Target Selectivity of Vertebrate Notch Proteins

All four mammalian Notch proteins interact with a single DNA-binding protein (RBP-jκ), yet they are not equivalent in activating target genes. Parallel assays of three Notch-responsive promoters in several cell lines revealed that relative activation strength is dependent on protein module and promoter context more than the cellular context. Each Notch protein reads binding site orientation and distribution on the promoter differently; Notch1 performs extremely well on paired sites, and Notch3 prefers single sites in conjunction with a proximal zinc finger transcription factor. Although head-head sites can elicit a Notch response on their own, use of CBS (CSL binding site) in tail-tail orientation is context-dependent. Bias for specific DNA elements is achieved by interplay between the N-terminal RAM (RBP-jκ-associated molecule/ankyrin region), which interprets CBS proximity and orientation, and the C-terminal transactivation domain that interacts specifically with the transcription machinery or nearby factors. To confirm the prediction that modular design underscores the evolution of functional divergence between Notch proteins, we generated a synthetic Notch protein (Notch1 ankyrin with Notch3 transactivation domain) that displayed superior signaling strength on the hes5 promoter. Consistent with the prediction that “preferred” targets (Hes1) should respond faster and at lower Notch concentration than other targets, we showed that Hes5-GFP was extinguished fast and recovered slowly, whereas Hes1-GFP was inhibited late and recovered quickly after a pulse of DAPT in metanephroi cultures.

The Notch pathway constitutes a short range cellular communication channel involved in all fundamental aspects of multicellular life. Mutations that cause misregulation or misexpression of Notch pathway proteins have been directly linked to multiple human disorders from developmental syndromes to adult onset diseases and cancer (1)(2)(3).
Biochemical experiments established that ligand-mediated activation of Notch receptors triggers ectodomain shedding (mediated by a metalloprotease); the intermediate peptide (NEXT) (4) is rapidly cleaved by the enzyme ␥-secretase to release the Notch intracellular domain (NICD). 3 NICD translocates to the nucleus where it interacts with a DNA-bound protein (CBP or RBP-j in vertebrates, Su(H) in Drosophila, Lag-1 in Caenorhabditis elegans; collectively referred to as CSL) (5,6)). In the absence of NICD, CSL proteins complex with ubiquitous co-repressor proteins, such as SMRT (silencing mediator of retinoic and thyroid) hormone receptors or Mint (7), to repress transcription. The limiting amounts of NICD generated upon ligand binding somehow compete with the abundant repressor proteins and form an NICD-CSL complex. The NICD/CSL interface is recognized by Mastermind (MAM)/Lag-3 protein, and this tri-protein complex recruits the ARC-L/MED mediator complex (8), the histone ubiquitin ligase, Bre-1 (9), and histone acetyltansferases to assemble an active transcription complex on target promoters. Subsequently, CycC/CDK8, a component of ARC-L/MED, phosphorylates the NICD PEST domain triggering NICD ubiquitination by Fbw/sel 10 ubiquitin ligase and proteasome-mediated degradation (8). This signaling paradigm is referred to as the canonical pathway.
All Notch proteins share similar domain architecture; the extracellular domain includes multiple epidermal growth factors (10) and three LNR (Lin-Notch repeats) (11). The intracellular domain contains the RAM (RBP-j-associated molecule) domain, followed by the highly conserved seven ankyrin (ANK) repeats, of which only the six C-terminal ones assume an ankyrin fold. Nuclear localization signal sequences and a PEST domain regulating protein stability are found C-terminal to the ANK domain (Fig. 1A). In mammals, four receptors and seven ligands have been identified thus far. It has been reported that Notch1 and -2, but not Notch3 or -4, contain a transactivation domain (TAD) located C-terminal to the ANK repeats (12,13). This region contains phosphorylation sites that may allow other signaling pathways to selectively modulate Notch activity (14,15). Notch1 and -2 have essential functions (16 -18), whereas Notch3 and -4 are not necessary for survival (19,20). Notch1 and -3 were considered to be at the extreme functional ends in their ability to activate transcription (21,22), perhaps because of the lack of TAD. However, a specific function for Notch3 in the differentiation of arterial smooth muscle cells was described recently (23).
NICD/CSL interaction is mediated mainly through the unstructured RAM domain (24,25), but the ANK domain also contributes to this interaction (24,26) and is absolutely essential for MAM recruitment into the NICD-CSL complex (25,27). A third CSL interaction domain was recently identified and mapped to the C-terminal 200 residues, overlapping the nuclear localization signal sequences (28). Because of their importance for Notch-mediated activation, the ANK and TAD domains are hypothesized to cooperate to provide Notch proteins with their specific transactivation properties. Despite the number of ligands and receptors, a single CSL protein mediates all canonical Notch signals in vertebrates (RBP-j (29)).
The presence of four mammalian Notch receptors and many ligands led to the suggestion that each Notch protein may target a discrete set of downstream genes. Support for this possibility came from several recent experiments. During the differentiation of naive CD4 cells to either Th1 or Th2 fate, it was proposed that the ligand Delta signals to Notch3 in order to drive Th1 fate (30) and Jagged signaled through Notch1 and -2 to trigger the Th2 fate (31). Notch paralogs may even play contrasting roles in the same tissue. Notch1 and Notch2 have opposite effects on embryonal brain tumor growth through activation of different target genes (32); Notch1 may act as an oncogene (33) and Notch2 may play a tumor-suppressor role in different stages of human breast cancer (34).
The molecular basis for context dependence was investigated in cultured cells, where each of the four Notch paralogs has been known to elicit different quantitative responses from different promoters (12,13,29,(35)(36)(37)(38)(39)(40). Beatus et al. (12) have discovered that target activation depends on the composition of the ANK repeats, although the isolated domain by itself has no intrinsic activation property (13,26). Given that all Notch proteins bind DNA by virtue of their association with the CSL protein, an attractive model for specificity is that the diversity within ANK repeats among Notch proteins will dictate recruitment of different Mastermind (MAM) proteins, leading to the assembly of discrete activation complexes. However, this model cannot fully account for the variation in activity because the same MAM protein imparts different transcriptional activities to different Notch paralogs (35,36).
One caveat in these earlier studies stems from the fact that the domain boundary between ANK and the C-terminal regions has been redefined recently through crystallographic analyses (5,42,43); some of the previous studies deleted part or all of the 7th ANK repeat (5). In addition, a systematic comparison of the transactivation potential of NICD1-4 on different promoters within the same cell type has not been done. To understand further the mechanism underlying variation in transcriptional activation of Notch homologs, we decided to re-examine the contribution of individual domains of different Notch paralogs to their distinct transcriptional activation properties.
We report that the ability of the four Notch proteins to activate a given promoter was only partially dependent on the cell type. The arrangement and spacing of CBS, as well as their distance from the transcriptional start site, were important determinants in target selectivity and activation amplitude (as shown recently to be the case in Drosophila (44)). Notch1 activates paired CBS (in both head-to-head and tail-to-tail orientation) very efficiently; Notch3, a poor activator overall, is best at activating promoters with a single CBS, but additional cis-elements are required. We confirmed that the key to the activation potency of all Notch paralogs is encoded in part within their respective RAM-ANK domains and in part within their C-terminal domain/TAD. Contrary to earlier reports, we demonstrate that Notch3 has a potent but specific TAD best suited for activation of hes5 promoters, probably through interaction with the putative zinc finger protein. To test the Notch3 TAD domain, we engineered a "super NICD" with Notch1 ANK and Notch3 TAD that strongly induced the weak hes5 promoter. This latter observation is of particular importance, given the role of Notch3 and zinc finger proteins in the differentiation of vascular smooth muscle cells (23,45,46). In addition, we used ␥-secretase inhibition in transgenic metanephroi expressing Hes1-GFP or Hes5-GFP to demonstrate that preferred targets identified in vitro are more resistant to Notch inhibition in vivo, most likely because they are active at a lower NICD concentration.

