5 (cid:1) Stem-Loop of Collagen (cid:1) 1(I) mRNA Inhibits Translation in Vitro but Is Required for Triple Helical Collagen Synthesis in Vivo*

The 5 (cid:1) stem-loop is a conserved sequence element found around the translation initiation site of three collagen mRNAs, (cid:1) 1(I), (cid:1) 2(I), and (cid:1) 1(III). We show here that the 5 (cid:1) stem-loop of collagen (cid:1) 1(I) mRNA is inhibitory to translation in vitro . The sequence 5 (cid:1) to the translation initiation codon, as a part of the 5 (cid:1) stem-loop, is also not efficient in initiating translation under competitive conditions. This suggests that collagen (cid:1) 1(I) mRNA may not be a good substrate for translation. Since the 5 (cid:1) stem-loop binds protein factors in collagen-producing cells, this binding may regulate its translation in vivo . We studied in vivo translation of collagen (cid:1) 1(I) mRNA after transfecting collagen (cid:1) 1(I) genes with and without the 5 (cid:1) stem-loop into Mov 13 fibroblasts. The mRNA with the (cid:1) 1(I) 5 (cid:1) stem-loop was translated into pepsin-resistant collagen, which was secreted into the cellular medium. This mRNA also produced more disulfide-bonded high molecular weight collagen found intracellularly. The mRNA in which the 5 (cid:1) stem-loop was

Three fibrillar collagen mRNAs, ␣1(I), ␣2(I), and ␣1(III), are coordinately regulated in fibrotic processes of various organs (1,2). In the 5Ј-UTR 1 of these mRNAs there is a stem loop structure (5Ј stem-loop) encompassing the translation initiation codon (3). The 5Ј stem-loop is located about 75 nt from the cap and has a stability of ⌬G ϭ 25-30 kcal/mol. The 5Ј stemloop is well conserved in evolution, differing by only two nucle-otides in Xenopus and human collagen mRNAs (4). The sequence flanking the 5Ј stem-loop is not conserved. A similar stem-loop structure is also found around the translation start codon of the sea urchin collagen gene (5). Evolutionary conservation of this sequence suggests an important function.
We analyzed previously a regulatory role of the 5Ј stem-loop in two experimental systems; quiescent versus activated hepatic stellate cells (HSCs) (6) and fibroblasts cultured in a three-dimensional matrix (7). Activated HSCs are responsible for excessive collagen production in liver fibrosis (8,9). We found that the 5Ј stem-loop prevented expression of the reporter genes in quiescent HSCs, which express low amounts of type I collagen, but allowed for expression in activated HSCs. This inhibitory effect of the 5Ј stem-loop was in part mediated by a decreased half-life of the corresponding mRNAs. Reporter genes with the mutated 5Ј stem-loop were constitutively expressed to a high level in both cell types. Therefore, expression of the reporter mRNA with the 5Ј stem-loop resembles expression of endogenous collagen ␣1(I) mRNA in HSCs; it is low in quiescent HSCs and elevated in activated HSCs and regulated by a post-transcriptional mechanism (6).
Second, we studied the role of the 5Ј stem-loop on collagen ␣1(I) mRNA expression in fibroblasts cultured in a three-dimensional matrix where the fibroblasts revert from an activated phenotype to a more quiescent phenotype. This is accompanied by destabilization of endogenous collagen ␣1(I) mRNA (10). The reporter collagen ␣1(I) mRNA with the intact 5Ј stem-loop was less stable than the identical mRNA with the mutated 5Ј stem-loop in the cells grown in the matrix. Thus, the 5Ј stem-loop is required for accelerated decay of collagen ␣1(I) mRNA in cells that down-regulate collagen synthesis (7).
The mechanism by which the 5Ј stem-loop targets mRNAs for turnover in HSCs and fibroblasts grown in a three-dimensional matrix is unknown. In quiescent HSCs we could not detect any protein binding to the 5Ј stem-loop in vitro. In activated HSCs a cytosolic protein factor(s) of unknown identity binds to the stem-loop and requires a 7mG cap on the RNA for binding (6). An excess of cap analogue completely prevents formation of this complex in vitro. The complex is also found in fibroblasts in postpolysomal cytoplasmic fraction. Its binding is greatly reduced if the cells are cultured in a three-dimensional matrix (7). One possibility is that the 5Ј stem-loop binding activity may increase the steady-state level of collagen mRNAs by diverting them from the degradative pathway (11).
