Regulation of Human Vascular Endothelial Growth Factor mRNA Stability in Hypoxia by Heterogeneous Nuclear Ribonucleoprotein L*

A 126-base region of human vascular endothelial growth factor (VEGF) 3′-untranslated region, which we identified as the hypoxia stability region, forms seven hypoxia-inducible RNA-protein complexes with apparent molecular masses ranging from 40 to 90 kDa in RNA-UV-cross-linking assays. In this study, we show that proteins that form the 60-kDa RNA-protein complex with the hypoxia stability region were present in both cytoplasmic and nuclear compartments. We purified the protein associated in the 60-kDa complex and identified it as heterogeneous nuclear ribonucleoprotein L (hnRNP L) by protein sequencing. Removal of hnRNP L by immunoprecipitation specifically abolished formation of the 60-kDa complex. Synthetic deoxyribonucleotide competition studies defined the RNA-binding site of hnRNP L as a 21-base-long sequence, 5′-CACCCACCCACAUACAUACAU-3′. Immunoprecipitation of hnRNP L followed by reverse transcription-polymerase chain reaction showed that hnRNP L specifically interacts with VEGF mRNA in hypoxic cells in vivo. Furthermore, when M21 cells transfected with antisense oligodeoxyribonucleotide to the hnRNP L RNA-binding site, the VEGF mRNA half-life was significantly reduced under hypoxic conditions. Thus, we propose that specific association of hnRNP L with VEGF mRNA under hypoxia may play an important role in hypoxia-induced post-transcriptional regulation of VEGF mRNA expression.

Vascular endothelial growth factor (VEGF), 1 also known as vascular permeability factor, is a potent angiogenic and endothelial cell-specific mitogen (1,2). VEGF is expressed and secreted at low levels by most normal cells but constitutively expressed at high levels by many human tumors and tumor cell lines (1)(2)(3)(4). Hypoxia up-regulates VEGF expression and several studies have demonstrated that increase in transcription alone does not account for all of the increase in VEGF mRNA (3)(4)(5)(6)(7)(8). The post-transcriptional regulation of VEGF mRNA stability also plays a critical role in the observed hypoxic induction (6 -8).
Post-transcriptional regulatory mechanisms, especially modulation of mRNA stability, has been shown to play a major role in gene expression (9). The turnover rate of a given mRNA can be determined by interactions of trans-acting factors with specific cis-element located within 3Ј-untranslated regions (3Ј-UTR) (9,10). Many labile mRNAs, including those that encode lymphokines, cytokines, transcription factors, and proto-oncogenes, contain AU-rich elements (AREs) in their 3Ј-UTR (9). Identification of the interaction of AREs with trans-acting proteins has been the first step in understanding the molecular regulation of mRNA stability (9,10). The presence of a reiterated pentamer (AUUUA) n in many AREs has been shown to be associated with rapid mRNA turnover and translation attenuation (10 -12). In the case of granulocyte-microphage colonystimulating factor, c-Fos, and c-Myc mRNAs, deletion of the ARE region enhances their stability, and insertion of the region into the 3Ј-UTR of a normally stable globin mRNA significantly destabilizes it (10 -15). A variety of AUUUA-binding proteins have been identified, and examples of these include (i) a 32-kDa nuclear protein from HeLa cells (15); (ii) AU-A, AU-B, and AU-C, 30 -43-kDa nuclear and/or cytoplamic proteins from human T lymphocytes (16,17); (iii) AUBF, a heterotrimeric protein formed by 15-, 17-, and 19-kDa subunits and present in both nucleus and cytoplasm (18,19); and (iv) AUBP, a 36-kDa cytoplasmic protein from human spleen identified as glyceraldehyde-3-phosphate dehydrogenase (20). However, the mechanisms of how these proteins affect mRNA turnover remains unclear.
Our previous work identified a 126-base hypoxia stability region (HSR) in human VEGF 3Ј-UTR that is critical for the stabilization of VEGF mRNA under hypoxia (21). This region is able to form seven hypoxia-inducible mRNA-protein complexes (21). Here we report hnRNP L as a protein that interacts with the VEGF HSR and forms a hypoxia-inducible 60-kDa RNA-hnRNP L complex. The cytoplasmic hnRNP L specifically interacted with VEGF mRNA in hypoxic cells in vivo and regulated VEGF mRNA stability. Thus, we propose that a specific interaction of hnRNP L with VEGF mRNA may play an important role in hypoxia-induced post-transcriptional regulation of human VEGF mRNA stability.

MATERIALS AND METHODS
Antibodies and Oligodeoxyribonucleotides-Monoclonal antibodies 4D11 (anti-hnRNP L) and 4F4 (anti-hnRNP C) were generously provided by Dr. Gideon Dreyfuss (University Pennsylvania, Philadelphia) (22,23). All of the oligodeoxyribonucleotides used in the study were synthesized from Genemed Synthesis (San Francisco, CA).
