An Erythroid-Specific Transcript Generates the Soluble Form of NADH-Cytochrome b₅ Reductase in Humans

The human erythroleukemia cell line K562 and the human rhabdomyosarcoma cell line TE671 were grown in RPMI 1640 medium, supplemented with heat-inactivated 10% fetal calf serum (FCS) in a 5% CO₂incubator. K562 cells were induced to differentiate by incubation in medium supplemented with 20 μmol/L hemin (Sigma Chemical Co, St Louis, MO) for 4 days. Induction of globin synthesis was assessed with the benzidine reaction. Benzidine positive cells were generally ≈70% of the total cell population, in comparison to 5% to 10% in untreated samples.

HeLa cells were grown in a 10% CO₂ incubator in Dulbecco's modified Eagle's medium (DMEM), supplemented with 1 mmol/L sodium pyruvate and with 10% heat inactivated FCS.

Erythroid cell cultures derived from normal human peripheral blood were set up as described by Fibach et al,17 but 10 ng/mL stem cell factor (Genzyme, Cambridge, MA) were included in the erythropoietin-dependent second phase of the culture. Human recombinant erythropoietin was from Cylag (Sulzbach, Germany). After 19 to 20 days of culture, cells were harvested and counted. Cell viability was determined by Trypan blue exclusion; cellular morphology and Hb-containing cells were assessed on cytocentrifuge slides stained with May Grunwald's Giemsa and benzidine-HCl.18 The erythroblast RNA used for the experiments in Figs 2 and 3 of this report were from a terminal culture containing 72% of erythroid cells (of which ≈70% were orthochromatic erythroblasts and ≈30% were polychromatophilic cells), with the nonerythroid component due to leukocytes or undifferentiated cells.

Heparinized blood, obtained from anemic patients, was passed through a microcrystalline cellulose-α–cellulose (Sigma) column to remove leukocytes.19 The percentage of reticulocytes in the final preparation determined by staining with brilliant cresyl blue was ≈15%. Leukocytes were undetectable.

Total RNA was isolated from cultured cells using the Total RNA Fast ΙΙ Isolation Kit (Molecular Systems, San Diego, CA). Poly A⁺ RNA was obtained from the total RNA using the Oligotex mRNA Kit from Qiagen GmbH (Hilden, Germany). To extract total RNA from reticulocytes, the reticulocyte-enriched red blood cell preparation described above was homogenized with a solution containing 4 mol/L guanidinium thiocyanate, 25 mmol/L sodium citrate pH 7, 0.5% sarcosyl, 0.1 mol/L 2-mercaptoethanol (2 mL of solution per 10⁹cells). The Chomczynski-Sacchi protocol20was followed until the first isopropanol precipitation. The resulting pellet was resuspended with the homogenization buffer of the Total RNA Fast ΙΙ Isolation Kit and processed further according to the instructions of the manufacturer. Approximately 50 μg total RNA/14 × 10⁹ erythrocytes were obtained.

To obtain the 5′ extremities of M and S cDNAs, 250 ng of poly A⁺ RNA obtained from TE671 cells was subjected to RT-PCR using the First Strand Synthesis Kit from Stratagene (La Jolla, CA) and recombinant Moloney murine leukemia virus Reverse Transcriptase supplied by the kit. All amplification reactions were performed in a DNA thermal cycler from Perkin Elmer-Cetus (Norwalk, CT).

For the 5′ extremity of M cDNA, the following oligonucleotides were used (underlined nucleotides correspond to extra sequences containing restriction sites): upper primer 5′-GGGAATTCCGGCGACAGAGCGAGCG-3′ (nts -32 to -16 in the 5′ UT of 1M of the b5R gene sequence in Tomatsu et al¹⁴; lower primer: 5′-CCCTCTA GATGGCGGCAGGGCAAAGC-3′ contained in exon 3 (nts 17671 to 17655 in Tomatsu et al14). After 30 cycles (1 minute at 94°C, 1 minute at 60°C, and 2 minutes at 72°C), a fragment of 230 nucleotides was obtained as expected.

