Identification of Function for CD44 Intracytoplasmic Domain (CD44-ICD)

Background: CD44, a multifunctional receptor, undergoes cleavage to produce an intracytoplasmic domain (CD44-ICD) that translocates into the nucleus. Results: CD44-ICD binds to a novel DNA consensus sequence and activates many genes. Conclusion: We finally explain the multifunctionality of CD44 and reveal new genes affected by CD44. Significance: Our findings will accelerate the understanding of how CD44-ICD regulates a multitude of cell functions. CD44 is a multifunctional cell receptor that conveys a cancer phenotype, regulates macrophage inflammatory gene expression and vascular gene activation in proatherogenic environments, and is also a marker of many cancer stem cells. CD44 undergoes sequential proteolytic cleavages that produce an intracytoplasmic domain called CD44-ICD. However, the role of CD44-ICD in cell function is unknown. We take a major step toward the elucidation of the CD44-ICD function by using a CD44-ICD-specific antibody, a modification of a ChIP assay to detect small molecules, and extensive computational analysis. We show that CD44-ICD translocates into the nucleus, where it then binds to a novel DNA consensus sequence in the promoter region of the MMP-9 gene to regulate its expression. We also show that the expression of many other genes that contain this novel response element in their promoters is up- or down-regulated by CD44-ICD. Furthermore, hypoxia-inducible factor-1α (Hif1α)-responsive genes also have the CD44-ICD consensus sequence and respond to CD44-ICD induction under normoxic conditions and therefore independent of Hif1α expression. Additionally, CD44-ICD early responsive genes encode for critical enzymes in the glycolytic pathway, revealing how CD44 could be a gatekeeper of the Warburg effect (aerobic glycolysis) in cancer cells and possibly cancer stem cells. The link of CD44 to metabolism is novel and opens a new area of research not previously considered, particularly in the study of obesity and cancer. In summary, our results finally give a function to the CD44-ICD and will accelerate the study of the regulation of many CD44-dependent genes.

CD44 is a multifunctional cell receptor that conveys a cancer phenotype, regulates macrophage inflammatory gene expression and vascular gene activation in proatherogenic environments, and is also a marker of many cancer stem cells. CD44 undergoes sequential proteolytic cleavages that produce an intracytoplasmic domain called CD44-ICD. However, the role of CD44-ICD in cell function is unknown. We take a major step toward the elucidation of the CD44-ICD function by using a CD44-ICD-specific antibody, a modification of a ChIP assay to detect small molecules, and extensive computational analysis. We show that CD44-ICD translocates into the nucleus, where it then binds to a novel DNA consensus sequence in the promoter region of the MMP-9 gene to regulate its expression. We also show that the expression of many other genes that contain this novel response element in their promoters is up-or down-regulated by CD44-ICD. Furthermore, hypoxia-inducible factor-1␣ (Hif1␣)-responsive genes also have the CD44-ICD consensus sequence and respond to CD44-ICD induction under normoxic conditions and therefore independent of Hif1␣ expression. Additionally, CD44-ICD early responsive genes encode for critical enzymes in the glycolytic pathway, revealing how CD44 could be a gatekeeper of the Warburg effect (aerobic glycolysis) in cancer cells and possibly cancer stem cells. The link of CD44 to metabolism is novel and opens a new area of research not previously considered, particularly in the study of obesity and cancer. In summary, our results finally give a function to the CD44-ICD and will accelerate the study of the regulation of many CD44-dependent genes.
CD44 is a multifunctional cell membrane receptor involved in cell adhesion, tumor invasion, and metastasis (1)(2)(3). In addition, it also regulates macrophage inflammatory gene expression (4) and vascular gene expression in proatherogenic environments (5). More recently, CD44 has been implicated in the multidrug resistance phenotype of some cancer cells and in protecting cells against apoptosis (6 -8). Importantly, CD44 has emerged as a marker of normal progenitor cells as well as cancer-initiating or stem cells, although its role in these cells is not known. CD44 exists in a variety of isoforms due to alternative splicing and post-transcriptional modifications. The most common isoform, CD44 standard (CD44s), 2 is a major receptor for hyaluronan (9,10). Also, CD44s has been implicated in the hyaluronan-dependent or -independent regulated expression of matrix metalloproteinases (MMPs), mainly MMP-2 and MMP-9 (11)(12)(13), and has been suggested to be involved in transcription (14). However, the mechanism of the CD44 multifunctionality is not known.
MMPs are a group of endopeptidases that degrade extracellular matrix. Hence, the enzymatic activities of MMPs play an important role in invasion and metastasis of tumors in which they are frequently overexpressed (15,16). Some MMPs, including MMP-9, are known to bind CD44 (17,18) or are reg-* This work was supported by National Institutes of Health Grants ulated by the CD44s-hyaluronan interaction (11,13). Hyaluronan also increases MMP-9 activity and gene expression (19). This effect is blocked with anti-CD44 antibodies, although a defined mechanism of how CD44 affects MMP-9 expression is not known.