Cell Culture and Transfection
Culture Conditions-N2a and HeLa cells were maintained in minimum essential Eagle's medium supplemented with 10% fetal bovine serum (FBS) and 2 mM L-glutamine. 293T and MCF7 cells were main-tained in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% FBS, 2 mM L-glutamine. C2C12 cells were maintained in DMEM supplemented with 10% FBS, 4 mM L-glutamine. NIH3T3 cells were maintained in DMEM supplemented with 10% bovine calf serum and 4 mM L-glutamine. MK4 cells were maintained in DMEM supplemented with 10% heat-inactivated FBS, 4 mM L-glutamine. D4T endothelial cells were grown in Iscove's modified Eagle's medium supplemented with 10% FBS, 2 mM L-glutamine, and 1-thioglycerol (Sigma). In addition, all media contained 0.1 mM nonessential amino acids, 1.0 mM sodium pyruvate, and 100 units/ml penicillin/streptomycin.
Optimization-NIH3T3 cells (seeded in 24-well plates) were transfected with an ascending concentration series of NICD DNA (2.5, 5, 10, 25, 50, 75, and 100 ng of DNA/0.5 ml of media) in 410 ng of total DNA that includes 10 ng of pCS2ϩ/␤-galactosidase, 50 ng of reporter construct, and pCS2ϩ as carrier DNA per well (0.5 ml). Each NICD paralog elicited optimal reporter expression levels at different DNA concentrations (for NICD-1 and -2, 50 ng of DNA was used; for NICD-3 and -4, 10 -25 ng was used). The NICD concentrations that elicited optimal luciferase activity were subsequently used to survey other cell lines. For Figs. 5 and 7, a similar titration experiment was conducted to determine the optimal response for each Notch hybrid construct or mutant promoter.
Experiments-The cells were transfected by calcium phosphate method in BES-buffered saline (BBS) adapted from Ref. 41. Cells seeded on gelatin-coated 24-well plates were re-fed 3 h before transfection. With the exception of N2a, which used BBS at pH 6.8, the reaction mixture for all cell types was prepared in BBS at pH 6.75. 410 ng of total DNA/0.5 ml of media was used for NIH3T3, C2C12, 293T, N2a, MK4 cells, whereas MCF7, HeLa and D4T cells used twice the amount (820 ng) of total DNA. Cells were washed and re-fed with medium 12-16 h post-transfection.
For the experiments in Figs. 5-7, the three different DNA concentrations that elicit optimal responses were transfected for each Notch construct. Only the highest activity was used for comparison. Expression level of NICD from the lysate was determined by Western blot analyses.

Luciferase Assays
Cells were harvested 48 h after transfection and assayed by methods adapted from Ref. 41. Cells were washed once with phosphate-buffered saline (PBS) and lysed in 100 l of lysis buffer (100 mM KPO 4 buffer, pH 7.8; 0.2% Triton; 1 mM dithiothreitol (DTT); protease inhibitors) at room temperature for 10 min. 5 l of lysate was used to determine ␤-galactosidase concentration to normalize for transfection efficiency. These assays were performed according to the Tropix Galacton chemiluminescent substrate instructions. 50 l of lysate incubated with luciferin assay buffer (30 mM Tricine, pH 7.8; 3 mM ATP; 15 mM MgSO 4 ; 10 mM DTT; 0.2 mM CoA; 1 mM luciferin) was used to determine luciferase activity using a Tropix TR717 luminometer.

Western Blot
Laemmli SDS sample buffer (ϩ10 mM DTT) was added directly to cell lysates from luciferase assay and boiled for 10 min. All protein samples were resolved by 10% SDS-PAGE in 25 mM Tris, 192 mM glycine, 0.1% SDS buffer. Proteins were then transferred to nitrocellulose in 25 mM Tris, 192 mM glycine, 20% methanol buffer. Blots were incubated with primary antibody (M2 monoclonal ␣-FLAG at 1:1000 dilution; ␣-Actin at 1:5000 (Sigma)) in 0.1% Tween and 5% milk in PBS at 4°C for overnight. After three washes, the membranes were incubated with secondary antibody (␣-mouse Ig-horseradish peroxidase (Amersham Biosciences)) diluted 1:5000 in 0.1% Tween PBS for 1 h at room temperature. After three rinses, the protein was visualized with Super-Signal West Pico Chemiluminescent kit (Pierce) as per the Hyperfilm TM MP instructions (Amersham Biosciences).