Translation and mRNA decay are coupled processes (12). Therefore, studies on the translation of collagen ␣1(I) mRNA are required to provide insight into the mechanism of stabilization of this mRNA. All collagen ␣1(I) mRNA is associated with membrane-bound polysomes and is not found on free polysomes or in postpolysomal supernatant. 2 It is not known if this association is because of targeting of the mRNA or targeting by the leader peptide after initiation of translation. Collagens are secreted proteins, and their translation is coupled to export of the peptides into the endoplasmic reticulum (13). There is substantial evidence that all three peptides initiate folding into the heterotrimer while still associated with polysomes on the endoplasmic reticulum (14 -17). When folding is initiated, the collagen trimer is released in the lumen of the endoplasmic reticulum. In human disease osteogenesis imperfecta (OI) certain mutations of ␣1(I) chain decrease the rate of assembly of collagen type I. Unassembled OI ␣1(I) chains are hypermodified on proline and lysine residues and degraded (18,19). This suggests that modification and assembly processes are in a kinetic equilibrium. It is possible that the 5Ј stem-loop binding activity may target collagen mRNAs to translation at such sites.
In this study we investigate the role of the 5Ј stem-loop in translation of collagen ␣1(I) mRNA in vitro and in vivo. We found that the 5Ј stem-loop inhibits translation in vitro and when more than one mRNA is competing for a limited amount of translational apparatus. In fibroblasts in vivo, the 5Ј stemloop is necessary for efficient folding or synthesis of stable triple helical collagen. To our knowledge this is the first example of an RNA element that affects protein folding.

MATERIALS AND METHODS
Constructs-Plasmid used for in vitro transcription of PSII mRNA was constructed by cloning of the double-stranded oligonucleotide with the sequence shown in Fig. 1B into HindIII and NarI sites of the pGL3 vector (Promega). Then, the double-stranded oligonucleotide with the sequence of the T7 promoter was cloned into BglII-HindIII sites of the above construct. This plasmid was linearized with HpaI and transcribed in vitro with T7 polymerase and Cap-scribe kit (Roche Molecular Biochemicals) to produce capped PSII mRNA. COLL-START and COLL-OPTSTART constructs were made by cloning double-stranded oligonucleotides with the sequence shown in Fig. 1B into SacI-NarI sites of pGL3 and recloning of the EcoRI-HpaI fragment into EcoRI-SmaI sites of the pGEM3 vector (Promega). After linearization with BamHI corresponding capped mRNAs were synthesized with T7 polymerase and the Cap-scribe kit. Plasmids for synthesis of the 5Ј WT-SL and 5Ј MUT-SL mRNAs were made by cloning of double-stranded oligonucleotides with the sequence shown in Fig. 2A into the HindIII-NarI site of the pGL3 vector and cloning of the T7 promoter as for PSII. Capped mRNAs were made as described for PSII mRNA. The clone producing competitor A mRNA was made by cloning 75 codons of an artificial protein into EcoRI-XhoI sites of the vector pCDNA3 (Stratagene), followed by cloning of an optimal translation start site (Fig. 1B) to allow expression. This plasmid was linearized with NotI, and competitor A mRNA was made as for PSII mRNA. All in vitro produced mRNAs were gel-purified and analyzed by agarose gel electrophoresis. ␤-globin mRNA and tobacco mosaic virus (TMV) mRNA were prepared from the Roche Molecular Biochemicals in vitro translation kit.
The 5Ј WT-MH-COLL gene was made by cloning the BglII-XbaI fragment of mouse genomic DNA clone containing 220 nt of the promoter and 115 nt of the 5Ј-UTR (a kind gift from Dr. M. Breindl) into BglII-XbaI sites of the pGL3 vector and inserting into the above construct the XbaI-BamHI fragment of human collagen ␣1(I) cDNA (a kind gift from F. Ramirez). This restores the 5Ј stem-loop, which is identical in mouse and human collagen ␣1(I) mRNA and includes the entire coding region and 3Ј-UTR of human collagen ␣1(I) cDNA. The 5Ј MUT-MH-COLL gene was made identically except that a substitution of 18 nt, shown in Fig. 2A, was introduced into the mouse genomic clone before reconstituting the full-size construct.
