms. Although we surveyed the regulation of only a limited number of AP-1 proteins by a subset of effectors, these results indicate that significant numbers of encoded proteins identified through GFAcT analysis are likely bona fide modulators of endogenous AP-1 activity. Intriguingly, analysis of human tissue mRNA levels revealed that many of these identified growth-associated AP-1 modulators and canonical AP-1 transcription factors were coincidentally expressed in cells of lymphoid origin (Fig. 2C and Fig. 5, which is published as supporting information on the PNAS web site), suggesting a functional alignment between these molecules in proliferating cell types.

To refine the mode of action by which putative growth regulatory proteins modulate the AP-1 pathway, we assessed their relative inductive activities in the presence of siRNA molecules directed against known components of AP-1 signaling. AP-1(PMA) reporter gene activities resulting from pairwise cDNA (novel activators) and siRNA (AP-1 pathway members) transfections were normalized to appropriate controls and analyzed through single-linkage hierarchical clustering (Fig. 3A and Table 1). For comparison, effects of the inhibitory RNAs on PMA-induced AP-1 reporter gene activity were also determined. siRNAs encoding PI-3 kinase (PIK3R3, PIK3CD), PKC (PKCD, PRKACA, DGKI, MEKK2), and RAS (SRC, RAF1) pathway members were most proficient in abrogating AP-1 induction by the bulk of the elucidated putative growth regulators. This demarcates a majority of identified AP-1 activators as functioning upstream of these proteins and suggests that they regulate AP-1 activity through modulation of PI-3 kinase, PKC, and/or RAS pathway members. For example, the activity of B lymphocyte kinase (BLK) on the AP-1 reporter gene was extinguished by siRNAs coding for ERK pathway effectors DGKI, MEKK1, SRC, RAF1, and PIK3R3 (blue), but not by siRNAs directed against JNK-related pathway molecules MAP3K7, MAPK9, and MAP4K4 (yellow). Although we cannot exclude the possibility of unintended (“off-target”) siRNA effects on this analysis (see Table 1), these activities are consistent with studies that have demonstrated the importance of PI-3 kinase, PKC, and RAS function for AP-1 response to proliferation stimuli (30, 31). In sum, this genetic suppression analysis approach led to the identification of AP-1 pathway components required for the inductive activities of novel modulators and further defined the junction at which these molecules intersect the AP-1 signaling cascade.

We additionally examined the roles of the predicted growth-associated AP-1 modulators in mitogenic response. By using DNA expression plasmids encoding shRNAs (19) directed against a subset of eight of the growth-related pathway activators, we screened for the inhibition of PMA-induced AP-1 activity through reporter gene analysis (data not shown). On the basis of their potent antagonism of PMA-induced reporter activity, we selected shRNAs encoding sequences homologous to the BLK and transitional epithelia response 1 (TERE1) genes for further characterization. Through Western blot analysis, we demonstrated that transfection of these shRNAs into PMA-stimulated cells resulted in a significant decrease in c-Fos induction in response to phorbol ester (Fig. 3B). These results reveal that these encoded proteins play critical roles in AP-1 regulation by mitogenic stimuli in a human embryonic kidney cell line. Examination of Blk and Tere1 protein complexes by MS analysis suggests that these proteins are possibly mediating AP-1 function through the indirect modulation of SRC and RAF-1 activities and of intracellular biosynthetic intermediate (i.e., ATP, serine derivatives) levels, respectively (Fig. 6, which is published as supporting information on the PNAS web site).

Given the role of AP-1-related proteins in proliferation and cancer-related phenotypes, we alternately assessed the ability of these molecules to impart a growth advantage to avian cells. Proliferative control in primary CEF cells is particularly susceptible to modulation by AP-1 effectors (31, 32). Growth analysis of CEF ectopically expressing a subset of putative AP-1 growth regulatory genes demonstrated that these molecules can increase rates of proliferation in this primary cell type (Fig. 4A). We further investigated the oncogenic potential of these genes in transformation assays. Previous reports have shown that expression of TYRO3 can mediate focus formation in an immortalized cell type (RAT1 fibroblasts), and an activated mutant of BLK has also previously been demonstrated to possess oncogenic activities (33, 34). Here, we have shown that expression of wild-type BLK induced focal microtumors on monolayers of CEF cells (Fig. 4B), and ectopic TYRO3 activity was sufficient to promote anchorage-independent growth of CEF (Fig. 4C). Infection of CEF with retroviruses encoding Centaurin-α 1 (CENTA1) and the hypothetical protein KIAA1847, which we have designated TIAP (Tudor-domain containing inducer of AP-1), also resulted in anchorage independence (Fig. 4C). Colonies derived from ectopic expression of these genes have more diffuse borders in contrast to the compact colonies induced by c-JUN or TYRO3. These results confirm the in vivo proliferative activities of several predicted growth-associated AP-1 activators identified through GFAcT analysis and suggest that the oncogenic activities of BLK, TYRO3, CENTA1, and TIAP are likely mediated, at least in part, through modulation of AP-1 activity.

Pathway and cell-based functional annotation is an important methodology in deciphering the biological activities of genes identified by the genome sequencing projects. We describe here a method for genome-scale functional annotation of human genes with respect to AP-1, p53, and hypoxic (Epo) signaling. This approach is currently limited to the study of individual gene function through ectopic expression. Although overexpression of genes may activate legitimate signaling molecules in the absence of appropriate stimuli, in certain cases it can also result in the induction of physiologically irrelevant phenotypes. We have endeavored to exclude artifacts of overexpression by characterizing a subset of AP-1 activators in expanded and independent tests.

Twenty-five of the most potent AP-1 modulators identified by GFAcT analysis were further classified through reporter-based pathway mapping, identifying those genes that are likely to mediate AP-1-related proliferative activities. siRNA-based functional suppression analysis was used to demonstrate that this subset of modulators probably work through RAS/PKC/PI-3K pathways, and further characterization of these proliferation-associated genes revealed the requirement of BLK and transitional epithelia response 1 for unabridged AP-1 response to mitogenic stimulation. Finally, cellular analysis confirmed that many of the identified genes can function to augment proliferation in a primary cell type and led to the identification of Centaurin-α 1 and TIAP as mediators of cellular transformation.