Immunohistochemistry
NIH3T3 cells seeded on a 24-well plate were transfected with NICD and Hes1-luciferase as described above. They were trypsinized 24 h post-transfection and seeded equally at 80% confluency onto three gelatin-coated 8-well chamber slides. Cells were harvested and examined 48 h post-transfection. One chamber slide was subjected to luciferase and ␤-galactosidase assays. The other slides were washed twice with PBS and fixed in 4% paraformaldehyde for 30 min at room temperature. After three washes, cells were permeabilized and blocked in 3% bovine serum albumin and 0.1% Triton X-100 in PBS. Incubation with primary antibodies (mouse M2 monoclonal ␣-FLAG at 1:5000 (Sigma) and goat ␣-luciferase at 1:1000 (Promega)) was carried out in blocking buffer overnight at 4°C. After three washes, cells were incubated sequentially with donkey ␣-goat Ig-Alexa Fluro-488 conjugate and goat ␣-mouse Ig-Alexa Fluro-594 conjugate (Molecular Probes) with 4,6-diamidino-2-phenylindole (1:2500) at room temperature for 1 h. Cells were washed three times and mounted for visualization. An area of 49 mm 2 was scanned using ImagePro-Plus and Volumescan software.

Transcription Factor Binding Sites, Search and Statistical Analyses
Putative TF-binding sites were identified using transcription element search system (www.cbil.upenn.edu/tess/). Student's two-tailed t test was used to evaluate statistical significance. F test was used to determine whether the two pools of values have equal or unequal variance. Statistical significance was defined as p value of 0.05 or less.

Identification of the ZF-CSL Regulatory Module
Because the CSL binding profile is not available in the TRANSFAC or the JASPAR data base, we constructed one using experimentally vali-dated binding sites collected from the literature. The detailed methods are provided in the Supplemental Material.

Metanephric Organ Culture and Immunohistochemistry
Metanephric organ cultures were performed as described (47). Briefly, embryos were harvested from CD1 females mated with Hes1-GFP or Hes5-GFP males. Vaginal plug detection was considered as embryonic (E) day 0.5; kidneys were removed at E14.5 and cultured on transwell filters (Falcon, pore size 1 m) at an air/fluid interface in a serum-free medium consisting of equal volumes of Dulbecco's modified Eagle's medium and Ham's F-12 medium containing 25 mM HEPES, sodium bicarbonate (1.1 mg/ml), 10 nM Na 2 SeO 3 ⅐5H 2 O, 10 Ϫ11 M prostaglandin E 1 , and iron-saturated transferrin (5 g/ml). From each embryo, one kidney was cultured with Me 2 SO (1 l/ml), and the other with N-(N-(3,5-difluorophenacetyl)-L-alanyl)-S-phenylglycine t-butyl ester (DAPT) in Me 2 SO (at a final concentration of 1 M in the medium, 100 ϫ IC 50 ). In all cases, the medium was refreshed twice a day to maintain the inhibition by DAPT. After various time points, DAPT and Me 2 SO were omitted. GFP fluorescence was recorded directly under a camera equipped with stereo microscope capable of UV epi-illumination at regular intervals during the experimental period.
For immunohistochemistry, the E17.5 kidneys were fixed in 4% paraformaldehyde for 4 h, immersed in 30% sucrose, and cryostat-sectioned at 16 m. The sections were incubated with anti-Pax-2 (1:200; Covance) antibody, followed by Cy3-conjugated anti-IgG to visualize the antigen. GFP fluorescence could be seen directly without any additional manipulation.

Activation Strength of Notch Proteins Is Dependent on the Promoter and Cellular
Context-Because all vertebrate Notch proteins recognize DNA targets by virtue of their association with the RBP-j protein, one possible explanation for differential responses to transfected Notch proteins could lie in their ability to interact with tissue-specific partner(s) (48). To reassess the mechanistic basis for target activation and selectivity by four vertebrate Notch paralogs, we constructed four truncated proteins that included the entire intracellular domain of the respective Notch tagged with three FLAG epitopes at their N terminus (Fig. 1A), and we conducted a series of parallel experiments with the four proteins in different cell lines.
First, we determined the DNA concentration that produced optimal activation for each Notch protein (see "Experimental Procedures" for details and see supplemental Fig. S1). Next, we assayed the ability of the four NICDs (each at its optimal DNA concentrations) to activate three different promoters (one synthetic, 4xCBS, and two endogenous promoter fragments from Hes1 and Hes5; see . Under these experimental conditions, we observed that although the absolute fold of promoter activation above base line was cell type-dependent, the relative strength of plasmid promoter activation by each Notch paralog was essentially the same in different cellular contexts (Fig. 1C). NICD1 elicited the strongest activation on all three promoters, whereas NICD3 was the weakest on the 4xCBS and Hes1, as observed by others (N1 Ͼ N2 ϭ N4 Ͼ N3 (35,36,38)). In contrast, the ability of Notch3 to activate Hes5 was consistently better than that of NICD4 in all the cells we examined ( Fig.  1C; supplemental Fig. S1E).
Dependence on cell context was observed only in 293T cells (where NICD3 was a better activator of 4xCBS than NICD4) and in MCF7 cells (where NICD2 was the most potent signaling molecule on all promoters). Most interestingly, activation of Hes1 by any NICD was consis- Arrowhead in forward direction indicates that the CBS is on the sense strand, and inverted arrowhead indicates that CBS is on the antisense strand of the DNA. Synthetic promoter 4xCBS contained four tandem repeats of high affinity CBS (black arrowhead) arranged in head-to-tail orientation. Both Hes1 (354 bp) and Hes5 (879 bp) are endogenous mouse promoters that contain both low affinity (gray arrowhead) and high affinity CBS. C, distribution of fold activation exhibited by four Notch paralogs on three promoters in different cell lines. 4xCBS and hes1 promoters generally exhibited an N1 Ͼ N2 ϭ N4 Ͼ N3 activation profile, and Notch3 consistently outperforms Notch4 on the hes5 promoter. The activity of each Notch paralog is first optimized in NIH3T3 by titration experiment (SF1). Results are representative of at least three independent experiments. FEBRUARY 24, 2006 • VOLUME 281 • NUMBER 8