In the current study, we utilized CD44-mediated overexpression of MMP-9 as a working platform to elucidate the role of CD44 in transcription. We used a CD44-ICD-specific antibody, a modification of a chromatin immunoprecipitation (ChIP) assay to detect small molecules, and extensive computational analysis. We found that CD44 induces MMP-9 transcription directly following the intramembranous proteolytic processing of CD44 by presenilin-1. We demonstrate that the resulting intracytoplasmic tail, CD44-ICD, is then transported into the nucleus, where it binds a novel promoter response element, thereby regulating transcription of target genes. Interestingly, the CD44-ICD response element (CIRE) is located downstream of Hif1␣ in CD44-ICD early response gene promoters; this group of hypoxic-responsive genes are turned on by CD44 during normoxic conditions independently of Hif1␣ expression. Taken together, these studies show that CD44-ICD activates multiple genes involved in cell survival during stress, atherogenesis, inflammation, oxidative glycolysis, and tumor invasion. These findings finally elucidate a mechanism for the many functions attributed to CD44, specifically cancer cell metastases and metabolism, and promise to accelerate the study of the regulation of many CD44-dependent genes.

EXPERIMENTAL PROCEDURES
Tissue Microarray (TMA)-Tissue arrays were prepared in collaboration with the Cancer Institute of New Jersey Tumor Retrieval Shared Facility and the Tissue Array Facility. Ovarian and breast carcinoma tissue arrays were constructed using formalin-fixed paraffin-embedded tissue blocks containing ovarian cancer tumors. Areas of invasive tumor and normal tissue were identified and marked for subsequent retrieval and analysis. Core biopsies of 0.6 mm in diameter were taken from each donor block and arrayed into a glass slide. The ovarian cancer TMA was constructed from patients who gave consent to have identifiable information; therefore, Institutional Review Board approval was obtained. The breast TMA was constructed without identifiable patient information and received Institutional Review Board-exempt approval. The TMAs were approved by Institutional Review Board 0220034452 and 020055381, respectively. Immunoreactivity was assigned as positive when more than 50% of the tumor cells stained for the particular antibody, regardless of intensity, or when focal strong staining was observed. At least 20% showed strong staining. Negative immunoreactivity was defined as no reactivity at all, weak staining, or staining of less than 5% of the tumor.
Enzyme-linked Immunosorbent Assay (ELISA)-Nuclear and cytoplasmic fractions from cell lines MCF-7/CD44-ICD-GFP and MCF-7/GFP vector were prepared using the Nuclear Extract Kit (Active Motif), and each fraction was analyzed by an activated Runx2-specific ELISA (TransAM, Active Motif) as suggested by the manufacturer.
Electrophoretic Mobility Assay (EMSA)-Oligonucleotides for mobility shifts were prepared by Integrated DNA Technologies. Nuclear extracts were prepared from MCF-7, MCF-7/ CD44s, and MCF-7 CD44-ICD-GFP monolayer cultures grown to 70% confluence. For nuclear extract harvest, cells were washed two times with ice-cold Ca 2ϩ /Mg 2ϩ -free PBS and transferred in PBS to microcentrifuge tubes, where they were centrifuged in a refrigerated centrifuge (4°C) at 1,850 ϫ g for 5 min. Cells were resuspended in 5 packed cell volumes of buffer A (10 mM HEPES, pH 7.4, 1.5 mM MgCl 2 , 10 mM KCl, 0.5 mM DTT, 0.1 mM EDTA, 0.1 mM EGTA, 1ϫ Halt Protease Inhibitor (Roche Applied Science), 0.3% IGEPAL) and centrifuged at 1,850 ϫ g for 5 min at 4°C. The pellet was resuspended in 3 packed cell volumes and lysed using a 25-gauge needle. The samples were centrifuged at 3,300 ϫ g for 5 min at 4°C and resuspended in buffer C (20 mM HEPES, pH 7.9, 7.5 mM MgCl 2 , 0.2 mM EDTA, 0.1 mM EGTA, 1.0 mM DTT, 0.4 M NaCl, Halt Protease Inhibitor (Roche Applied Science)). Cells were centrifuged in a refrigerated centrifuge (4°C) at 14,000 ϫ g for 30 min, and supernatants were stored in 20-ml volumes at Ϫ80°C. Protein content of the extracts was determined using the modified Lowry assay with a bovine serum albumin standard curve (Pierce). Ten g of nuclear protein combined with binding buffer (20 mM HEPES, pH 7.9, 25% glycerol, 1.2 mM MgCl 2 , 0.3 mM EDTA, 0.5 mM DTT, 0.5 mM PMSF, 0.42 M NaCl, Halt Protease Inhibitor (Roche Applied Science), and 1 mg of poly(dI-dC)) totaling 10 ml were incubated with 30,000 cpm of [␣-32 P]dCTP (GE Healthcare) double-stranded probes for 30 min at room temperature. After incubation, binding reactions were electrophoresed on a 7% acrylamide gel in 0.5ϫ Tris borate/EDTA at 125 V. Gels were dried, exposed on a Phosphor-Imager screen, and scanned using a PhosphorImager.