Preparation of Cytoplasmic and Nuclear Extracts-Cytoplasmic and nuclear extracts were obtained as described by Claffey et al. (21). Following exposure to normoxia or hypoxia (3% O 2 ), M21 cells were washed three times in ice-cold PBS followed by lysis in 1% Triton X-100 lysis buffer containing 50 mM Hepes, pH 7.5, 10 mM sodium pyrophosphate, 150 mM NaCl, 100 mM NaF, 0.2 mM NaOVa 4 , 1 mM EGTA, 1.5 mM MgCl 2 , 10% glycerol, and 5 mM 4-(2-aminoethyl)benzene-sulfonyl fluoride (Sigma). The cytoplasmic extract was collected and centrifuged at 14,000 ϫ g for 15 min, and the nuclei pellet were further extracted with the 1% Triton X-100 lysis buffer containing 400 mM NaCl for 15 min on ice. The nuclear extract was recovered after centrifuged at 14,000 ϫ g for 15 min.
In Vitro Transcription-The sense and antisense HSR of human VEGF were transcribed in vitro as described previously (21), using T7 RNA polymerase transcription of NotI-linearized and T3 RNA polymerase transcription of EcoRI-linearized plasmid, respectively. Both biotin-UTP-labeled and [ 32 P]UTP-labeled RNA transcripts were generated using an RNA in vitro transcription kit (Stratagene, La Jolla, CA) according to the manufacturer's protocol and then treated with RNasefree DNase (Promega, Madison, WI) for 15 min at 37°C. The [ 32 P]UTPlabeled transcripts were extracted once with phenol:chloroform and loaded onto RNase-free G-50 spin columns (Boehringer Mannheim) to remove free ribonucleotides. The biotin-UTP-labeled sense and antisense HSR transcripts were precipitated with ethanol and dissolved in diethyl pyrocarbonate-treated H 2 O. The transcripts were then immobilized to an avidin-Sepharose 4B column (Amersham Pharmacia Biotech, Uppsala, Sweden) at 4°C for 1 h and used in mRNA-binding protein affinity purification.
RNA Binding and UV Cross-linking (RNA-UVXL)-RNA-UVXL was performed as described earlier (21). Radiolabeled RNA transcripts (250,000 cpm/reaction) were incubated with cytoplasmic or nuclear proteins (40 g/reaction) in 30 l of RNA binding buffer containing 10 mM Hepes, pH 7.5, 5 mM MgCl 2 , 50 mM KCl, 0.5 mM EGTA, 0.5 mM dithiothrietol, 10% glycerol, 100 g/ml tRNA, and 5 mg/ml heparin for 20 min at 30°C. The mixtures were UV cross-linked by UV Stratalinker (Stratagene) at room temperature for 10 min (total energy 1800 J/cm2) followed by RNase digestion (40 units of RNase T1 (Boehringer Mannheim) and 1 g of RNase A (Sigma)) for 15 min at room temperature. The sample was then denatured in SDS-PAGE sample buffer under reducing conditions and RNA-protein complexes were analyzed by 8% SDS-PAGE and autoradiographed with Eastman Kodak Co. MR film.
Affinity Purification of mRNA-binding Proteins-Cytoplasmic extracts were prepared after M21 cells were cultured in hypoxia for 24 h. The lysate was sequentially loaded onto poly(A)-, poly(U)-, antisense HSR-and sense HSR-Sepharose 4B RNA affinity columns at room temperature. The sense HSR-Sepharose column was then extensively washed with phosphate-buffered saline, 0.2% Triton X-100. The mRNA-binding proteins were then eluted with 0.9 M NaCl and precipitated with 3 volumes of ethanol.
Immunoprecipitation-M21 cytoplasmic lysates were incubated with a 1:500 dilution of anti-hnRNP C ascites (4F4), anti-hnRNP L ascites (4D11; both 4F4 and 4D11 were obtained from Dr. Gideon Dreyfuss, University of Pennsylvania, Philadelphia, PA), or control antibody (normal mouse serum) for 2 h at 4°C. The lysates were further incubated with 30 l of protein A-Sepharose beads (Amersham Pharmacia Biotech) for 2 h at 4°C. Beads were recovered by brief centrifugation, washed three times in lysis buffer, and RNA was extracted for RT-PCR or denatured directly with SDS-PAGE buffer for Western blot.
Western Blot-The immunoprecipitated material was denatured under reducing conditions, and proteins were analyzed by 8% SDS-PAGE. Western blots were performed using a 1:5000 dilution of anti-hnRNP L monoclonal antibody, anti-hnRNP C monoclonal antibody, and normal mouse serum, followed by a 1:10,000 dilution of horseradish peroxidaseconjugated goat anti-mouse IgG (Amersham Pharmacia Biotech) in a blocking buffer containing 1% bovine serum albumin and 0.1% Tween 20 in Tris-buffered saline. The blots were then developed with the ECL system (Amersham Pharmacia Biotech).