JOURNAL OF BIOLOGICAL CHEMISTRY 5109
tently higher than Hes5 in most of cell types except for N2a cells; in these neuroblastoma-derived cell, Hes5 was activated to a higher fold than Hes1. The basis for this difference in response was not pursued further in this study.
The different response to NICD proteins could reflect differences in the probability of promoter occupancy in a transfected cell (digital model (49)) or in the amplitude of activation (analog model, the same relative fraction of transfected cells activates the promoter but some NICD-containing transcription complexes are less stable and produce fewer transcripts). Because these transfection-based assays do not address chromatin-packed, endogenous targets, a pseudo-analog model can also exist; each of multiple transfected targets responds digitally (either on or off), but the sum of responding promoters in a transfected cell differs for each paralog. To differentiate between these possibilities, we determined the fraction of transfected 3T3 cells containing Notch and the reporter proteins using antibodies to FLAG and luciferase. Comparing NICD1 (strongest activator) or NICD3 (weakest activator) activation of Hes1-lucifease in individual transfected cells revealed a similar percentage of responding cells (Notch1, 93.5%, n ϭ 170; Notch 3, 86%, n ϭ 215) but higher intensity of luciferase in NICD1-transfected cells than NICD3-transfected cells as determined by image quantification (supplemental Fig. S2, A and B). 34.1% of Notch1-transfected cells fall on or above the median intensity level, whereas only 8.8% of Notch3transfected cells reach this plateau. Thus, either an analog or a pseudoanalog model best explains the quantitative differences between Notch paralogs that will be further analyzed here.
Orientation of CBS Is Critical for NICD Activation on hes1 Promoter-Notch signaling is mediated through CSL that can bind to either a low (RTGRGAR) or high affinity (YGTGRGAA) CBS (50). Many of the best characterized Notch target promoters (E(spl) complex in Drosophila and Hes1 in mouse) contain a highly conserved "SPS" (Su(H) paired sites (50)) and paired, inverted CBS (CBS-A and CBS-B) separated by 15-22 nucleotides ( Fig. 2A). A conserved hexameric site (G/CAAAGT) is often located between the paired CBS sites. Mutating this site did not alter Hes1 activation by Notch (51). A low affinity CBS (site C) lies upstream of CBS-A in the hes1 promoter. the region between Ϫ248 to ϩ1 on hes5 promoter and a single CBS (site E) is sufficient to elicit an N3 Ͼ N4 response. C, different arrangement of paired CBS on Hes5 has contrasting effects on four Notch paralogs. Tail-to-tail paired CBS with Hes1 intervening sequences specifically enhances Notch1 activity (⌬(Ϫ214 to Ϫ79)), whereas tail-to-tail paired CBS in endogenous context specifically reduces Notch3 activity (D16). CBS in head-tohead orientation strongly increases the overall activation of all Notch proteins, albeit to a lower extend for Notch3 (InvD16). Results are representative of at least three independent experiments.
As reported previously, we find that intact CBS A and B are necessary for activation of the hes1 promoter (Fig. 2B, Single), and loss of either the low affinity CBS-C (Cm) or the conserved hexameric site (Hxm) had no effect on Hes1 activation by any of the four paralogs (Fig. 2B, data shown for WT construct). In fact, a minimal promoter containing only the paired site and a TATA sequence is sufficient to elicit ϳ60% luciferase activity (Fig. 2B, ABtata), confirming that the SPS is both necessary and sufficient for Notch-dependent gene expression (52,53) in vertebrate cells.
Next, we wished to determine whether restoring a high affinity CBS at site C will compensate for the loss of site B. Most surprisingly, converting site C to a high affinity site did not restore activation despite the presence of two paired high affinity CBS ( Fig. 2A, C-H, and Fig. 2B, T-T).
Restoring the spacing between C and A as to 16 nucleotides did not improve activation (C-H 16 ). In this configuration, the sites were in tailto-tail (T-T) orientation, and the potent SPS are head-to-head (H-H). To test if the orientation of CBS within the paired configuration might be important, we inverted site B (invB) without altering its sequence to get a head-to-tail orientation (Fig. 2B, H-T). This manipulation abrogated all Notch-mediated activations but not RBP-j binding or RBPj-NICD complex formation (Fig. 2C). Thus, although SPS is sufficient to enable Notch activation, the orientation of individual CBS within the Hes1 SPS is critical for its activation by all vertebrate Notch proteins (and perhaps for the activity of all paired sites (44)).