Immunofluorescence Microscopy-Cells were fixed with 4% paraformaldehyde in PBS, incubated with anti-CD44-ICD antibody and Alexa 488-and Cy3-conjugated secondary antibodies (Invitrogen), and counterstained with TOPRO-3 (Molecular Probes). Cells were viewed using the Nikon Eclipse TE2000U confocal microscope using the ϫ60 oil immersion Plan Apo objective, and images were acquired with EZ-C1 3.50 software. Image analysis was done using ImageJ software (National Institutes of Health).
Luciferase Assay-MCF-7 cells were transiently transfected with a firefly luciferase reporter vector (pGL3, Promega) driven by a region of the MMP-9 promoter (pGL3/Ϫ670MMP-9 construct; a kind gift of Dr. Ernst Lengyel, University of Chicago Medical Center) using Lipofectamine Reagent (Invitrogen) as suggested by the manufacturer. Twenty-four h after transfection, cells were detached and lysed, and luciferase activity was analyzed using the Dual-Glo Luciferase Assay System (Promega).
Microarray-Experiments were run for 28 samples on the Affymetrix Human Exon 1.0 ST exon microarray platform (1.4 million probes). Using GeneSpring GX 11 (Agilent Technologies, Inc.), raw exon expression signals were combined and summarized with ExonRMA16 (RMA) using only Core transcripts (17,800 transcript clusters from RefSeq and full-length GenBank TM mRNAs). The data were further quantile-normalized with base-line transformation by the median of all samples. Further, the normalized expression signals were averaged between biological replicates. Gene expression data were first filtered by percentile cut-off, resulting in removal of genes with low signal (Ͻ10 percentile of all expression values across all samples) in all samples. Gene lists displaying differentially expressed behavior were generated by performing pairwise comparisons. The 251 genes that have been shown to change in "tail" were acquired by looking for a significant -fold change (Ͼ2.0) between any pairs of the tail time points (tail 12 h versus tail 24 h, tail 24 h versus tail 48 h, and tail 12 h versus tail 48 h). To obtain the 64 genes that change not only in tail, but in "full" as well, the 251 genes mentioned previously were further filtered by -fold difference (Ͼ2.0) between pairs of the full time points (full 12 h versus full 24 h, full 24 h versus full 48 h, and full 12 h versus full 48 h). Fig. 6 depicts a heat map of the expression levels of the 64 genes of interest, highlighting the genes most significantly differentially expressed in CD44-ICD and fulllength CD44s using -fold change as the criterion. The heat map was generated using hierarchical conditions, and gene clustering was performed using the Euclidean distance metric and centroid linkage rule.
Statistical Analyses-Microsoft Excel software was used for statistical analysis. In all studies, comparison of mean values was conducted with two-tailed two-sample equal variance Student's t tests. In all analyses, statistical significance was determined at p Ͻ 0.05.

RESULTS
CD44s Induces MMP-9 Expression and Activity in Carcinoma Cells-We previously reported an increase in migration and in vitro invasion of MCF-7 breast carcinoma cells stably transfected with CD44s (MCF-7/CD44s) as well as the CD44dependent transcriptional regulation of other genes (7). MMP-9 has been reported to be regulated by CD44s-ligand interactions (11) and is implicated in cancer invasion. These findings led us to investigate how CD44 regulates the expression of MMPs in cancer cells.
MMP-9 protein expression and activity was evaluated in MCF-7 cell lines expressing different levels of endogenous or ectopic CD44 using Western blot analysis. Fig. 1, A and B, shows increased levels of active MMP-9 in MCF-7/Adr and MCF-7/CD44s cells compared with parental MCF-7 or MCF-7 empty vector cells. To validate these results gelatin zymogram analyses were performed. MMP-9 activity appeared in the cells that express CD44 while it was barely visible in parental MCF-7 cells and empty vector cells (Fig. 1C). Interestingly, MMP-2 activity remained relatively constant regardless of the CD44s expression (Fig. 1C). This observation indicates a CD44-specific effect on MMP-9 activity. Of note, the MMP-2 gene promoter is generally regarded as constitutively regulated, whereas the MMP-9 gene has a more inducible promoter (20). We also tested whether endogenously induced CD44 induced MMP-9 expression. We previously showed that TNF-␣ up-regulates endogenous CD44 expression in SKOV-3 ovarian cancer cells (21). Fig. 1, D and E, shows a correlation between endogenous CD44s protein expression and MMP-9 activity after TNF-␣ treatment. In this case, both of them were up-regulated in a dose-dependent manner. The modulation of MMP-9 in the breast and ovarian cancer cell lines analyzed suggests that CD44s has a direct role in the up-regulation of the invasive phenotype of these cancer cells through the functional expression of MMP-9.