A riboprobe for analyzing expression from MH-COLL genes was made by cloning the XbaI-KpnI fragment of human cDNA clone into XbaI-KpnI of Bluscript SK vector (Invitrogen). This plasmid was linearized by NotI and transcribed by T7 polymerase in the presence of [ 32 P]UTP as described (20).
In Vitro Translation Reactions-0.08 pmol of gel-purified mRNAs was translated in a 50-l reaction using nuclease-treated rabbit reticu-locyte lysate (Roche Molecular Biochemicals), according to the manufacturer's instructions. In preliminary experiments 0.08 pmol of mRNA was found to be a nonsaturation concentration of mRNA for a 50-l reaction. Competitor mRNA was added in 10-fold molar excess (0.8 pmol) to the test mRNA prior to mixing with the lysate. Reactions were incubated for 30 min at room temperature when a 5-l aliquot was analyzed for luciferase activity. Incubations longer than 30 min did not further increase luciferase activity. For the same test mRNA preparation the reaction was done with and without competitor, and the ratio of luciferase activity was normalized to that of PSII mRNA. All experiments were done with two different mRNA preparations, each done in duplicate.
Transfection of Mov 13 Fibroblasts-Transient transfections were done with the calcium phosphate technique using 10 g of corresponding MH-COLL plasmids. 24 h after transfection, equal number of cells were split into two dishes and incubation continued for an additional 24 h. The cells were then incubated in 0.2% serum for 24 h and either treated with 4 ng/ml of TGF␤1 (R&D Systems) or left untreated for an additional 24 h. Cells and cellular medium were collected and analyzed by RNase protection assay or Western blot. Stably transfected Mov 13 fibroblasts were developed by transfection of MH-COLL genes and pCDNA3 vector in a ratio of 10:1 and selection with G418 for 3 weeks. G418-resistant cells were pooled and processed as above.
Western Blots-50 g of cellular proteins were run on 7.5% SDS-PAGE gels under reducing or nonreducing conditions, as indicated. 100 ng of purified rat tail collagen type I (Collaborative Biomedical Products) was included as control. After transfer, the blots were probed with 1:1000 dilution of anti-collagen type I antibody (600-401-103, Rockland) and developed using the ECL system (Amersham Biosciences). Cellular medium was concentrated on Centricon 100 columns (Amicon), and equivalent amounts (corresponding to 4 ϫ 10 5 cells) were analyzed by Western blot as above. For pepsin digestions, 40 l of concentrated medium was adjusted to pH 2.5 with acetic acid and digested with 1 l of 64,000 units/ml of pepsin (Sigma) for 30 min at room temperature. After neutralization, the samples were analyzed by Western blot. For collagenase digestion, 1 l of 4 units/ml of bacterial collagenase (Roche Molecular Biochemicals) was added to 40 l of concentrated medium and digested for 30 min at room temperature.
RNase Protection Assay-50 g of total cell RNA was simultaneously hybridized with collagen-specific riboprobe and glyceraldehyde-3-phosphate dehydrogenase-specific riboprobe (Ambion), as previously described (20). The collagen-specific band has an expected size of 145 nt, and the glyceraldehyde-3-phosphate dehydrogenase-specific band has an expected size of 120 nt. Fig. 1A shows the sequence of the 5Ј stem-loop of mouse collagen ␣1(I) mRNA. To the right is shown the consensus 5Ј stem-loop sequence, which can be derived from ␣1(I), ␣2(I), and ␣1(III) mRNAs of evolutionary distant organisms (ranging from fish to humans). Since the sequence around the collagen ␣1(I) mRNA start codon, as a part of the 5Ј stem-loop, does not match the consensus sequence derived by Kozak (22,23), one set of mRNAs was constructed to investigate how the sequence of collagen ␣1(I) mRNA immediately 5Ј to the start codon affects translation. Therefore, we constructed a reporter mRNA containing only the last 25 nt of the mouse collagen ␣1(I) 5Ј stem-loop linked in-frame with a luciferase mRNA (COLL START, Fig. 1B). In this construct the 5Ј stem-loop cannot form because its 5Ј-region was deleted, but it contains the collagen ␣1(I) start codon in its natural sequence context. Control mRNAs had a short 5Ј-UTR of 35-36 nt without any structural elements or short upstream open reading frames (uORF) (PSII and COLL-OPTSTART, Fig. 1B). The COLL-OPTSTART differs from the COLL-START mRNA by 9 nt preceding the start codon, which were optimized to conform to the Kozak rules, while in COLL-START they were from mouse collagen ␣1(I) mRNA. PSII mRNA had a 5Ј-UTR derived from the pGL3 vector, which is optimized for efficient translation and was used as control. 3Ј to the start codon all constructs had the rest of the sequence of the 5Ј stem-loop (underlined in Fig.  1B), followed by the luciferase ORF. The mRNAs were made in vitro with 7mG cap, and their integrity was analyzed by agarose gel electrophoresis (Fig. 1C). These mRNAs did not contain a poly(A) tail, because of the small effect that the poly(A) tail has on translation in vitro (21). These test mRNAs were translated in rabbit reticulocyte lysate with or without of 10-fold molar excess of a competitor mRNA. The competitor mRNA had the optimal start site followed by an ORF of 75 amino acids (competitor A). The sequence of the 5Ј-UTR of competitor A is shown in Fig. 1B (COMP A). Without competitor A all three test mRNAs yielded similar amounts of the luciferase protein.