Additional CIS Elements Are Necessary for Notch Activation of hes5 Promoter-The observation that Notch3 activated the hes5 promoter more efficiently than NICD4 was surprising because previous studies proposed that Notch3 was either a repressor or the poorest activator of target promoters (13,35,36,38,40). The hes5 promoter does not contain paired CBS; instead, it has two high affinity CBS located 134 bases apart and three low affinity sites further upstream (Fig. 3A). To investigate Hes5 activation by the Notch paralogs, we mutated each of the two most proximal high affinity sites (Fig. 3A, CBS-D and CBS-E). Mutating CBS-E abolished Notch-mediated activation (Fig. 3A, Em), but mutating CBS-D resulted in 50% reduction of activation (Fig. 3A, Dm). CBS-D thus cooperates with CBS-E to attain maximum activation of Hes5 but is not able on its own to sustain any activation.
To determine whether placement of a high affinity CBS the same distance and orientation as in Hes5 could rescue a mutant SPS in Hes1, we inserted a CBS-D from hes5 into hes1-Bm promoter 134 bases upstream to the intact CBS-A ( Fig. 2A, BmD, and Fig. 2B, T-T). The addition of site D to Hes1 (BmD) did not restore Notch-mediated activation, suggesting that other essential cis-acting element(s) are present in the hes5 promoter and that these site(s) must work in concert with CBS-E to facilitate Notch-dependent activation. A shorter construct (Ϫ248 to ϩ1) delivers ϳ50% of Hes5 activity in response to NICD, perhaps because of the loss of three low affinity CBS or other upstream elements (Fig. 3B). This observation suggests that the cis-elements defining the core enhancer reside between Ϫ248 to ϩ1 region. More importantly, this truncated region with a single CBS still responds better to NICD3 than to NICD4.
Distance of SPS to the Transcriptional Start Site Affects Activation-We digressed to ask if inserting an SPS would alter the activation profile of the hes5 promoter. We used three different strategies, and surprisingly, each resulted in distinctive activation patterns. When CBS-D and -E were paired by replacing the intervening 134 endogenous bases with 16 nucleotides from the Hes1 paired sites (Fig. 3C, construct ⌬(Ϫ214 to Ϫ79)), Notch1-mediated activity was elevated 3-fold (from initial 14 to 50-fold activation, n ϭ 5), whereas Notch2-mediated activation was increased less than 2-fold (Fig. 3C). Most interestingly, ⌬(Ϫ214 to Ϫ79) responded better to Notch4 than to Notch3. These observations were in contrast to the hes1 promoter and Drosophila E(Spl) where the tail-totail orientation failed to trigger a Notch response. Because CSL proteins bind all high affinity sites, this observation was consistent with the notion that additional cis-acting elements present in Hes5 (but missing in Hes1 and E(spl)) were cooperating with the added CBS. The second construct (Fig. 3C, D16), where tail-to-tail CBS-D and -E pairing was created by transposing CBS-D 16 nucleotides upstream of CBS-E while retaining the intervening endogenous Hes5 sequences, significantly impaired Notch3-mediated activation (from 7-fold in Fig. 3A to less than 3-fold in Fig. 3C). This alignment demonstrates that the Hes1derived spacer provides optimal context for Notch1, and any SPS degrades the ability of Notch3 to activate.
Next, we recreated the SPS with head-to-head symmetry and the 16 nucleotides spacer from Hes1 by introducing site D immediately downstream of E (Fig. 3C, invD16). This construct was activated very efficiently by all NICD (Fig. 3D). Most interestingly, although Notch1, -2, and -4 proteins showed greater than 25-fold improvement, Notch3 exhibited less than 5-fold increase (from 7 to 35; n ϭ 4). Similar results were obtained when we fused either truncated or complete hes5 pro-  6). B, unlike activation, the two distantly spaced high affinity CBS did not exhibit differences in their repressive roles. However, the presence of a paired CBS in either head-head or tail-tail orientation always leads to enhanced repression. #, p ϭ 5e-6 (n ϭ 5); ##, p ϭ 9e-6; ###, p ϭ 1e-5; ####, p ϭ 3e-6 versus Hes5. Data shown are the average values of six independent experiments. Results are represented as mean Ϯ S.D. moter sequences upstream to the Hes1 or 4xCBS promoters (data not shown). Collectively, these experimental data support three additional conclusions. First, the distance between CBS is critical to elicit an optimal response from NICD (Fig. 3). Second, Notch paralogs evolved preferences for different configuration of CBS (Fig. 3C). Third, in the right context, tail-to-tail orientation SPS can mediate activation by Notch proteins.
To test if SPS acts as an enhancer that can activate from a distance, we fused the Hes1 SPS onto the 5Ј end of complete or truncated hes5 promoter sequences. In all cases, increasing the distance diminishes the effectiveness of SPS (supplemental Fig. S3). This enhanced the overall activation of the hes5 promoter only in the presence of two proximal CBS. Most interestingly, although removal of either CBS lowered the activation by all paralogs, the more distant CBS-D can complement SPS activity less efficiently than the proximal CBS-E (compare construct AB.(Ϫ248 to ϩ1)Dm to AB.(Ϫ248 to ϩ1)Em and AB.Hes5Dm to AB.Hes5Em in supplemental Fig. S3). The addition of SPS improved activation by Notch4 such that it now surpassed Notch3. Thus, the SPS is not a potent enhancer, and from a distance it works in collaboration with an additional promoter proximal to CBS.