To obtain the 5′ extremity of S cDNA, two consecutive amplifications were required. For the first amplification, the upper primer was (underlined nucleotides as above): 5′-CCCAAGCT TAAAACATTGGAGCTTGG-3′, corresponding to nts 2189-2210 (or 2761-2782 by the numbering of Du et al15) in the 1S exon; the lower primer was the same one used to obtain M cDNA. After 30 cycles of amplification (1 minute at 94°C, 1 minute at 54°C, and 2 minutes at 72°C) and electrophoretic analysis of the samples, no fragment of the expected size (270 nts) could be detected by ethidium bromide staining. The entire PCR sample was loaded on a preparative agarose gel, the region corresponding to 200-300 nts was excised, and the DNA was eluted. All of the eluted material was subjected to a second round of amplification, in which the upper primer of the first extension was maintained, but a nested lower primer was used: 5′-CTTGATGTCGGGGTTCTCGAG-3′ corresponding to nts 12364-12344 in exon 2¹⁴ and containing an internalXho 1 restriction site. At the end of this second amplification (performed under the same conditions as the first one), a 205-nt fragment was obtained as expected.

The PCR-generated M and S fragments were digested and subcloned into appropriate vectors, either to produce labeled antisense RNA probes or to reconstruct the entire M and S cDNAs from a b5R clone lacking the 5′ extremity (see below).

The 230-bp M fragment and the 205-bp S fragment were inserted, respectively, into the EcoR1-Xba 1 sites of pBluescript II KS (+) and into the Hind3-Sal 1 restriction sites of pGEM3. The plasmids, truncated at the Hind2 and Hind3 sites for M and S, respectively, were used to generate³²P-cytidine triphosphate (CTP) antisense-labeled transcripts with T7 RNA polymerase. A 316-bp fragment of human glyceraldehyde-3-phosphate dehydrogenase (GAPDH) cloned into the pTRI plasmid (Ambion Inc, Austin, TX) was used as control. The plasmid was linearized with Hind3 and generated a 404-bp fragment when transcribed with T7 RNA polymerase. Production of³²P-CTP–labeled antisense transcripts, hybridization of the probes to the total or poly A⁺ RNA from human cells, and digestion with RNase A+T1 cocktail were performed with reagents of the Maxi Script and RPA 2 kits from Ambion Inc. Amounts of probe and RNA, as well as temperatures of hybridization, are specified in the figure legends. Protected probe fragments were analyzed on 6% sequencing gels. To provide DNA molecular size markers, Msp1-digested pBR322 DNA (New England Biolabs, Beverly, MA) was labeled with ³²P-dCTP using the Klenow fragment of DNA polymerase I. Band intensities of the autoradiograms were determined with the NIH-Image program (version 1.55).

The entire coding sequences of the M- and S-transcripts were reconstructed by splicing the fragments obtained by RT-PCR to an incomplete human reductase cDNA lacking the 5′ extremity and inserted into pUC13 (b5R-pUC13) kindly provided by Dr T. Yubisui (Kochi Medical School, Japan).5 The RT-PCR–generated M fragment (see above) was excised from Bluescript with Sal 1 and Xho 1 and subcloned into b5R-pUC13. The RT-PCR–generated S fragment was subcloned into the Hind3 andXho 1 sites of b5R-pUC13. The resulting M and S constructs were subcloned into the modified mammalian expression vector pCB6.21,22 

HeLa cells were plated at ≈20% of confluence, in 100-mm Petri dishes or on 1.7 × 1.7 cm coverslips, and transfected 12 to 24 hours later by the calcium phosphate method,23 with pCB6 containing M- or S-cDNA (1 μg DNA/cm² of culture dish). Thirty-six hours after transfection cells were either fixed or collected for biochemical analyses.