Translation of Reporter mRNAs with Collagen 5Ј-UTR Sequences-
However, in the presence of a 10-fold amount of competitor A the efficiency of translation was reduced 5-fold for PSII mRNA, 19-fold for COLL-START mRNA, and 8.4-fold for COLL-OPT-START mRNA. In Fig. 1D this result is shown normalized to the inhibition of PSII mRNA. Because the highest inhibition was observed when the collagen ␣1(I) sequence preceded the start codon (COLL-START), we concluded that this sequence is suboptimal in promoting translation initiation when competing with another mRNA for the translation machinery.
Another set of reporter mRNAs was designed to address the role of the 5Ј stem-loop in translation. 5Ј WT-SL reporter contained 63 nt of the mouse collagen ␣1(I) 5Ј-UTR with the 5Ј stem-loop in-frame with LUC, while the 5Ј MUT-SL reporter had substitutions in the 5Ј stem-loop to abolish its formation ( Fig. 2A). Translation of these reporters was compared in vitro to the PSII mRNA (described above) without competitor mRNA or under competitive conditions. The integrity of the mRNAs is shown in Fig. 2B. Without competitor, the 5Ј WT-SL mRNA was translated about 3-fold less efficiently than PSII (arbitrarily set as 1) and 5Ј SL-MUT mRNAs. The latter two were translated with comparative efficiency (Fig. 2C). This was not due to preferential degradation of the 5Ј WT-SL mRNA in the lysate, because extraction of the RNAs from the lysate after a 1-h incubation and retranslation in fresh lysate yielded the same result (not shown).
Next we compared translational efficiency of the 5Ј WT-SL mRNA and 5Ј MUT-SL mRNA to PSII mRNA in the presence of 10-fold molar excess of competitor mRNAs. Three competitor mRNAs were added in 10-fold molar excess to the reaction, competitor A (described above), ␤-globin mRNA, and TMV. With competitor A 5Ј WT-SL reporter mRNA was translated about 30-fold less efficiently, when compared with PSII mRNA. 5Ј MUT-SL mRNA was translated only 5.5-fold less efficiently relative to PSII mRNA (Fig. 2D). When ␤-globin mRNA was used as competitor the respective ratios were 4.6-and 3.6-fold (Fig. 2E). When viral RNA was used as a competitor the translation of 5Ј WT-SL and PSII mRNAs was not affected, while translation of the 5ЈMUT-SL mRNA was increased 2-fold (Fig.  2F). Based on the results in Fig. 2, we concluded that the 5Ј stem-loop is inhibitory for translation in vitro in the absence of a competitor mRNA, while various competitor mRNAs have either a strong inhibitory effect or show only a small effect. We analyzed if reticulocyte lysates contain the 5Ј stem-loop binding activity by performing gel mobility shift analysis using capped 5Ј stem-loop RNA or inverted 5Ј stem-loop RNA as probes, as described (6). We could not detect specific binding to the 5Ј stem-loop (data not shown).