Enhanced Promoter Repression in the Absence of NICD Did Not
Contribute to SPS-mediated Increase in Activity-A significant change in fold activation was detected with both hes1 and hes5 promoters when we manipulated the distance and orientation of CBS (Figs. 2 and 3), but CBS sites can deliver both activation and repression (reviewed in Ref. 54). Because we measure fold activation over base line, our results could reflect improved repression or de-repression in the absence of NICD. To address the possibility that de-repression or enhanced repression contributed to our observations, we compared the basal activity of all constructs tested in Figs. 2 and 3. NIH3T3 cells were transfected with individual reporters with a transfection control in the absence of NICD. The level of luciferase activity was normalized against mock transfection with control LacZ plasmid only (Fig. 4, A and B). On the hes1 promoter, mutation of either site A or B caused roughly a 2.5-fold increase in basal activity ( Fig. 4A; n ϭ 5), whereas alteration of site C did not lead to a significant change (Cm ϭ 0.86). Construct C-HA (site B mutated and C converted to high affinity) showed lower basal activity than Bm but was not identical to repression mediated by SPS, suggesting that the paired high affinity sites mediated not only improved activation but FIGURE 5. Contribution of discrete domains to overall transcription was promoter-dependent. A, diagram of the hybrid Notch constructs used in the analyses. B-E, the influence of discrete Notch domain on the overall fold activation was promoter-dependent. Notch3 Ram-Ank domain exhibits repressive effect (B), and its C-terminal region is highly potent on hes5 promoter (D). There is an increase dependence on the Ram-Ank domain when the promoter contains an inverted paired CBS (E). In each experiment, the activity used for comparison was extrapolated by transfecting four different DNA concentrations of each NICD/variant that encompassed the optimal responses (see "Experimental Procedures"). Results are representative of at least three independent experiments. F, Western blot with anti-FLAG antibody of the lysate used in the hes1 promoter luciferase assay (arrowhead indicates nonspecific band). IB, immunoblot. FEBRUARY 24, 2006 • VOLUME 281 • NUMBER 8 also improved repression. Switching the orientation of site B lowered repression (invB ϭ 1.74, p ϭ 0.003, n ϭ 6), indicating that repression is also dependent on orientation.