Transfected HeLa cells, grown on glass coverslips, were fixed in paraformaldehyde, permeabilized, and processed for immunofluorescence as described,22 using rabbit affinity-purified antirat b5R antibodies cross-reactive with the human enzyme, followed by a fluorescein-labeled antirabbit IgG (Jackson Immunoresearch, West Grove, PA) as secondary reagent. Preparations were observed under an Axioplan microscope equipped for epifluorescence (Zeiss, Oberkochen, Germany). Mitochondria were stained with 100 nmol/L Mitotracker CMX Rose from Molecular Probes (Eugene, OR) as described.22 

Cells grown on 100-mm Petri dishes were washed free of medium with cold phosphate-buffered saline (PBS) containing 5 mmol/L EDTA, incubated in the same buffer for 5 minutes on ice, detached with a rubber policeman, and collected by centrifugation. The pellet was resuspended with cold hypotonic buffer (1 mmol/L EDTA, 15 mmol/L KCl, 30 mmol/L NaCl, 1 mmol/L Tris-HCl, pH 7.5) containing 1 mmol/L phenylmethylsulfonyl fluoride plus the cocktail of protease inhibitors described in Holt and Hart.²⁴. A total of 300 μL of buffer for every 10⁷ cells was used. After 5 minutes, cells were ruptured by passage through a cell cracker with a 0.0007-in clearance. The volume of the homogenate was measured and the same volume of cold two times concentrated isotonic buffer (0.5 mol/L sucrose, 0.2 mmol/L EDTA, 2 mmol/L Tris HCl, pH 7.4) was added. After sedimentation of nuclei, (500g for 10 minutes at 4°C), the resulting postnuclear supernatant (PNS) was centrifuged at 150,000g for 50 minutes at 4°C in a TLA 100.3 rotor (Beckman Instruments, Inc, Palo Alto, CA) to obtain a membrane pellet and a supernatant (cytosol) fraction. The pellet was resuspended with a small volume of isotonic buffer containing protease inhibitors.

Reduced and alkylated pellet and supernatant fractions were analyzed by SDS-PAGE (11% gels) followed by blotting onto nitrocellulose and radioimmunostaining as described.25 The affinity-purified polyclonal antirat b5R antibodies described above were used at 2.5 μg/mL, followed by ¹²⁵I-protein A (Amersham Life Science Ltd, Buckinghamshire, UK) at 200,000 dpm/mL.

Protein was assayed by the method of Lowry et al.26 b5R enzyme activity in cell fractions was determined by the rate of reduction of the ferrihemoglobin-ferrocyanide complex (MetHb-FeCN) according to Hegesh et al,27 slightly modified as described.28 Triton X-100, at a final concentration of 0.1%, was routinely included in the assay, a condition that allows measurement of the activity of both the membrane-bound and the soluble enzyme.28,29 

Basing ourselves on the published sequence of the b5R gene,14,15 we constructed probes for the 1M exon (the first exon of M-mRNA, which specifies the membrane-bound form of b5R) and for the recently identified 1S exon (the first exon of S-mRNA, which is thought to specify the soluble form). The probes, prepared by RT-PCR from polyA⁺ RNA purified from a human rhabdomyosarcoma cell line (TE671), were designed to contain, in addition to the first alternative exon, part of the downstream sequence common to the two transcripts (Figs 2A and3A). The M probe was obtained with 30 cycles of amplification, while for the S probe, two consecutive 30-cycle amplifications were required to obtain a product visible by ethidium bromide staining (see Materials and Methods), indicating that the expression of the 1S exon is much lower than that of the 1M exon in these cells. The sequences of three clones for each PCR-generated fragment were determined and found to be identical to the published ones,14,15 except for two differences, probably due to polymorphic replacements: (1) a silent C to G transversion at position 12,274,14 in exon 2, at position 3 of a val codon; this difference was seen also in a similar fragment obtained from HeLa cell RNA; (2) a G-T transversion in exon 1S (5′ UT region) at position 2206 (Tomatsu et al14 or position 2778 of the sequence in Du et al15).