RT-PCR and Southern Blot Detection-The protein A-Sepharosebound RNA was extracted with an equal volume of Trizol (Life Technologies, Inc.) and 1 ⁄5 volume of chloroform. The supernatant was collected after centrifugation at 14,000 ϫ g for 15 min, and RNA was precipitated with an equal volume of isopropyl alcohol at Ϫ80°C for 1 h. Reverse transcription was carried out with oligo(dT) (15-mer) as primer and incubated at 42°C for 1 h in the presence of reverse transcriptase and dNTPs (Boehringer Mannheim). The transcribed cDNA was collected after removal of free dNTPs through RNase-free G-50 spin column (Boehringer Mannheim). PCR amplification was carried out with a Taq DNA polymerase and two VEGF primers (VEGF 5Ј primer: 5Ј-TATGCGGATCAAACCTCAC-3Ј, corresponding to nucleotides 374 -393 of VEGF coding region encoding amino acids 82-88 of the mature protein; VEGF 3Ј primer: 5Ј-ATAACATTAGCACTGTTAATTT-3Ј, corresponding to nucleotides 435-457 3Ј to the translation strip codon, GenBank TM accession number AF022375) and amplified for 30 cycles. The 716-base-long PCR products were resolved by 1% agarose gel electrophoresis and detected by Southern blot with a [ 32 P]dCTP-labeled human VEGF165 AccI/NcoI fragment (823 base pairs) encompassing the coding region and 330 base pairs of 3Ј-UTR of the full-length cDNA. Probes were prepared by the random-primed synthesis method using the Multiprime kit (Amersham Pharmacia Biotech). DNA was blotted by capillary transfer to a nylon membrane (NEN Life Science Products) in 10ϫ SSC. Blots were cross-linked with UV Stratalinker 1800 (Stratagene), baked at 80°C for 15 min, and prehybridized for 2-4 h at 65°C. Hybridization was carried out overnight at 65°C, and blots were washed at 0.5% SDS, 0.1% SSC at 55°C before they were exposed to Kodak MR film.
Oligonucleotide Transfection and Northern Blot Detection-Antisense oligodeoxyribonucleotides, AS1 (1 M) and AS3 (1 M), were transfected into M21 cells by Lipofectin following the instructions of the supplier (Life Technologies, Inc.). The transfected cells were then cultured in hypoxia for 8 h, and total cellular RNA was isolated by using the RNeasy RNA extraction kit (Qiagen, Chatsworth, CA) according to the manufacturer's instructions. For the VEGF mRNA stability study, 5 g/ml actinomycin D was added to AS1, AS3, or mock-transfected M21 cells after 6 h of hypoxia incubation. The time intervals of actinomycin D treatment were 30 min, 1 h, and 2 h, and the total hypoxia incubation time was 8 h. Total RNA was isolated and electrophoresed in 1% agarose gels containing 2.2 M formaldehyde (10 g/lane). RNA was blotted by capillary transfer to a nylon membrane (NEN Life Science Products) in 10ϫ SSC. Blots were cross-linked with UV Stratalinker (Stratagene), baked at 80°C for 15 min, and prehybridized for 2-4 h at 65°C. Hybridization was carried out overnight at 65°C with a [ 32 P]dCTP-labeled human VEGF165 AccI/NcoI fragment and GLUT-1 BamHI fragment (2.47 kilobase pairs). A ribosome-associated protein cDNA, 36B4, was used as a control (21). Blots were washed with 1% SDS, 1% SSC at 55°C and exposed to Kodak MR film. VEGF was normalized to 36B4 expression using PhosphorImager analysis (Molecular Dynamics, Sunnyvale, CA).
Studies have shown that RNA-binding proteins are also present in nuclear compartments (15)(16)(17). Thus, we prepared both cytoplasmic and nuclear extracts from hypoxic or normoxic cultured M21 cells and compared RNA-protein complex formation by RNA-UVXL with 32 P-labeled VEGF HSR. As shown in Fig. 1B, 24-h hypoxia (3% O 2 ) substantially up-regulated the formation of the 90-, 88-, 72-, and 60-kDa complexes when compared with cytoplasmic extracts made from normoxic (21% O 2 ) cells. The 90-, 88-, 72-, and 56-kDa complexes were selectively cytoplasmic with little or no nuclear localization (Fig. 1B). Significantly higher levels of the 60-kDa complex were formed with nuclear extracts than cytoplasmic extracts, and the complex was markedly increased by hypoxia treatment in both extracts (Fig. 1B). Other than the 60-kDa complex, the nuclear extract also formed three unique complexes with ap-parent molecular masses of 45, 48, and 120 kDa (Fig. 1B). Taken together, these results suggest that proteins that form the 90-, 80-, 72-, and 56-kDa complexes with HSR mainly appear in cytoplasm. In contrast, the protein that forms the 60-kDa complex with HSR is present in both nuclear and cytoplasmic compartments with the majority in the nucleus. Thus, given the distribution and the hypoxia induction of the 60-kDa complex, we established a protocol to purify the RNAbinding protein in the 60-kDa complex.