Target Selectivity of Vertebrate Notch Proteins
Another conclusion that can be reached is that the contribution of individual CBS to repression does not always mirror its role in mediating Notch activation; on the hes5 promoter, mutation to either site D or E showed insignificant changes in the basal activity (Fig. 4B, Dm and Em). As predicted, engineering a paired CBS (in three different contexts and orientation: ⌬(Ϫ214 to Ϫ79), D16, and InvD16) decreased basal activity to the same extent as removing the 5Ј region (⌬ (Ϫ214 to Ϫ79)). This observation suggests that repression by a paired site acts to negate enhancer elements located in the deleted distal fragment; these same elements cooperate with NICD when it is present on site E (Fig. 3C). These results confirmed that paired sites could act as effective repressive elements in the absence of Notch. More importantly, the lower basal activity resulting from repression was not sufficient to explain the vast improvement in reporter activation when SPS was introduced.
NICD3 Contains a Potent Activator Domain-The experiments described above highlight the dual role of SPS in regulating gene expression; SPS acts as either an effective repressor or enhancer element, the latter depending on the availability of NICD protein or its mimics (EBNA2 (55,56) and RTA (57)). Our promoter analyses also revealed that Notch3 is particularly inept in capitalizing on SPS. Next, we wished to examine the contribution of different domains within each Notch paralog to the activation of target promoters. Earlier studies have suggested that the ANK domain was important, as was the C-terminal TAD (12,13). However, domain boundary was not clarified until recently (5,43); we therefore re-tested the roles of the ANK and TAD in target activation. To perform these domain-swap experiments, we preserved the seventh ANK repeat in each of the constructs made; details of domain boundary are shown in Fig. 5A.
By using optimal concentrations for each paralog (see "Experimental Procedures"), we confirmed the importance of the Notch1 TAD as the primary determinant for overall activation of the 4xCBS promoter (Fig.  5B, compare N1 with Notch4 RAM-ANK fused to Notch1 C-terminal region (N4A1)). However, comparison of the other fusion constructs on 4xCBS indicates that the ANK repeats also contribute to activation fold; despite the presence of a strong N1TAD, N3A1 is a poor activator of 4xCBS, suggesting that Notch3 RAM-ANK repeats are a dominant repressing factor in determining activation strength (Fig. 5B). In addition, the minimal hes1 promoter (ABtata) resembles the intact hes1 promoter in its response to activators (Fig. 5E), whereas the 4xCBS promoter is distinct in its response. This suggests that RAM-ANK repeats are capable of interpreting specific arrangement of CBS on a target. Analyses of Hes1 and Hes5 further confirm that the RAM-ANK repeats are critical on natural promoters (Fig. 5, C and D) (13).
Most interestingly, the construct N1A3 is more active on the hes5 promoter (n Ͼ10) than any natural Notch protein. Therefore, the ability of Notch3 to activate Hes5 better than Notch4 maps to its C-terminal region, which defines it as a TAD.
The Notch-responsive Enhancer of Hes5 Includes a Proximal CBS and Putative Zinc Finger Transcription Factor/CUP-binding Sites That Are Conserved across Species-Although SPS forms an optimal Notch-responsive element, they are not found in the regulatory regions of many Notch target genes (for example, Hes5; Vg (58); cyclin D1 (59); p21 (60); Nrarp (61); Hey (40); and Nodal (62)). Notch3-mediated activation of Hes5 is dependent critically on elements 5Ј to CBS-E in a region between Ϫ248 and ϩ1 (Fig. 3). Moreover, the spacing between the cis-regulatory element(s) and CBS-E might be critical for synergistic interaction as it is in SPS; interaction with these elements was lost after the insertion of CBS-D (Fig. 3C, D16).
With the hyperactive N1A3 fusion protein as a sensitive reporter, we systematically analyzed the region between Ϫ248 and ϩ1 for the presence of this putative site. We focused on TF sites that are present in Hes5 but absent in Hes1 (Fig. 6A). Our deletion analyses revealed that sequences located between positions Ϫ98 and Ϫ80 on the hes5 promoter were critical for N1A3 activity (Fig. 6B). Within this 18-base region, the putative zinc finger transcription factor-binding site (ZFBS; GCCGCCATT) (63, 64) was disrupted in earlier experiments when CBS-D was inserted near CBS-E (Fig. 3C, D16). A binding site for CCAAT/enhancer-binding protein ␣ (C/EBP␣) undifferentiated protein (CUP) (CCGCCG) (65) was located between the ZFBS and CBS-E (Fig. 7A). All three sites occupied 27 bases within the hes5 promoter. To address the possibility whether either one or both of these putative sites were critical for Notch activity on Hes5, we mutated the sites separately on the hes5 promoter, and we tested the effect on promoter activation at optimal Notch DNA concentrations (Fig. 7).
We speculated above that the spacing between CBS and ZFBS might be critical. To test this, we replaced the sequence "CCGCC" located between ZFBS and CBS-E with a single "A." The only possible consensus TF-binding site generated by this replacement (the complement TCNCNCNCCC resembles Yi-binding site (66)) was also generated earlier with no affect on Notch activation (Fig. 7C, CUPm). This deletion resulted in significant reduction in reporter activation by all paralogs (Fig. 7D), consistent with the hypothesis that Notch may interact simultaneously with a zinc finger protein bound to ZFBS and the single RBP-j molecule bound to the CBS-E, and that the Notch3 TAD has evolved to excel in mediating this interaction. Alternatively, CUP and the zinc finger proteins are both required in the correct spacing to form a complex that the Notch3 TAD can recognize. The Hes5 core enhancer is analogous to the SPS, where the integrity of the 27 bases encompassing these two sites is critical for Notch activation of Hes1. Further biochemical experiments, outside the scope of this study, are required to identify the zinc finger TF and confirm its synergistic interaction with the Notch3 protein.
Sequence analysis revealed that the Hes5 core enhancer is completely conserved in human, mouse, and rat but showed moderate to low homology in chicken and Zebrafish (supplemental Fig. S4A). To test if this regulatory module may be present in the promoter region of other genes, we performed in silico analyses of 1000-bp upstream sequences from all annotated human genes. We identified several putative targets that fit our search criteria (supplemental Fig. S4C). One of the putative targets, Hes4, is related to Hes5. Cross-species analysis of the hes4 promoter showed moderate to low conservation in the ZFBS-CBS module (supplemental Fig. S4D). Experiments establishing these genes as Notch3-specific targets are underway.
CBS Architecture Affects Expression Kinetics in Vivo, Defining between Primary and Secondary Notch Targets-Our analysis indicates that the CBS architecture within the promoter context is critical in defining the response to Notch paralogs based on their unique module (ANK/TAD) combination. It is possible that the in vitro data reflect target utilization in vivo. Proximal SPS may facilitate recognition of RBP-j by Notch and thus support expression in the presence of lower levels of Notch or its mimics (Fig. 2B). Targets with unpaired CBS may require higher nuclear concentrations. Previously, we demonstrated that Hes1 activation occurred before NICD1 could be detected in the nucleus (67). The analog or pseudo-analog behavior of hes1 and hes5 may thus reflect the amounts of NICD required to recognize and activate these promoters.
To investigate the merit of this hypothesis in vivo, we used a pharmacological approach to reduce NICD concentration and to examine the response of chromatin-embedded promoters in a living tissue, the kidney anlagen (metanephroi) from Hes1-GFP or Hes5-GFP transgenic mice (68). Histological examination showed that Hes5-GFP is expressed where active Notch1 is detected (47); however, Hes1-GFP showed a broader expression domain within the S-shaped body in the developing kidney, perhaps because it could respond to lower levels of Notch signal FIGURE 7. Integrity of 27 bases containing putative ZF and CUP-binding sites was critical for Hes5 activation by Notch. A, comparison of the sequences of the different mutant variants of hes5 promoter (Ϫ107 to Ϫ69 region) used in the analyses. B, mutation in putative zinc finger site significantly affects the activity of Notch3 and N1A3 proteins. C and D, although mutation to CUP site did not affect Notch activity, its removal drastically impaired Notch activation of hes5 promoter (optimal activation of different Notch proteins was extrapolated from titration experiment (see "Experimental Procedures"). Notch1 activity was used as the reference value and set as 100%. Results are represented as mean Ϯ S.D. Statistical significance was defined as p value of 0.05 or less. FEBRUARY 24, 2006 • VOLUME 281 • NUMBER 8 (Fig. 8, A and B, left panel). Our hypothesis predicted that the hes1 promoter would be more resistant to the ␥-secretase inhibitor, DAPT, which will eventually block Notch proteolysis and thus signaling, than Hes5. Similar expression kinetics between these reporters would mean that the hes1 and hes5 promoters are not different in their affinity to Notch-CSL complex; the involvement of other transcription factors would predict complete resistance to DAPT.