Labeled antisense RNA probes, transcribed from the RT-PCR–generated fragments, were used in protection assays performed on RNA prepared from various human cell lines, as well as from primary erythroid cultures and from reticulocytes. The results obtained with the 1M-containing probe are shown in Fig 2. The presence of M-mRNA in the RNA samples was predicted to entirely protect the probe, with generation of a 225-nt–labeled fragment after digestion of the portions of the probe transcribed from the cloning vector (Fig 2A). In contrast, the presence of b5R mRNA containing any alternative first exon would result in only partial protection of the probe, with generation of a 174-nt–labeled fragment. As can be seen in Fig 2B, RNA from Hela cells (lane H), from K562 cells induced (lane Ki) or not induced (lanes Kc) to differentiate, as well as from TE671 cells (not shown) yielded a major species with the expected size for the fully protected probe (migrating approximately at the position of the 217 nt marker fragment). Some minor lower molecular weight fragments were also visible, including a band at approximately 174 nt with an intensity less than 10% that of the major 225 nt species, as determined by scanning analysis. This band was slightly more intense in the primary erythroid sample (lane E), however, its significance was difficult to evaluate because of the background in the lanes and because of its variability (compare the Kc lanes of the left and right panels). In any case, in these RNA samples, obtained from erythroid as well as nonerythroid cells, the major b5R mRNA was of the M type. A different situation was found for reticulocyte RNA (lane R). In this case, in addition to the fully protected species, a fragment at the position expected for the partially protected probe could clearly be seen (arrowhead). By scanning analysis, the intensity of this band was 80% that of the completely protected species. Taking into account the smaller size of the partially protected probe, this result suggests that in reticulocytes, a b5R mRNA with an alternative first exon is present in roughly equal amounts to the ubiquitous M-mRNA. It should be noted that the absolute level of expression of both b5R transcripts in reticulocytes was considerably lower than in the other tested cells; moreover, GAPDH mRNA, present at high levels in the other cells, was undetectable in the reticulocyte RNA preparation (row marked GAPDH in Fig 2B).

To determine whether the alternative transcript detected in reticulocytes contains the 1S exon, RNase protection assays were performed with the 1S probe, which is predicted to generate a 181-nt–labeled fragment when fully protected, and an 89 nt fragment if partially protected by a b5R mRNA not containing the 1S exon (Fig 3A). As can be seen in Fig 3B, by this assay, the S transcript was detectable not only in reticulocyte RNA (lanes R), but now, without the background problem, also in RNA from the terminal primary cultures of erythroid cells (lane E). The much higher ratio of M- to S-transcript in the primary cultured erythroid sample compared with reticulocytes can be explained by the presence in the cultures of nonerythroid cells (see Materials and Methods), which could contribute most of the M-transcript detected in this sample. In contrast to the situation with reticulocyte and erythroblast RNA, even prolonged exposure of the Hela (H), TE61 (T), induced and noninduced K562 (Ki and Kc) lanes failed to show any fully protected probe, indicating that levels of S transcript in these cells are below the limits of detection of the RNase protection assay. Scanning analysis of the reticulocyte lanes showed that the band intensity of the partially protected probe was 44% of that of the fully protected fragment. Again, taking into account the different size of the fully and partially protected probe, this result indicates that the two b5R transcripts are present in approximately equal amounts in reticulocytes, in agreement with the data in Fig 2.