Purification of the RNA-binding Protein in the 60-kDa RNA-Protein Complex-The VEGF HSR is a highly AU-rich element containing 54 A bases and 52 U bases out of 126 bases, resulting in 43% A and 41% U distribution (Fig. 1A). To purify the binding protein that forms the 60-kDa complex, hypoxiatreated M21 cell lysates were loaded sequentially onto poly(A)-, poly(U)-, and antisense HSR-coupled columns to remove nonspecific RNA-protein interactions prior to loading to the sense HSR-coupled column. To ensure that the protein that formed the 60-kDa complex was not removed by poly(A)-, poly(U)-, and antisense HSR-coupled columns, the flow-through from each column was collected, and RNA-UVXL was performed with 32 P-labeled HSR. As shown in Fig. 2A, the poly(A) column effectively interacted with proteins forming the 90-and 72-kDa complexes and removed them from the cell lysates, whereas the poly(U) column removed the proteins that formed the 88-, 56-, 46-, and 40-kDa complexes. Further incubation of the lysate with the antisense HSR column completely removed proteins forming the 72-kDa complex, whereas the 60-kDa complex selectively bound to the sense HSR column (Fig. 2A). These results suggest that the proteins forming the 90-and 72-kDa complexes can interact with poly(A) sequences, whereas proteins forming the 88-, 56-, 46-, and 40-kDa complexes can interact with poly(U) sequences. In contrast, the proteins present in the 60-kDa complex only specifically interact with a unique sequence in the HSR.
To analyze proteins that bound to the four mRNA columns, small aliquots of the column material were directly analyzed in reducing 6% SDS-PAGE. As shown in Fig. 2B, all four mRNA columns interacted with numerous proteins. However, the sense HSR column contained one unique protein band with an apparent molecular mass near 60 kDa (indicated by an arrow). When eluted with various concentrations of NaCl, the 60-kDa protein was successfully released from the sense HSR column with 0.9 M NaCl (Fig. 2B, indicated by an arrow). The purified 60-kDa protein was then separated in a preparative 6% SDS-PAGE, stained with Coomassie Blue, and sequenced after trypsin digest and HPLC purification (Harvard Microchemistry Protein Sequencing Facility). A 16-amino acid sequence of an HPLC-purified peptide was determined as SDALETLGFLN-HYQMK, which is identical to hnRNP L, amino acid residues 522-537 (22).
The RNA-binding Protein in the 60-kDa Complex Is hnRNP L-To further confirm the RNA-binding protein in the 60-kDa complex as hnRNP L, we used anti-hnRNP L monoclonal antibody to immunoprecipitate hnRNP L from M21 cell lysates followed by detection of the 60-kDa complex formation by RNA-UVXL. For hnRNP control, anti-hnRNP C monoclonal antibody was used. Direct Western blot analysis of M21 cytoplasmic extracts showed that the anti-hnRNP L monoclonal antibody identified three proteins with apparent molecular masses of 66, 60, and 56 kDa, with the 60-kDa band being the most abundant of the three (Fig. 3A), indicating that hnRNP L is expressed as three different isoforms or differentially modified proteins in M21 cells. Anti-hnRNP L immunoprecipitation removed all of the 56-and 66-kDa bands and most of the 60-kDa hnRNP L band (Fig. 3A). Control antibodies (anti-hnRNP C and normal mouse serum) did not interact with any of the three hnRNP L proteins (Fig. 3A). Fig. 3B shows that removal of hnRNP L molecules by immunoprecipitation abolished formation of the 60-kDa RNA-protein complex in RNA-UVXL, whereas formation of the other RNA-protein complexes was unaffected. Immunoprecipitation with anti-hnRNP C did not affect any of the seven HSR-protein complexes, suggesting that hnRNP C is not one of the RNA-binding proteins present in these complexes. Taken together, these results further prove that hnRNP L is indeed the RNA-binding protein responsible for the 60-kDa RNA-protein complex. Removal of hnRNP L from cell lysates did not affect the formation of the other six RNA-protein complexes, suggesting that hnRNP L does not form specific com- plexes with other RNA-binding proteins interacting with VEGF HSR.
Hypoxia Regulation of hnRNP Expression-Hypoxia substantially increases the steady state level of VEGF mRNA (4 -8). The formation of hnRNP L-VEGF 3Ј-HSR complex markedly increased when cytoplasmic protein extracts were obtained from M21 cells exposed to hypoxia (3% O 2 ) as compared with those exposed to normoxia (21% O 2 ) (Fig. 1B). Thus, it was of interest to compare the effect of hypoxia on synthesis and distribution of hnRNP L in nuclear versus cytoplasmic compartments. As shown in Fig. 4, M21 cells grown in normoxic conditions contained high levels of hnRNP L, especially in the nuclear compartment. A minor induction of hnRNP L can be seen under hypoxia in both cytoplasm and nucleus, with the 56-kDa immunoreactive band in cytoplasm being the most significantly increased (Fig. 4). Conversely, the hnRNP C level in cytoplasm decreased to an undetectable range after a 24-h hypoxia incubation, although it remained high in the nucleus (Fig. 4). These results suggest that hypoxia differentially regulates the distribution of hnRNPs in different cellular compartments.