CD44-ICD Localizes to Nucleus to Regulate MMP-9 Expression-CD44 undergoes regulated intramembranous proteolysis (22) by presenilin-1/␥-secretase, which generates a cytoplasmic tail known as CD44-ICD and is translocated to the nucleus. However, the function of nuclear CD44-ICD is not known. We hypothesized that nuclear CD44-ICD could be responsible for the transcriptional up-regulation of MMP-9 expression. To test this hypothesis, we transiently transfected MCF-7 cells with one of the following plasmids: pcDNA3.1 vector (vector), pCDNA/CD44 WT full-length (CD44s), or pCDNA/CD44-ICD (CD44-ICD) and measured MMP-9 mRNA expression by qRT-PCR. Fig. 1F shows that CD44s or CD44-ICD induced a 3.5-fold and a 14-fold increase in MMP-9 mRNA over empty vector, respectively. To confirm that CD44-ICD indeed translocates into the nucleus, we GFPtagged CD44-ICD. As shown in Fig. 1G, GFP-tagged CD44-ICD was found in the cytoplasm as well as in the nucleus of stably transfected MCF-7 cells. To further confirm the nuclear localization of the CD44-ICD, nuclear and cytoplasmic extracts from MCF-7/ empty HA vector and MCF-7/CD44-ICD-HA cells were analyzed by Western blotting. Fig. 1H illustrates that CD44-ICD-HA is present both in the cytoplasm and nuclear fractions of the cell extract.
In order to determine whether CD44 processing from fulllength into CD44-ICD occurs constitutively (i.e. without exogenous stimulation) in cancer cells, we examined the nuclear and cytoplasmic expression of the CD44-ICD by immunoprecipitation of CD44-overexpressing cell lysate with an anti-CD44-ICD antibody. Endogenously processed CD44-ICD was detected both in the cytoplasm and in the nucleus (Fig. 1I).
MMP-9 Regulation by CD44 Is Clinically Relevant in Human Tumors-We determined the clinical relevance of our findings by studying the co-expression of CD44 and MMP-9 in human ovarian and breast carcinomas. Tissues were obtained from women who underwent surgery to resect an ovarian or a breast cancer. We constructed a TMA and performed immunohistochemistry using anti-CD44 or anti-MMP-9 monoclonal antibodies ( Table 1). The breast carcinoma microarray was composed of 159 invasive breast adenocaricinomas (Table 1). We observed that 81% of the breast tumors were positive for CD44 expression. Fifty-five percent of the tumors co-expressed CD44 and MMP-9. As expected, only 8% of tumors expressed MMP-9 without CD44 expression. Because the breast TMA was biased by large tumors, we also tested an ovarian TMA for MMP-9 and CD44 co-expression (Table 1). Of 81 ovarian tumors, 73% were CD44 (ϩ), 59% co-expressed CD44 and MMP-9, and only 11% expressed MMP-9 without immunoreactivity to CD44. These results validate and strengthen the results of the in vitro data indicating CD44 as a regulator of MMP-9 expression in human cancers.
Processing of Full-length CD44 Is Required for Induction of MMP-9-To further evaluate whether the processing of CD44 into CD44-ICD was necessary for the transcriptional modulation of MMP-9, we co-transfected MCF-7 cells with CD44-expressing vectors and a luciferase reporter plasmid driven by the MMP-9 promoter ( Fig. 2A). Fig. 2B shows that when CD44s or CD44-ICD was overexpressed, the MMP-9 promoter-driven luminescence was increased by 10-and 11-fold, respectively. However, when CD44 could not be cleaved to produce CD44-ICD (CD44 uncleavable mutant (CD44-U) construct) luciferase expression was similar to empty vector. These data indicate that the CD44-ICD is capable of modulating the expression of MMP-9 at the level of its promoter and that the uncleaved CD44 is not capable of this induction. Immunofluorescence data of the co-transfected MCF-7 cells validated the expected cellular localization of the CD44 and the CD44-ICD. As shown in Fig. 2C, transfection with WT full-length CD44-expressing vector generates CD44 associated with the cell membrane as well as the cytoplasm and the nucleus. Transfection of CD44-ICD-expressing vector generates CD44-ICD associated with the nucleus but not the cell membrane. Transfection of cleavage mutant full-length CD44 generates CD44 mainly associated with the cell membrane but not the nucleus.
Because presenilin-1 is the enzymatic component of the ␥-secretase complex (23), a specific alteration of its expression

CD44-ICD Regulates MMP-9 via CIRE
should affect the CD44-dependent MMP-9 regulation. To test this possibility, we knocked down the presenilin-1 mRNA transcript and examined the effect of presenillin down-regulation on MMP-9 activity. We observed a decreased amount of MMP-9 activity (Fig. 2D). The down-regulation of MMP-9 activity correlated with the down-regulation of presenilin-1 expression tested by qRT-PCR (Fig. 2D, bar graph) and Western blotting. These data confirm a regulatory role of CD44-ICD in the up-regulation of MMP-9 expression.
CD44-ICD Interacts with Novel DNA Binding Sequence on MMP-9 Promoter-Because Runx2 is a known MMP-9 key transcription factor (24), we initially explored the possibility that CD44-ICD physically interacts with Runx2 at the MMP-9 promoter by analyzing ChIP nuclear extracts. Fig. 3A (lanes 7  and 8) depicts that nuclear Runx2 co-immunoprecipitated with nuclear CD44-ICD.