Expression of Collagen ␣1(I) Reporter Genes in Mov 13 Fibroblasts-Mov 13 fibroblasts were derived from mice in which insertion of a retrovirus into the first intron of collagen ␣1(I) gene had inactivated the transcription of this gene (24). Mov 13 activity was determined. The ratio of luciferase activity with and without competitor mRNA was calculated and arbitrarily set as 1 for PSII mRNA. The result shown is from two independent experiments each performed in duplicate. The error bar Ϯ S.D. is shown. fibroblasts transcribe the ␣2(I) gene and thus provide a unique opportunity to study translation of collagen type I mRNAs and assembly of the collagen trimer when the ␣1(I) mRNA is encoded by various transgenes. Fig. 3 shows characterization of Mov 13 fibroblasts. No collagen ␣1(I) polypeptides can be detected by Western blot among cellular proteins of Mov 13 fibroblasts (Fig. 3A, lane 1). For comparison, cellular proteins of NIH 3T3 fibroblasts and purified collagen from rat tail were analyzed in lanes 3 and 2, respectively. Pro-␣1(I) (about 175 kDa) and ␣1(I) (about 120 kDa) peptides were seen in the NIH 3T3 sample. The antibody used did not detect the ␣2(I) chain. The ␣1(I) monomer and higher molecular weight cross-links of type I collagen are seen in the rat tail sample, which served as markers.
To assess the role of the 5Ј stem-loop in collagen type I synthesis in vivo we constructed two genes. One gene contained 220 nt of the promoter of the mouse collagen ␣1(I) gene followed by the mouse collagen 5Ј-UTR including the 5Ј stem-loop, ligated to the full-size human collagen ␣1(I) cDNA (5Ј WT-MH-COLL, Fig. 3B). The gene has an open reading frame encoding a full-size human collagen pro-␣1(I) polypeptide. The other gene is identical, except it has an 18-nt mutation within the 5Ј stem-loop, which destroys its formation (5Ј MUT-MH-COLL). This mutation does not affect the coding region of the gene, it encodes for the identical polypeptide as the 5Ј WT-MH-COLL gene. The genes were transiently transfected into Mov 13 fibroblasts and mRNA analyzed by RNase protection assay (  (Fig. 4A). We also treated the cells with 4 ng/ml of active TGF␤1, to assess how this profibrogenic cytokine (25,26) would affect the expression. The blot was done under nonreducing conditions to assess the synthesis of disulfide-bonded collagen species. The major collagen detected was the pro-␣1(I) chain, and both genes synthesized a similar level of the peptide. Its steady-state level was unaffected by TGF␤. However, the 5Ј WT-MH-COLL gene yielded some of the disulfide-linked higher molecular weight collagen species (HMW COLL), while the 5Ј MUT-MH-COLL gene did not (Fig. 4, compare lanes 1 and 2 to lanes 3 and 4). We could not distinguish, with our antibody, whether these species were homo or hetero multimers of type I collagen, although these collagen moieties comigrated with the collagen species found in rat tail type I collagen (lane 5). The result suggested that the 5Ј stem-loop, although an RNA element, is involved in more efficient formation of disulfide-bonded collagen monomers, suggesting better registration of collagen chains. This prompted us to investigate if the collagen synthesized by 5Ј  WT-MH-COLL mRNA is more efficiently secreted into the cellular medium.
We collected the cell medium and analyzed the equivalent amounts by Western blot under reducing conditions. Very little of pro-␣1(I) polypeptide (175 kDa) was secreted out of nonstimulated cells and in the same amount for 5Ј WT-MH-COLL and 5Ј MUT-MH-COLL genes (Fig. 4B, lanes 2 and 4). Only tracing amounts were processed to mature ␣1(I) chain of about 120 kDa. With TGF␤1 stimulation a much higher amount of pro-␣1(I) chain was found in the cellular medium, but both genes produced similar amount of the protein (lanes 3 and 5).
Since there was no change in mRNA level with TGF␤1 stimulation (not shown), we concluded that TGF␤1-stimulated translation or secretion of type I collagen independent of the 5Ј stem-loop. Alternatively, TGF␤ may have decreased extracellular degradation of collagen. In stably transfected Mov 13 fibroblasts TGF␤1 also stimulated extracellular accumulation of collagen, but no difference between the 5Ј WT-MH-COLL and 5Ј MUT-MH-COLL genes was seen (Fig. 4C).