Identification of hnRNP L mRNA Binding Site-To define the hnRNP L mRNA binding site, six antisense oligodeoxyribonucleotides were synthesized from different regions of the HSR (Fig. 5A, AS1-AS6). To ensure the stability of oligonucleotides in the RNA-UVXL, we synthesized oligodeoxyribonucleotides instead of oligoribonucleotides. As shown in Fig. 5B, AS1 (1 M) efficiently blocked the interaction of hnRNP L with 32 P-labeled HSR and abolished formation of the 60-kDa RNA⅐hnRNP L complex. The blocking of HSR⅐hnRNP L complex formation can be seen as low as 0.05 M AS1 (data not shown). AS6 abolished the formation of the 90-and 88-kDa complexes and partially inhibited the HSR⅐hnRNP L complex, whereas AS2, AS3, AS4, and AS5 did not show any effect (Fig.  5B). These results suggest that the complementary region of AS1 on HSR contains the hnRNP L binding site and that hnRNP L interacts with the single-stranded mRNA region. AS1 and AS6 lack sequence homology; thus, the partial inhibition of hnRNP L interaction with HSR observed with AS6 may be functioning by blocking the 90-and 88-kDa complexes that may promote hnRNP L binding in some manner.
To further explore the hnRNP L mRNA binding site, we chemically synthesized the sense oligodeoxyribonucleotide (S1), which was complementary to AS1 (Fig. 6A) and attempted to compete for hnRNP L binding to 32 P-labeled HSR. As shown in Fig. 6A, S1 (10 M) successfully competed with 32 P-labeled HSR for hnRNP L binding and substantially reduced HSR-hnRNP L complex formation (85% inhibition of hnRNP L binding). The competition of S1 for hnRNP L binding to HSR can be seen as low as 0.1 M AS1 (data not shown). To define the hnRNP L binding site, eight different oligodeoxyribonucleotides were synthesized after modification of the internal base sequences of S1 (Fig. 6). As shown in Fig. 6A, substitution of base 333 G to A in S1.1 or further substitution of base 335 C to A in S1.2 showed strong competition for hnRNP L binding to the 32 P-labeled HSR (S1.1 and S1.2 showed 74 and 82% inhibition, respectively), thus demonstrating that neither 333 G or 335 C were required for hnRNP L binding. However, substitu- tion of base 337 C to A in S1.3 showed a substantial reduction in the ability to compete for hnRNP L binding (Fig. 6A, 57% inhibition), and any modifications after base 337 showed a similar affect (Fig. 6A; S1.4 and S1.5 showed 54 and 50% inhibition, respectively). These results suggest that bases 332-336 (AGACA) in S1 are not essential for hnRNP L binding and that S1.2 contains the hnRNP mRNA binding site.
The S1.2 oligodeoxyribonucleotide has three ACAU repeats. To study the potential importance of the ACAU repeats for hnRNP L binding, three oligodeoxyribonucleotides (S1.6, S1.7, and S1.8) with 1-3 ACAU repeats were synthesized (Fig. 6B). As shown in Fig. 6B, removal of any one of the ACAU repeats from S1.2 substantially reduced the ability to compete for hnRNP L binding with 32 P-labeled HSR (S1.6, S1.7, and S1.8 showed only 44, 36, and 31% inhibition, respectively). This result suggests that all three ACAU repeats are essential for hnRNP L binding. Overall, these results suggest that the minimal hnRNP L mRNA binding site appears to be 337 CACCCAC-CCACAUACAUACAU 357 . Using a GenBank TM Blast search, this 21-base sequence was only found in human and bovine VEGF 3Ј-UTRs but not in other VEGF species or other genes.
The hnRNP L Interacts with VEGF mRNA in Hypoxic Cells in Vivo-To determine if hnRNP L interacts with VEGF mRNA in vivo, we immunoprecipitated hnRNP L with anti-hnRNP L monoclonal antibody from 24-h hypoxia-cultured M21 cell lysates and performed RNA extraction, RT-PCR, and Southern blot detection. The VEGF primers used in RT-PCR contained part of the VEGF coding region and the entire 3Ј-HSR and should produce a 716-base VEGF165 isoform product. As shown in Fig. 7A, a unique PCR product of approximately 700 base pairs in length appeared with hnRNP L immunoprecipitation, whereas immunoprecipitates for hnRNP C and normal mouse serum control antibodies did not show any distinct band in the same region. Southern blot hybridization with VEGF cDNA probe showed a strong hybridization to the hnRNP L PCR product, thus confirming the PCR product as VEGF cDNA (Fig. 7B). A weak interaction of VEGF cDNA probe with the hnRNP C PCR product was observed, whereas the normal mouse serum control did not have any detectable signal (Fig.  7B). Studies have shown that hnRNP C can interact with the poly-U region of mRNAs (30), and thus it is possible that hnRNP C is associating with many mRNAs including VEGF in vivo. Taken together, these results verify that hnRNP L interacts with VEGF mRNA in hypoxic cells in vivo.