Target Selectivity of Vertebrate Notch Proteins
Pilot experiments established that 8 h after DAPT treatment, NICD1 levels drop; NICD1 became undetectable at 16 h (data not shown and see Ref. 47). We have shown that if DAPT is removed after 48 h of inhibition, NICD is restored, and normal metanephroi development resumes (47). We then compared the kinetics of GFP disappearance and reappearance after DAPT withdrawal in metanephroi from the two transgenic lines. After 8 h of DAPT treatment, Hes1-GFP expression was unchanged (Fig. 8A, panels on the right) whereas Hes5-GFP was extinguished (Fig. 8B, panels on the right). The Hes1-GFP signal was eventually lost at the 18-h time point. DAPT was removed 24 h after the experiment started, and the Hes1-GFP signal regained normal intensity 6 h later (Fig. 8A, right panel). In contrast, and consistent with our hypothesis, Hes5-GFP intensity was only partially restored after 21 h (Fig. 8B, right panel).

DISCUSSION
Many envisioned mammalian Notch paralogs as transcription factors activating specific subsets of downstream targets (29 -32, 34, 69). Although an attractive proposition, the mechanistic basis for such selectivity is not easily understood because all bind to promoters via an interaction with the same protein, RBP-j. Given the revised assignment of the ANK domain boundaries (5), our objective was to readdress the question of divergence between mammalian Notch paralogs and to begin to understand the mechanism through which it may have been achieved.
Promoter Activation by the Four Mammalian Notch Paralogs Is Largely Cell Type-independent-Promoter activation assays in transfected cells measures the response of up to ϳ10 6 targets in each cell. Different activation probabilities underlie the observed differences in activation "strength" of each Notch paralog, resulting in a pseudo-analog enhancement of transcription; Notch1 provides the highest activation probability, whereas Notch3 provides the lowest probability on SPS at an optimal concentration (supplemental Fig. S2). Our analyses demonstrated that the potency of the four vertebrate Notch paralogs in a transient transfection assay is dependent on the promoter, the amino acid composition of the individual Notch protein, and to a lesser degree on a cellular context (Fig. 1). We hypothesized that this reflects dependence on Notch concentration; the Hes1 SPS is activated at lower NICD concentrations than the hes5 promoter. We demonstrate that this differential response to Notch is seen in vivo, where Hes1 has a broader expression domain and is more resistant to reduction in Notch levels than Hes5 in the same organ (Fig. 8).
Collectively, our data suggest that a matrix of parameters that include NICD concentration (a reflection of receptor and ligand concentration), amino acid composition (specific paralog composition in the cell proteome), CBS orientation, and promoter context (48,70) will determine which promoters respond to a particular Notch protein at a given level of ligand binding. Some of these ideas will have to be tested by structural and biophysical approaches; evidence for the importance of others are discussed below.
Orientation and Context-Why then is the orientation so important? The ability of SPS to improve repression of promoters (Fig. 4B) would argue that binding to SPS does not require pre-assembly of the RBP-j-NICD complex. Correctly spaced and oriented sites may increase the efficiency with which both repressing and activating complexes assemble on the promoter. A higher probability of activation could result from Notch binding to RBP-j cooperatively when two RBP-j molecules are properly positioned in proximity to each other. However, our EMSA data do not support this interpretation. In both H-H (WT) and H-T (InvB) orientations, RBP-j displays noncooperative binding; most probes bound only one RBP-j (complex 1), and a double-bounded probe can be faintly seen between complex 2 and 4 (Fig. 2C). Truncated NICD1 forms two identical complexes (complexes 3 and 4), both smaller than the probe bound by two RBP-j molecules on either H-H or H-T probes, suggesting perhaps that two NICD molecules can interact with each RBP-j molecule in this assay. Therefore, head-to-tail CBS arrangement abolished Notch activation without affecting formation of RBP-j-NICD complexes and without any obvious enhancement for a given Notch/RBP stoichiometry (2:1 or 1:1). Instead, it is possible that Notch proteins differ in their ability to assemble the mediator complex (8), dissociate the repressor complex, or both. We observed that Notch3 activated the hes5 promoter better than Notch4 in all cell types, suggesting that perhaps the Hes5 represented a class of preferred Notch3 targets; Hes5 is activated in vascular smooth muscle cell where Notch3 plays an essential role (23,71). To understand this preference, we dissected the promoter (Fig. 6) and discovered that synergism between a putative ZFBS and a CBS within a 27-bp span on the hes5 promoter formed a preferred Notch3 target and that the Notch3 TAD was sufficient to confer efficient utilization of this promoter, perhaps by interacting to the neighbor (Fig. 7).
At this time, we do not know which proteins facilitate Notch3 activity on the Hes5 core enhancer. However, the information we gathered permitted us to address some possibilities. The 27-bp Notch-responsive enhancer of Hes5 includes binding sites to a zinc finger protein, CUP, and RBP-j (CBS-E; all are underlined). The site distribution may favor interaction between Notch and the zinc finger TF: RBP-j occupies 11 bp (72) and its footprint (italic) may prevent simultaneous binding of CUP (lowercase) within the enhancer sequence GCGCCGCCATTG-GccgccgAG TGTGGGAACG. Genome-wide search of conserved ZF-CBS regulatory modules in upstream sequence regions in multiple species reveals a few interesting putative targets (supplemental Fig. S4), some of which are reported to act in vascular smooth muscle cells, a specific tissue where Notch3 function is important. We are currently working to validate these candidate genes in Notch3 null animals.
While this manuscript was in preparation, in vitro and in vivo evidence emerged that an intact "SPS ϩ A" (proneural A site) transcriptional code functions to mediate Notch transcription of the m8 gene in specific Drosophila cells (44). Alteration to the specific head-to-head architecture of SPS abolished this synergy. The authors speculated that other configurations of the Su(H)-binding site could provide for cofactor(s) specificity, resulting in selective Notch target gene activation in the fly. Candidates for such paralog-selective interactions include SMAD3 (73)(74)(75), bHLH activator (44,50,52,76), and grainyhead (48), the latter two contribute to Drosophila E(spl)-C expression. To this list we add a yet to be identified zinc finger protein. Our observations establish that activation by vertebrate Notch differs from Drosophila where the presence of additional E box elements between the SPS and TATA is necessary (44). First, we uncover no evidence for the involvement of a bHLH in regulation of Hes1 (Fig. 2B, ABtata) (49). Second, and more importantly, tail-to-tail sites can confer significant levels of Notch1 activation in the right context (Fig. 3C, ⌬(Ϫ214 to Ϫ79)). These observations concur with previous studies that define the H-H orientation as optimal Notch target; all other orientations require contributions from additional CIS elements.
Distance-We find that placing SPS further from the promoter start site diminishes their usefulness in enhancing Notch-mediated activation; however, if an additional promoter proximal CBS was present, SPS still provided potent enhancement (supplemental Fig. S3). Cooperativity with a proximal site decreases with distance. Most interestingly, we found no evidence showing synergism between low affinity CBS and a distant SPS, suggesting these sites may not mediate CSL-dependent activation.
The following two mechanisms for RBP-j-mediated suppression were proposed: lock-and-key (77), in which RBP-j interferes with access to the promoter, and general repression by the recruitment of repressor complexes (78). The first requires placement between the TATA box and the repressed enhancer, and the second can work over distance. Conversion of CBS-E to a paired-CBS inhibited the basal activity of the hes5 promoter, consistent with either mechanism. Distally located paired CBS were not as potent in reducing basal activity (data not shown), a surprising result suggesting that despite the documented ability of CSL to recruit the SIN3A repressor complex, it acts to suppress Hes5 in accord with a lock-and-key mechanism.
Amino Acid Composition Integrates Variability Within Enhancer Elements-The pictures that emerge from the literature and from the data provided here support the hypothesis where each mammalian Notch paralog evolved to utilize different promoters based on spacing, distribution, orientation and distance of CBS within a promoter, and the presence of specific cis-elements recognized by other factors.
Mammalian Notch paralogs can be separated in two functional modules that translate specific cis-regulatory elements into transcriptional output (12,13). The four Notch paralogs utilize modular protein domains to integrate different configurations of CBS with other transcription factors binding in their vicinity. Promoters lacking SPS will be expected to show the greatest divergence in response to Notch paralogs. Optimal activation is achieved via integration of ANK and TAD domains with a specific enhancer sequence. Perhaps Notch3, whose C-terminal region has no strong activation capability in heterologous promoter assays (13) (Fig. 5B), evolved to favor promoters with zinc finger protein-binding sites near a CBS. We demonstrate this by creating a fusion protein containing the effective Notch1 ANK and the specialized Notch3 TAD (N1A3) to gain optimal activation of the Hes5 core enhancer (Figs. 5-7). As expected, the reciprocal protein (N3A1) was a poor activator of any promoter (Fig. 5) (13). FEBRUARY 24, 2006 • VOLUME 281 • NUMBER 8

JOURNAL OF BIOLOGICAL CHEMISTRY 5117
Concluding Remarks-The matrix described above integrates our results with published reports and is likely to underlie paralog-specific gene regulation in mammals. In the effort to construct the perfect "Notch-reporter" mouse, several promoters were chosen, including the three under analysis in this study (68,79,80). Our data suggest that Notch proteins activate a different chorus of targets in each relevant developmental or disease context rather than a quantum of signature targets. Therefore, reliance on any single reporter could be misleading because it is bound to miss important subtleties in gene regulation. Instead, multiple reporters, each validated to reflect accurately a subset of Notch activities, should be investigated by those interested in visualizing Notch pathway activity.