Only the 5Ј WT-MH-COLL Gene Expresses Properly Folded Collagen-Next, we probed the structure of secreted collagen from Mov 13 fibroblasts by digestion of cellular medium with pepsin and collagenase. We used the medium of cells treated with TGF␤1, because it contained the higher amount of collagen that facilitated the analysis. The medium was subjected to digestion with pepsin and collagenase as described under "Material and Methods" and analyzed by Western blot. In undigested medium under reducing conditions, the predominant collagen species was the pro-␣1(I) chain (Fig. 5A, lane 1). When the medium of cells expressing the 5Ј WT-MH-COLL gene was digested with pepsin the molecular mass of this chain was reduced to about 120 kDa (lane 2). This suggests cleavage of the globular domains, but folding of the core domain into pepsinresistant triple helix. When the medium was digested with bacterial collagenase, no collagen peptides could be detected, suggesting the specificity of the bands (lane 3). Thus, mRNA with the 5Ј stem-loop directs synthesis of triple helical collagen, which accumulated in the cell medium. When the medium of 5Ј MUT-MH-COLL-expressing cells was digested with pepsin no pepsin-resistant fragment was obtained (lane 5), although the amount of collagen secreted was comparable to that of the 5Ј WT-MN-COLL gene (lane 4). Thus, mRNA without the 5Ј stemloop directs synthesis of structurally aberrant collagen that could not resist limited pepsin digestion. Digestion with collagenase served as the specificity control (lane 6). The result obtained with the medium from stably transfected Mov 13 cells is shown in Fig. 5B. Again, the 5Ј WT-MH-COLL gene yielded secreted triple helical collagen (lane 2), while the 5Ј MUT-MH-COLL gene produced a pepsin-sensitive collagen (lane 5). We concluded from these experiments that the 5Ј stem-loop is necessary for productive collagen synthesis and, although it was mutated without affecting the coding region of the mRNA, has a profound effect on collagen protein folding or stability. DISSCUSSION Unique features of the three fibrillar collagen mRNAs, ␣1(I) mRNA, ␣2(I) mRNA, and ␣1(III) mRNA, is the 5Ј stem-loop structure that encompasses the start codon (3). The 5Ј stemloop has an important role in regulating ␣1(I) mRNA stability (3,6,7) and is conserved in collagen mRNAs of evolutionary distant species (4,5). Since the start codon is part of this stem-loop, the sequence constraints required to maintain the 5Ј stem-loop dictate the sequence around translation initiation. Therefore, the start codon in collagen mRNAs is not in the sequence context necessary for optimal translation initiation (22,23). We have shown here that it is not efficiently recognized in vitro if more than one mRNA species are competing for the translation machinery. When the start codon was optimized we could increase translation 4-fold under competitive conditions (Fig. 1C). The 5Ј stem-loop structure has a stability of 25-30 kcal/mol. This is insufficient to block scanning ribosomes to reach the start codon, because stem-loop structures of about 70 kcal/mol are needed (27,28). Nevertheless, reporter mRNA with the 5Ј stem-loop (5Ј WT-SL) was translated 3-fold less efficiently, even in the absence of competitor, than similar mRNA in which the stem-loop was mutated (5Ј MUT-SL, Fig.  2C). It seems that the structure of the 5Ј stem-loop together with its suboptimal translation start site is responsible for this effect.
Various competitor mRNAs inhibit in vitro translation of a reporter mRNA with collagen ␣1(I) 5Ј stem-loop (5ЈWT-SL) to a different degree. Competitor A inhibited translation of 5Ј WT-SL mRNA 30-fold relative to a control mRNA (PSII) and 6-fold relative to the identical mRNA without the stem-loop (5Ј MUT-SL) (Fig. 2D). ␤-globin mRNA had a smaller effect. Viral mRNA showed no inhibition on 5Ј WT-SL mRNA, while 5Ј MUT-SL mRNA is translated better. The reason for this is unclear, but TMV mRNA with its Omega sequence may titrate an inhibitor of translation (29). The result with competitors corroborates the finding that the 5Ј stem-loop may compromise translation of collagen ␣1(I) mRNA, which may be a weak substrate for translation in the absence of its RNA-binding proteins. Since the 5Ј stem-loop binds protein factors in collagen-producing cells, it is possible that binding of these factors regulates translation of collagen ␣1(I) mRNA. Cloning of these proteins and their addition to the in vitro translation reaction will address this question.