Interaction of hnRNP L and VEGF mRNA Is Important for VEGF mRNA Stability-To study whether hnRNP L plays an important role in the regulation of VEGF mRNA expression, M21 cells were transfected with antisense oligodeoxyribonucleotide AS1 to block interaction of hnRNP L with VEGF mRNA in hypoxic cells in vivo. The AS3 oligodeoxyribonucleotide, which did not block the formation of any HSR-protein complex formation (Fig. 5), was used as control. As shown in Fig. 8, M21 cells transfected with AS1 had substantially lower levels of VEGF mRNA when compared with mock-transfected control cells (54.3% by PhosphorImager analysis), whereas the GLUT-1 mRNA level was not affected by the transfection (97.3%), indicating that AS1 is specifically affecting VEGF mRNA expression. The control oligodeoxyribonucleotide AS3transfected cells did not show changes for either VEGF mRNA (109.5%) or GLUT-1 mRNA (94.5%). These results suggest that blocking of hnRNP L and VEGF mRNA interaction in vivo will specifically affect VEGF mRNA accumulation in hypoxia.
The hnRNPs are a family of abundant nuclear proteins that are involved in pre-mRNA processing and splicing (22)(23)(24). To address whether the effect of AS1 repression of VEGF mRNA accumulation is through VEGF mRNA stability, we transfected M21 cells with AS1, AS3, or mock reagent and examined VEGF mRNA decay under hypoxic conditions. The rate of decay of the mature VEGF mRNA was determined by Northern blot hybridization after treatment of cells with the transcriptional inhibitor actinomycin D. As shown in Fig. 9A, M21 cells transfected with AS1 showed a substantial decrease in VEGF mRNA level after 30 min of actinomycin D treatment, whereas mock-and AS3-transfected cells maintained higher VEGF levels for up to 1 h of actinomycin D treatment. Conversely, GLUT-1 mRNA did not show any changes by AS1 when compared with mockand AS3-transfected cells (Fig. 9A). When the VEGF mRNA decay curve of a triplicate assay was plotted over a 2-h actinomycin D treatment and VEGF mRNA half-life (t1 ⁄2 ) was measured, AS1-transfected cells showed a half-life of 32 Ϯ 5.7 min, whereas mock-and AS3-transfected cells showed half-lives of 53 Ϯ 4.1 and 57 Ϯ 5.6 min, respectively (Fig. 9B). AS1-transfected cells showed statistically significant differences (p Յ 0.0005) in VEGF mRNA levels when compared with mocktransfected cells at both 30 and 60 min. These results suggest that blocking of the hnRNP L and VEGF mRNA interaction decreases VEGF mRNA stability under hypoxia. These also suggest that although hnRNP L is important for mRNA processing, interaction of hnRNP L with VEGF mRNA 3Ј-UTR plays a critical role in post-transcriptional regulation of VEGF mRNA stability.
FIG. 6. Identification of hnRNP L binding site on HSR. S1 is a sense-orientated oligodeoxyribonucleotide that is complementary to AS1 and comprises bases 332-357 of the HSR. A, competitive RNA-UVXL with sense (S) oligodeoxyribonucleotides with modifications in the 5Ј-end. Competitive RNA-UVXL was performed with 32 P-labeled VEGF HSR and M21 cytoplasmic extract (24-h hypoxia) in the absence (Ϫ) or presence of S1 or its derivatives (10 M). Band intensities of the 60-kDa HSR-hnRNP L complex were quantified by PhosphorImager analysis. The percent inhibition was calculated as the percentage decrease in band intensity compared with control (Ϫ). B, competitive RNA-UVXL with sense oligodeoxyribonucleotides with 3Ј modifications. C, minimal defined 21-base binding site for hnRNP L-VEGF HSR binding. FIG. 7. Interaction of hnRNP L with VEGF mRNA in hypoxic cells in vivo. M21 cytoplasmic extracts (24-h hypoxia) were immunoprecipitated with anti-hnRNP L monoclonal antibody (L), anti-hnRNP C monoclonal antibody (C), or normal mouse serum (Ϫ). The co-immunoprecipitated mRNAs were extracted and used as templates for RT-PCR. The PCR products were resolved by 1% agarose gel electrophoresis (A) and detected by Southern blot with a [ 32 P]dCTP-labeled human VEGF (B).

DISCUSSION
With a half-life of only 30 -45 min under normal growth conditions (6 -8), VEGF mRNA falls within a class of labile mRNAs encoding for many transiently expressed proteins in-cluding cytokines, lymphokines, oncogenes, and transcriptional activators (9). The mechanisms whereby normally labile mRNAs are stabilized by stimuli such as hypoxia, growth stimulation, and viral infection are unclear. On the basis of RNAbinding and UV cross-linking, the interaction of specific cellular proteins with mRNAs have been shown to be altered in response to stimuli, which correlates with changes in mRNA stability (9 -21). This has led to the hypothesis that mRNA turnover is mediated principally through mRNA-binding proteins that specifically recognize AREs and other sequence motifs.