The entire coding sequences of M- and S-cDNAs, reconstructed by splicing the RT-PCR–generated fragments described above to an incomplete b5R cDNA clone, lacking the 5′ extremity,5were subcloned into a mammalian expression vector and transfected into HeLa cells. Membrane pellets (P), and cytosol fractions (Sup), prepared from the transfected cells, were analyzed by radioimmunoblotting (Fig 4). The amounts of protein of the pellet and supernatant fractions loaded on the gel were adjusted to correspond to the same number of cells. The pellet fraction of cells transfected with the vector alone (not containing a b5R insert) contained a polypeptide with an apparent M_r of ≈35 000, corresponding to b₅R (calculated M_r = 34,232) (Fig 4B, lane 1), whereas no band at the position expected for soluble b5R (calculated M_r = 31,600) was detected in the cytosol fraction (Fig 4B, lane 2, Sup). This was true also for other cell lines that we analyzed (TE671 and K562 induced and not induced to differentiate, not shown), even with higher protein loads and prolonged exposure of the blots, consistently with the results of the RNase protection assays. A faint band comigrating with the membrane-bound form was detected after prolonged exposure (not shown), probably due to contamination of the cytosol fraction with small membrane fragments, which failed to sediment. In addition, the cytosol fraction from all cell lines analyzed contained a cross-reactive band at ≈28 kD (marked by asterisk in Fig 4B). The b5R band of the pellet fraction of cells transfected with M-cDNA was of increased intensity (Fig 4B, lane 3), while the cytosol fraction remained unaltered compared with the control (lane 4). Total protein load of the fractions of cells transfected with vector alone or with b5R-cDNA were roughly the same, as can be seen in Fig 4A. A striking result was obtained after transfection with S-cDNA: in this case, there was no increase in b5R in the pellet fraction, while a band at the position of the soluble enzyme (≈32 kD) appeared in the supernatant, representing about one half of the total b5R of the sample.

To investigate whether the transfected cDNAs generate enzymatically active b5R, we also determined enzyme activity in the cell fractions using an assay that detects the soluble, as well as the membrane-bound forms, of the enzyme.28,29 As can be seen in Fig 5, more than 90% of the enzyme activity of control cells was found in the pellet fraction. The small amount (≈8%) of activity assayed in the supernatant may be due to contaminating membrane fragments, as discussed above. The distribution of enzyme activity was unaffected by transfection of M-cDNA, whereas expression of S-cDNA resulted in a shift of the activity to the supernatant (39% of the total activity). Figure 5 also shows that rat S-cDNA, which is known to generate soluble b5R,10 when transfected into HeLa cells, yielded results indistinguishable from those obtained with the human clone (last set of columns in Fig 5). The results in Fig 5 are in good agreement with those obtained by radioimmunoblotting and demonstrate that the human S-cDNA generates enzymatically active soluble b5R.

The rat M-mRNA generates a protein that associates both with endoplasmic reticulum (ER) and with mitochondrial outer membranes (MOM),30 and there is indirect evidence that this is the case also in man.2 We directly observed the subcellular localization of the products of b5R mRNAs by immunofluorescence analysis of the transfected HeLa cells (Fig6). Cells transfected with S-cDNA and stained with antireductase antibodies (at dilutions at which the endogenous b5R is undetectable) displayed uniform fluorescence throughout the cell, as expected for a soluble protein (Fig 6A). In contrast, cells expressing the M-cDNA product showed a reticular staining pattern (Fig 6B). The distribution of the membrane-bound form of b5R was compared with that of mitochondria in double-labeling experiments, using the mitochondrial fluorescent dye Mitotracker (Fig 6C and D). b5R was clearly present on mitochondria, as demonstrated by the coincident staining of antireductase antibodies and Mitotracker (arrows in Fig 6C and D show some of the most obvious identities). In addition, however, b5R showed a more widespread reticular staining and, importantly, was present also on the nuclear envelope (arrowhead in Fig 6C), indicating an ER localization. Thus, the human membrane-bound form of b5R localizes both to the ER and to MOM, like its rat homologue.

The molecular biology of b5R is of interest both because the gene offers an attractive model for the study of the regulation of alternative exon expression during erythroid maturation and because detailed knowledge of the biogenesis of the membrane-bound and soluble forms of b5R may contribute to the understanding and eventual treatment of hereditary methemoglobinemia caused by b5R deficiency. Although the mechanism of biogenesis of the two enzyme forms from one gene had been elucidated in the rat,10 less information was available in man. In the present study, we show that the basic features of the alternative promoter mechanism governing the production of the two enzyme forms in the rat are present in man, as well, although there are also some interesting differences between the two species.