Furthermore, CD44-ICD was present in actively transcribed chromatin (H3K4(ϩ)) but not in silenced chromatin (H3K9(ϩ)) ( Fig. 3A, lanes 3-6). To confirm these findings we performed ChIP assays by amplifying the region containing the Runx2 consensus sequence of the MMP-9 promoter (Fig. 3B). We found that CD44-ICD-GFP immunoprecipitated a region of the MMP-9 promoter (Fig. 3B, lane 4). Immunoprecipitated Runx2 was also bound to this specific region of DNA (Fig. 3B, lane 5). These results support an interaction between CD44-ICD, Runx2, and the MMP-9 promoter within the same promoter region where Runx2 binds to DNA. To analyze whether this interaction was also present in cells expressing full-length CD44 and thus containing endogenously processed CD44-ICD, we performed a Runx2 consensus sequence-specific ELISA assay. We found that the binding of Runx2 to its consensus sequence was increased by almost 2-fold in nuclear extracts of cells endogenously producing CD44 (Fig. 3C). To test whether the CD44-ICD could directly bind to the DNA of the same region of the MMP-9 promoter, we performed a similar ELISA using recombinant GST-tagged CD44-ICD and the same Runx2 consensus sequence probe. Notably, CD44-ICD-GST was found to bind to the Runx2 consensus sequence in a dose-dependent fashion (supplemental Fig. S1A).
To further determine additional details of this interaction, we performed EMSAs. We incubated recombinant CD44-ICD-GST with a radiolabeled Runx2 consensus sequence probe. Fig. 3D shows a dose-dependent interaction between To better understand the interaction between endogenously processed CD44-ICD and the MMP-9 promoter, we performed additional EMSAs. Preliminary EMSAs indicated a CD44-related interaction with the MMP-9 probe because nuclear extracts from MCF-7/CD44s cells but not from parental MCF-7 cells produced shifted bands (supplemental Fig. S1B). Fig. 3E (lane 2) shows a shift of two bands upon combining CD44(ϩ) nuclear extracts with the WT MMP-9 promoter probe. We competed this binding with a "cold" MMP-9 WT promoter probe. Fig. 3E (lane 3) shows that, as expected, cold WT probe competed both complexes. We concluded that one of the competed bands was Runx2 bound to its consensus sequence (TGCGGT; underlined in Fig. 3F). To test whether the other band was bound CD44-ICD, we carried out competition experiments using mutated oligonucleotides within the Runx2 or neighboring sequences. In Fig. 3E, lane 4, we observed that when mutant 1 was used, there was still some competition when compared with the WT competitor. Similarly, when the whole Runx2 consensus sequence was mutated and used as competitor (Fig. 3E, lane 5), we found reduced competition. If Runx2 was the only protein binding to the DNA, adding a completely mutated Runx2 consensus binding site cold oligonucleotide would have resulted in no competition at all. Because we observed partial competition (compare lane 5 with lane 2), FIGURE 3. CD44-ICD interacts with Runx2 on the MMP-9 promoter. A, antibodies used in ChIP experiment were as labeled. The chromatin complex was cross-linked using ethylene glycol-bis (succinic acid N-hydroxy succinimide ester) followed by formaldehyde, (ϩ crosslink). Half of the sample was treated with hydroxylamine HCl to revert the cross-link (Ϫ crosslink). The above ChIP assay extracts were probed with anti-GFP antibody, which shows an interaction between CD44-ICD-GFP tagged and Runx2, which localizes to the transcriptionally active MMP-9 promoter in MCF-7/CD44s cells. B, we stably transfected MCF-7 cells with a GFP-tagged CD44-ICD expression vector and amplified 300 bp of the MMP-9 promoter in ChIP assays. This ChIP assay consisted of a two-step cross-linking method that improves the detection of small molecules by making use of a protein-protein cross-linker, Di(N-succinimidyl)gluterate, followed by a protein-DNA cross-linking step using formaldehyde (25). C, Runx2 consensus sequence-specific ELISA. Binding to Runx2 consensus sequence was observed on nuclear extracts from MCF-7 cells overexpressing CD44s. MCF-7 and MCF-7 vector-only cells were used as controls. Positive control was extracts from SAOS-2 cells; negative control was buffer only. D, EMSA shows specific CD44-ICD binding to a Runx2 binding sequence region of the MMP-9 promoter. The addition of salmon sperm DNA demonstrates the specificity of this binding interaction. E, EMSA analysis using a 32 P-labeled MMP-9 WT promoter oligonucleotide containing the Runx2 consensus sequence shows two shifting bands. These bands correspond to the binding of Runx2 to its known binding sequence (TGCGGT) and to the binding of CD44-ICD to a novel binding sequence (CCTGCG). F, DNA sequences of the radioactive MMP-9 WT promoter probe and the competing cold WT and mutant probes used. Runx2 binding sequence is underlined, and mutated nucleotides are shown in red. G, EMSA shows supershifted bands upon the addition of an anti-Runx2 antibody to confirm the presence of Runx2 in the binding complex. H, luciferase assay using pGL3-670MMP-9 reporter construct WT or without CCTGCG in its sequence (del mut). I, secreted luciferase assays show increased activity over control in supernatant from cells co-transfected with a CD44-ICD expression vector and either a WT CCTGCG sequence repeat-driven reporter vector (wt sr) or a mutant sequence repeat AAGTAT (mut sr). Reporter vector without the CCTGCG sequence was used as control. Error bars, S.D.