Our previous work demonstrated a 126-base HSR in human VEGF 3Ј-UTR, which is responsible for hypoxia-induced VEGF mRNA stability and forms seven hypoxia-inducible RNA-protein complexes with M21 human melanoma cell extracts (21). In the present study, we showed that the protein that forms the 60-kDa RNA-protein complex is present in both nuclear and cytoplasmic compartments (Fig. 1). Protein sequencing and immunodepletion studies identified that protein in the 60-kDa RNA-protein complex as heterogeneous nuclear ribonucleoprotein L (hnRNP L) (Figs. 3 and 4). The specific mRNA binding site of hnRNP L was identified as 3Ј-CACCCACCCACAUA-CAUACAU-5Ј, a 21-base single-stranded element, which is unique to human and bovine VEGF 3Ј-UTR sequences (Figs. 5 and 6). Immunoprecipitation of hnRNP L followed by RT-PCR showed that hnRNP L specifically interacted with VEGF mRNA in hypoxic cells in vivo (Fig. 7). Furthermore, when M21 cells were transfected with antisense oligodeoxyribonucleotide to the hnRNP L RNA-binding site, the hypoxia-induced VEGF mRNA half-life decreased from 53 Ϯ 4.1 min to 32 Ϯ 5.7 min (Fig. 8). This study identifies for the first time that hnRNP L as a protein present in human cells that is capable of interacting with VEGF mRNA 3Ј-UTR and is functionally involved in the post-transcriptional regulation of VEGF mRNA under hypoxic conditions.
It is well documented that unstable mRNAs containing AREs generally consists of either scattered AUUUA pentanucleotides, contiguous AUUUA repeats, nonamer motif UUAUUUA(U/A)(U/A), or a stretch of AU residues lacking either motif (10 -15). These AREs interact with cytoplasmic and/or nuclear proteins and usually form ARE-protein complex with average molecular mass less than 40-kDa (10,(15)(16)(17)(18)(19)(20). The VEGF HSR is also a highly AU-rich element (43% A and 41% U) and contains one AUUUA pentanucleotide as well as two stretches of AU residues. However, interaction of the HSR with cellular proteins formed seven RNA-protein complexes with higher molecular masses (ranging from 40 to 90 kDa). The HSR also formed same protein complexes with five other cell lines we tested (data not shown), indicating that those RNA-binding proteins are common proteins to different cell lines. The 3Ј-UTRs of two other hypoxia up-regulated genes, erythropoietin (25) and tyrosine hydroxylase (26), interacted with mRNA-binding proteins with apparent molecular masses ranging from 66 to 140 kDa. However, the VEGF HSR and the 3Ј-UTRs of both erythropoietin and tyrosine hydroxylase lack significant homology. Moreover, the interaction site for a 66-kDa protein to tyrosine hydroxylase 3Ј-UTR is different from the hnRNP L mRNA binding site. Thus, VEGF HSR-binding proteins may be distinct from those that recognize other hypoxia-inducible genes.
Studies of rat VEGF mRNA demonstrated a 600-base region covering nucleotides 1251-1877 in the 3Ј-UTR required for rat VEGF mRNA stability (27,28). However, little homology exists between the human VEGF HSR sequence and the rat 600-base region. In addition, the protein complexes observed by EMSA in the rat ARE-binding proteins are 17, 28, and 34 kDa (27, 28), FIG. 8. Blocking of hnRNP L and VEGF mRNA interaction reduces steady state VEGF mRNA level. M21 cells were transiently transfected in the absence (Ϫ) or presence of 1 M AS1 or AS3 oligodeoxyribonucleotides, and VEGF, GLUT-1, and 36B4 mRNA levels were detected by Northern blot analysis. VEGF and GLUT-1 mRNA signals were normalized with the ribosome-associated mRNA, 36B4, from each lane and quantified by PhosphorImager. The mRNA from AS1-and AS3-transfected cells were calculated as percentage of VEGF or GLUT-1 of mock-transfected mRNA (Ϫ) signals designated as 100%.