Previous reports suggested that N-and C-terminal peptides of type I collagen inhibit translation of collagen ␣1(I) mRNA (30,31). We did not see any inhibition when these recombinant peptides were added to the in vitro translation reaction (data not shown). Also, the ␣1(I), ␣2(I), and ␣1(III) collagen mRNAs contain two short uORF preceding the start codon. We did not see any change in translation of our reporter mRNAs in vitro when these uORF were abolished (data not shown). Although it is known that uORFs can regulate translation in yeast (32,33), there are only a few examples of their role in translational regulation in higher organisms (34 -36). Based on electron microscopy data, assembly of the collagen type I heterotrimer occurs on the membrane of the endoplasmic reticulum, while the individual chains are still associated with polysomes or shortly after their release (14,17). Lysyl hydroxylase, one of the key enzymes in collagen modifications, is also associated with the membrane of the endoplasmic reticulum (37). Membrane association may couple folding starting from the C terminus of collagen chains, to concomitant modifications of the selected lysine residues. For collagen type I, this implies that ␣1(I) and ␣2(I) chains may be synthesized by ribosomes positioned in close proximity on the endoplasmic reticulum membrane. Such coordinated translation would greatly increase local concentration of the chains. We hypothesized that the sequence elements that modulate loading of ribosomes on collagen ␣1(I) mRNA may be involved in targeting for such coordinated translation. Therefore, we mutated the 5Ј stemloop and analyzed production of collagen trimers in vivo from the hybrid mouse-human collagen genes (5Ј WT-MH-COLL and 5Ј MUT-MH-COLL) (Fig. 5). The human collagen ␣1(I) gene could rescue the phenotype of Mov 13 mice, proving that human collagen ␣1(I) polypeptide is functional in mouse (38,39). The 5Ј WT-MH-COLL gene produced triple helical collagen in Mov 13 fibroblasts; however, the 5Ј MUT-MH-COLL gene produced collagen, which was sensitive to digestion with pepsin (Fig. 5). The structurally aberrant collagen was produced although the 5Ј MUT-MH-COLL gene had the identical coding region. This pepsin-sensitive collagen may represent individual ␣1(I) chains, which were not efficiently folded into triple helix and were secreted as monomers, or alternatively, the monomers were not properly modified and an unstable triple helix was secreted. The 5Ј MUT-MH-COLL chains had identical electrophoretic mobility to the 5Ј WT-MH-COLL chains, excluding a major difference in post-translational modification, although subtle differences may remain undetected. If the chains were not efficiently folded, the 5Ј stem-loop may be required to increase their local concentration, which would facilitate the chain registration. The result shown in Fig. 4 where the 5Ј WT-MH-COLL gene produced more disulfidelinked high molecular weight collagen suggests that this may be the case. If the chains were not properly modified, the 5Ј stem-loop may target collagen mRNAs for translation to discrete regions of the endoplasmic reticulum where there is optimal concentration of collagen-specific modifying enzymes and molecular chaperones (40). Although we do not have direct evidence for this, we think that the 5Ј stem-loop also couples the translational machinery to the rest of the collagen biosynthetic pathway, because collagen biosynthesis requires coordinate action of translational apparatus, modifying enzymes, and molecular chaperones (41). In patients with OI, where folding of collagen type I chains is impaired, the mutant chains are hypermodified and subjected to degradation (19,42). Interestingly, one patient with OI type I was described who had a mutation in the 5Ј stem-loop in the absence of any other mutation of the collagen ␣1(I) gene (43).
There are many examples that mRNAs are targeted for translation at discrete subcellular sites to produce proteins with the concentration gradient within the cell. Most targeting signals are located in the 3Ј-UTR of these mRNAs (44). To our knowledge this is the first example that a RNA element located in the 5Ј-UTR is involved in synthesis of a secreted multisub-unit protein. This study demonstrates that the conserved collagen 5Ј stem-loop has specific functions. In the absence of RNA-binding proteins, the 5Ј stem-loop renders the collagen mRNAs inefficient for translation and therefore susceptible to regulation, such as by TGF␤. In collagen-producing cells, the 5Ј stem-loop has a novel function of directing the post-translational modification of collagen to produce mature triple helices. The 5Ј stem-loop almost certainly acts through its cognate RNA-binding proteins. Cloning of these protein factors will help us elucidate the complex biosynthesis of type I collagen.