we hypothesized that the CD44-ICD is also binding outside the Runx2 consensus sequence. To test this premise, four additional mutant probes with sequence changes outside the Runx2 binding site were used (Fig. 3E, lanes 6 -9). These mutant probes fully competed the shift of the top band but not the lower band. Because the only sequence in common among the mutant probes and the 32 P-labeled DNA promoter was the WT Runx2 consensus sequence, we conclude that the top band reflects Runx2 binding to its consensus sequence. However, only mutant probe 5 was able to compete the bottom band as well (Fig. 3E, lane 8). The difference between mutant probe 5 and the other mutant probes is that this probe contained WT sequences immediately flanking the Runx2 consensus sequence. Fig. 3F shows that the WT MMP-9 promoter has a CCTGCG sequence that flanks the 3Ј-end Runx2 consensus sequence and overlaps the 5Ј-end by four nucleotides (i.e. TGCG). We hypothesized that these flanking CCTGCG repeats were the CD44-ICD binding sites.
To test this possibility, we mutated both flanking sequences but not the Runx2 consensus sequence. Fig. 3E, lane 9, shows that mutant probe 6 failed to compete the binding of the lower band but, as expected, completely competed the binding of the top band. These results indicate that CD44-ICD binds to CCTGCG. To further confirm this premise, we designed competing probes with multiple sequence repeats (SR) of the putative CD44-ICD response element (CD44-ICD SR) or the Runx2 consensus sequence (Runx2 SR). Fig. 3E, lanes 10 and 11, shows that probes Runx2 SR and CD44-ICD SR competed the top and the bottom band, respectively.
To validate the identity of Runx2, we performed a supershift assay using an anti-Runx2 antibody (Fig. 3G, lane 3). We observed that the anti-Runx2 antibody shifted the radiolabeled probe. Furthermore, we competed this supershift using the cold Runx2 SR competitor probe in a dose-dependent fashion (supplemental Fig. S2). We also competed the CD44-ICD band (lower band) but not the Runx2 band (top band) with a CD44-ICD SR probe (supplemental Fig. S2). This result indicates that the bottom band in fact corresponds to the MMP-9 probe interaction with CD44-ICD.
In order to confirm that CCTGCG was a novel CD44-ICD response element sequence, we performed two different luciferase reporter expression assays, one with the same luciferase construct used in Fig. 2B (pGL3/Ϫ670MMP-9 promoter) and the other with a minimal promoter sequence plasmid (pTK) in which the putative CD44-ICD response element (CCTGCG) was cloned upstream of the minimal promoter. Fig. 3H shows that when CD44-ICD was co-transfected with a CCTGCG deletion mutant in the Ϫ670MMP-9 promoter luciferase reporter construct, luciferase activity decreased by 56% compared with the WT reporter construct. In a similar experiment but using the pTK/CCTGCG luciferase reporter construct, Fig.  3I shows that when CD44-ICD was co-transfected with a CCTGCG sequence-driven luciferase reporter construct, luciferase activity increased by 1.8-fold (3.3/1.8) over empty vector (p Ͻ 0.01). Consistent with this result, the mutation of the CCTGCG sequence to AAGTAT resulted in a decrease in luciferase activity back to vector-only levels. These experiments confirm CCTGCG as the CD44-ICD DNA response element.
CD44-ICD Response Element Is Present in CD44-regulated Genes-To identify additional genes that could be regulated by CD44s via CD44-ICD, we performed an NCBI PubMed Web site literature review and found 12 genes that had been reported as regulated by CD44. Using the Web-based program Transcriptional Regulatory Element Database, we found that all of the CD44-regulated genes found in our search contained one or more repeats of CCTGCG, the CIRE, in their promoters (Table  2). Furthermore, a recent report of a microarray of differentially expressed genes in the aortic arch of ApoE(Ϫ/Ϫ) CD44(Ϫ/Ϫ) mice, compared with ApoE(Ϫ/Ϫ) CD44(ϩ/ϩ) mice, validated 50 genes that are regulated by CD44 (5). The promoter sequences of 47 of these genes were available, and of these, 41 contained the CIRE sequence (CCTGCG). To determine what other genes CD44-ICD could induce in cancer cells, we performed a Gene-Chip microarray analysis (Fig. 4A). Hierarchical clustering analysis of the genes that showed a change in mRNA message expression by at least 2-fold and p Ͻ 0.05 upon transfection of CD44-ICD identified changes in 251 genes (supplemental Fig. S3 and supplemental Table 1). We narrowed down the analysis to require a change in expression by at least 2-fold in both the full-length CD44s and the CD44-ICD transfectants. This analysis identified 64 genes ( Fig. 4A and supplemental Table 2). A Gene Ontology analysis using the Ingenuity IPA tool identified four network functions (supplemental Fig.  S4) and several biological functions associated with this group of 64 genes, including cancer, cellular growth and proliferation, tumor morphology, and cell-to-cell signaling among others.