FIG. 9. Blocking of hnRNP L and VEGF mRNA interaction by AS1 oligodeoxyribonucleotide decreases VEGF mRNA stability. M21 cells were transiently transfected in the absence (Ϫ) or presence of 1 M AS1 or AS3 oligodeoxyribonucleotide, and VEGF mRNA stability was analyzed in an 8-h hypoxia assay. Actinomycin D (5 g/ml) was periodically added beginning at 6 h hypoxia and incubated for 2, 1, or 0.5 h, respectively, to triplicate plates. Total mRNA was extracted, and VEGF, GLUT-1, and 36B4 mRNA levels was detected by Northern blot. A, representative Northern blot analysis of VEGF mRNA level after actinomycin D treatment. B, VEGF mRNA decay curves and mRNA half-life (t[itinf,1,2]). VEGF mRNA signal was normalized with the ribosome-associated mRNA, 36B4, from each lane after quantification by PhosphorImager and calculated mean Ϯ S.D. from a triplicate experiment. VEGF mRNA signal without actinomycin D treatment was defined as 100%, and actinomycin D treated VEGF mRNA levels was calculated as percentage of decay. The asterisk indicates p Յ 0.0005 when AS1-and mock (Ϫ)-transfected cells were compared at 30-and 60-min actinomycin D treatment using Student's t test.
none of which correlate to the hypoxia-induced complexes described here. Recent studies identified that the 34-kDa rat VEGF ARE-binding protein as HuR, which interacted with a 45-base element of rat VEGF 3Ј-UTR and stabilized the mRNA under hypoxia (28). This element is also present in the human 3Ј-UTR covering nucleotides 1682-1726 3Ј to the translation stop codon, a region close to the poly(A) tail. Taken together, these studies suggest that there are a variety of mRNA-binding proteins present in cells, and the regulation of mRNA turnover probably depends on the mRNA sequences, the metabolic and activation state of the cells, and the regulation of the RNAbinding protein expression and distribution.
The hnRNPs are a family of abundant nuclear proteins that are involved in pre-mRNA processing and splicing (22)(23)(24). The hnRNP L differs from other ARE-binding hnRNPs in that hnRNP A, hnRNP C, and hnRNP D interact with the AUUUA motif and may participate in mRNA destabilization (15, 29 -32), whereas hnRNP L interacts with a unique sequence in the human and bovine VEGF 3Ј-UTR (3Ј-CACCCACCCACAUA-CAUACAU-5Ј) and probably has a predominant stabilization effect. In support of this hypothesis, hypoxia substantially increased hnRNP L in both nucleus and cytoplasmic compartments, whereas hnRNP C was dramatically decreased to undetectable levels in the cytoplasm after a 24-h hypoxic treatment. This differential regulation of hnRNPs under hypoxia, a decrease in the hnRNP C and an increase in the hnRNP L in cytoplasm, may define a key balance regulating mRNA metabolism during metabolic stress. The 21-base-long mRNA binding site of hnRNP L is juxtaposed to the only AUUUA pentanucleotide and a long stretch of AU residues. Therefore, it is possible that the interaction of hnRNP L may block other proteins such as hnRNP C from interacting with and destabilizing the mRNA. Furthermore, under hypoxia we observe increased expression of hnRNP L and the specific interaction of hnRNP L with VEGF mRNA in vivo, along with reduced VEGF mRNA stability observed with transfection of antisense oligodeoxyribonucleotide to the hnRNP L binding site. Thus, hnRNP L appears to plays a significant role in the post-transcriptional regulation of VEGF mRNA during hypoxic stress.
Using Western blot analysis, anti-hnRNP L monoclonal antibody identified three proteins with apparent molecular masses of 66, 60, and 56 kDa, with the cytoplasmic 56-kDa protein being the most strongly up-regulated by hypoxia. Since there was only one RNA-hnRNP L complex formed in RNA-UVXL, it is unlikely that all three hnRNP L isoforms interacted with HSR. Interaction of hnRNP L with the entire 21base binding site would account for approximately 6 kDa of the 60-kDa HSR-hnRNP L complex after UV-cross-linking and RNase treatment. Thus, it is reasonable to predict that the 56-kDa hnRNP L isoform is the molecule that is specifically interacting with HSR. Comparing the minor induction of hnRNP L levels in hypoxia by Western blot to the marked increase of hnRNP L and HSR interaction in RNA-UVXL assays leads us to speculate that post-translational modification of hnRNP L isoforms may promote a higher hnRNP L binding affinity to HSR under hypoxia, an interesting possibility we are currently investigating.
To effectively define the complex cellular mechanisms that control mRNA stability such as VEGF, identification of the mRNA-binding proteins is required. Purification and identification of hnRNP L protein as one of the VEGF mRNA-binding proteins is a significant step in defining the molecular regula-tion of VEGF mRNA expression under hypoxic conditions. Here we have detailed the characterization of hnRNP L as one of the hypoxia-inducible RNA-binding proteins that recognizes a specific region of the human VEGF HSR. The potential signaling mechanisms that regulate hnRNP L isoform expression and association with VEGF mRNA can now be investigated in detail. Since a complex of at least seven proteins bind to VEGF HSR, interaction of hnRNP L with other proteins in the same complex may be required to protect the VEGF mRNA from nuclease digestion. Interference of hnRNP L binding to VEGF HSR significantly affected VEGF mRNA expression in hypoxia, indicating that it plays an important and novel role in VEGF expression under hypoxic stress. Identification of the other mRNA-binding proteins that recognize VEGF HSR, how they are regulated, and their potential interaction with hnRNP L will be necessary to get a complete understanding of the molecular mechanisms resulting in hypoxia-mediated VEGF mRNA stability.