TABLE 2 Published genes that are regulated by CD44 and the presence of the CD44-ICD response element in their promoters
A PubMed literature review was performed for genes that are affected by different levels of CD44 expression. The promoter sequence within 1,000 or 5,000 bp (underlined numbers) upstream and 299 bp downstream of the transcription initiation site was analyzed for the presence of CIRE using the Transcriptional Regulatory Element Database (Michael Zhang Laboratory, Cold Spring Harbor, NY). We also found that 28 of 46 genes (61%) in which promoter region sequences were available contained at least one CIRE (CCTGCG) within 1,000 bp upstream of the transcriptional start site. Of interest were 16 genes that were up-regulated by CD44-ICD after only 12 h post-transfection but needed 48 h post-transfection of full-length CD44s to exhibit a similar upregulation (Fig. 4B). The lag time of gene expression between CD44-ICD and the full-length CD44s suggests that these genes are directly regulated by CD44-ICD, which is readily available to translocate into the nucleus when transfected as CD44-ICD but requires more time when CD44-ICD is generated by the cleavage of full-length CD44s. Interestingly, 8 of the 13 genes (62%) with known promoter sequence contained the CIREs (CCTGCG) in their promoters. We then evaluated the up-regulation of 10 of these genes by qRT-PCR. We validated the CD44-ICD-mediated up-regulation in seven genes. Of note, six of the seven genes contained the CIRE (CCTGCG). These results indicate that CD44-ICD transcriptionally activates genes that contain the CIRE, in most cases within 1,000 bp of their transcriptional start site.

Gene/Protein
To further understand the group of 16 CD44-ICD-targeted "early response" genes, a Gene Ontology analysis using the Ingenuity IPA tool was performed. One network and several biological functions, including cancer, were identified (Fig. 4C). Interestingly, many of the genes were directly associated with the hypoxia-induced transcription factor Hif1␣ and aerobic glycolysis. We found it intriguing that many of the genes activated by Hif1␣ were also activated by CD44-ICD but that Hif1␣ itself was not elicited in our very stringent Gene-Chip analysis although it does have a CD44-ICD response element in its promoter. One possibility could be that CD44 modulates Hif1␣ expression only under anaerobic conditions. We examined the expression of Hif1␣ in MCF-7 and MCF-7/CD44s cells by Western blotting. Fig. 4D shows that, as expected, Hif1␣ pro-tein was not expressed in either cell line under normoxia. Under hypoxic stress caused by CoCl 2 , Hif1␣ expression was clearly induced in MCF-7 cells but not in MCF-7/CD44s cells in which Hif1␣ expression was suppressed. These results indicate that CD44-ICD can induce Hif1␣-related genes in the absence of hypoxia and bypasses Hif1␣-dependent induction.

DISCUSSION
We previously reported that the expression of CD44s in human cancer cell lines induces a more invasive and drug-resistant cell phenotype (7). However, the augmentation in the invasive phenotype of CD44s-expressing cells is not well understood. One possible mechanism is that CD44 transcriptionally modulates tumor invasion-related genes. The purpose of this study was to use the CD44-mediated overexpression of MMP-9 as a working platform to elucidate the role of CD44 (specifically CD44-ICD) in transcription.
As a group, MMPs degrade most of the extracellular matrix, and therefore they play an important role in invasion and metastasis of tumors (16), where they are frequently overexpressed. In breast cancer, the levels and activity of these MMPs have been shown to display an important correlation with the tumor cell phenotype. For example, MMP-9 is significantly increased and more frequent in malignant tumors compared with benign breast tumors or normal breast tissue (27,28). Furthermore, invasive breast carcinomas with the highest aggressiveness display the highest levels of MMP-9 (27). In the APC-Min model, MMP-9 contributes to early intestinal tumorigenesis (29). Also, the inhibition of expression of MMP-9 in different cancer cell lines causes a marked reduction in invasiveness and CD44 expression, suggesting a functional linkage between CD44 and MMP-9 (30).
In our study, we characterize the functional relationship between CD44 and MMP-9 and demonstrate that the expres-   Table 3 for raw data. C, a gene ontology analysis using the IPA tool identifies a close relationship between Hif1␣ and the "early response" genes in the microarray containing CIRE sequence in their promoters. D, anti-Hif1␣ Western blotting of CoCl 2 -treated cell extracts show a strong induction of Hif1␣ in MCF-7-treated cells, as expected, but a repressed induction of Hif1␣ in MCF-7/CD44s-treated cells.
sion and proteolytic processing of CD44s, with the subsequent generation of the CD44-ICD, is directly associated with the up-regulation of the MMP-9 gene. Our data also demonstrate that the up-regulation of gene expression depends on the CD44-ICD nuclear translocation and binding to its novel response element, CCTGCG. The following experimental data support these conclusions.
First, we show that the stable expression of CD44s in MCF-7 cells not only correlates with an increase in MMP-9 activity but also induces an increase in pro-MMP-9 activation (Fig. 1). Of note, the increment of mature MMP-9, but not pro-MMP-9, is probably a function of the ability of CD44 to act as a cell membrane docking molecule for the pro-MMP-9 zymogen, which allows its successive activation (31,18). Interestingly, the expression of MMP-2 remains the same regardless of CD44s expression. This apparent discrepancy in expression between MMP-2 and MMP-9 is known to be due to differences in cis-element composition of their promoters (32). Our data show a CD44related differential regulation of the promoters of these two MMPs and thus assign a transcriptional inducer function to CD44s in the regulated expression of MMP-9. Interestingly, the hyaluronanmediated activation of CD44 induces its proteolytic cleavage at the cell surface to promote cell motility and subsequently tumor progression (33,34). This extracellular cleavage is followed by additional intracellular cleavages to generate the CD44-ICD fragment (14), which we show is directly involved in the regulated expression of MMP-9. We also show for the first time that the processing of CD44s may occur constitutively without the need of exogenous induction in breast cancer cells.
Second, we demonstrate that the CD44-mediated expression of MMP-9 directly correlates with the enzymatic activity of presenilin-1, as part of the ␥-secretase complex, the underlying enzymatic machinery that generates CD44-ICD after regulated intramembranous cleavage of CD44 (22,35). We demonstrate that when presenilin-1 activity is down-regulated, MMP-9 enzymatic activity decreases (Fig. 2). We support this conclusion by showing that the expression of an uncleavable CD44 is not capable of inducing MMP-9 expression. In agreement with these data, the proteolytic processing of CD44 has been detected in numerous carcinomas, including breast carcinoma, in which MMP-9 has also been reported to be overexpressed (36 -38). We show in Table 1 that not only are MMP-9 and CD44 usually co-expressed in ovarian and breast cancers, but also MMP-9 is rarely expressed in the absence of CD44 expression in more than 200 human tumors studied, making our findings biologically relevant.
Third, we show that ChIP assays localize CD44-ICD in a complex with Runx2 to the MMP-9 promoter (Fig. 3). These findings start to connect several components and mechanisms associated with the complex regulation of the MMP-9 gene in cancer cells. For example, MMP-9 expression in prostate (i.e. LNCap and PC3) and breast (i.e. MCF-7 and MDA-MB-231) cancer cell lines correlated with their Runx2 expression and transcriptional activity (24). Interestingly, the CD44s protein expression in those cells directly correlates with the Runx2-dependent MMP-9 expression (39,40). We used ELISAs as well as an extensive EMSA analysis to demonstrate that CD44-ICD binds directly to the DNA of the MMP-9 promoter. Further-more, we uncover a CD44-ICD DNA binding sequence and corroborate its importance using an Affymetrix assay. The majority of the genes up-regulated by the transfection of CD44-ICD have the sequence CCTGCG in their promoters, usually within 1,000 bp upstream of the transcription start site.
Among the early response genes induced by CD44-ICD listed in Fig. 4B, three important enzyme-encoding genes in oxidative glycolysis were found. They are ALDOC, which encodes aldolase c, fructose biphosphate; PDK1, which encodes pyruvate dehydrogenase kinase-1; and PFKFB4, which encodes 6phosphofructose-2-kinase/fructose-2,6-biphosphatase 4. This uncovers a possible role of CD44 in oxidative glycolysis. Early in the past century, Warburg observed that cancer cells underwent anaerobic glycolysis in the face of adequate oxygen supply (50). The key enzyme of the glycolytic pathway, 6-phosphofructose-1-kinase, is controlled allosterically by the CD44-ICD-inducible fructose-2,6-bisphosphatase. Fructose-2,6-P(2) is an allosteric effector that coordinately regulates two opposing pathways, glycolysis and gluconeogenesis, depending on the nutritional and endocrine state of the cell. PFKFB4 is a bifunctional enzyme that can acutely modulate fructose-2,6-P(2) levels and is known to be elevated in tumors (41). Another gene that we validated as induced by CD44-ICD is PDK1. The product of this gene inhibits pyruvate dehydrogenase and forces glycolysis to convert pyruvate into lactate rather than going into the citric acid cycle. Lactate increases the acidity of the tumor microenvironment, which then allows the cells to invade the extracellular matrix (42). Lactate also regenerates NAD ϩ for continued glycolysis (43). The induction of these metabolic genes implicates CD44 for the first time in the transcriptional regulation of cell metabolism. It suggests that by increasing the transcription of key regulatory enzymes in oxidative glycolysis, CD44 may promote the Warburg effect in cancer cells that are CD44(ϩ) and known to preferentially use aerobic glycolysis for survival regardless of oxygen concentration. Because CD44 is a known stem cell marker for many tumors, it is reasonable to speculate that by inducing aerobic glycolysis, CD44 maintains the metabolic needs of stem cells that are usually found in niches of intermittent hypoxia and would benefit from oxidative glycolysis as a way to survive. Most recently, Saya's group (44) reported that CD44 interacts with pyruvate kinase M2 (PKM2) and induces aerobic glycolysis in p53 mutant cells. Future work will be needed to test these hypotheses. Based on the work presented, a hypothetical model of the CD44-ICD-dependent CD44 signaling to modulate transcription anticipates direct as well as indirect interactions with promoters and transcription factors (Fig. 5).