In a recent report, Du et al15 identified an alternative first exon in the human b5R gene, called 1S, which was suggested to generate soluble b5R but which, because it was detected by RT-PCR in various nonerythroid cells, appeared not to be tissue-specific. Although we also could amplify S-cDNA from nonerythroid cells, our conclusion regarding the tissue specificity of exon 1S expression is different from that of Du et al.15 Using an RNase protection assay with antisense probes designed to simultaneously detect transcripts with alternative first exons in the same RNA sample, we found that the only cells expressing appreciable proportions of the S transcript were erythroid cells at late stages of maturation, ie, reticulocytes and terminally differentiated cultured erythroblasts. Indeed, with a 1S-containing antisense probe, the S-transcript was undetectable in nonerythroid cells, as well as in hemin-treated K562 cells, used as a model for erythroid cells at early stages of differentiation.31 Conversely, the use of a 1M-containing antisense probe demonstrated that in those same cells the major expressed transcript was of the M type. It must be noted that the presence of low concentrations of transcripts with alternative first exons different from 1S could not be ruled out from the experiments with the M probe because minor amounts of partially protected probe were difficult to evaluate over the background. In the rat, in addition to the 1M and 1S exons, two additional alternative first exons, apparently expressed ubiquitously at very low levels, have been described.32 We have not investigated whether these exons have counterparts in humans.

The discrepancy between the results with RT-PCR and RNase protection can, of course, be explained by the different sensitivities of the two methods. PCR can detect as little as 1 molecule of DNA, a sensitivity well above that of the RNase protection assay. Thus, although the 1S exon appears to be transcribed at very low levels in nonerythroid cells (detectable only by RT-PCR), its expression is upregulated specifically in developing red blood cells. It should be noted that “illegitimate” expression of various tissue-specific genes has been reported to occur at levels detectable by RT-PCR.33Thus, the presence of the S-transcript in nonerythroid cultured cells is of questionable biologic significance, considering also that these same cells do not have any soluble b5R detectable by radioimmunoblotting.

In addition to RT-PCR, Du et al15 also performed RNase protection experiments and detected low levels of the S-transcript in liver by this method. Their assay was not designed to compare the relative amounts of S and M transcript in the same sample, but because liver expresses high amounts of the membrane form of b5R,10 it is likely that the S transcript detected in liver represented a minor proportion of the total b5R transcripts. In rat liver, we previously also detected some S-transcript with a semiquantitative PCR assay, whereas none could be seen in skeletal muscle, heart, brain, or kidney.10 In that study, we suggested that the S-transcript detected in liver is due to the blood component and not to hepatocytes, however, it is also possible that hepatocytes do express slightly higher levels of the 1S exon than other nonerythroid cells.

Because the 1S exon is expressed preferentially in erythroid cells, one would expect to find typical features of erythroid promoters in its 5′ flanking region. We were surprised that consensus sequences for the erythroid transcription factors NF-E2 and GATA-116are not present in this region, at variance with the situation in the rat.10 Closer inspection of the region (Fig 7) shows, however, that upstream to the transcriptional initiation sites determined by Du et al,15 there are three potential erythyroid Krüppel-like factor (EKLF) binding sites (Fig 7, regions boxed with continuous lines). EKLF is an erythroid cell-specific factor, which is required for expression of the β-globin gene, which binds the consensus CCNCNCCCN,34,35 and which interacts physically and synergistically with GATA-1.36 In this context, it is worth mentioning that a sequence with one mismatch to the consensus for GATA binding is present in between the two potential EKLF binding regions (dotted rectangle in Fig 7, AGATGG in reverse orientation, where the G in boldface replaces the A of the consensus). Binding of GATA-1 to a sequence not matching the consensus at position 5 has been reported,37thus, this sequence may also have functional significance. Clearly, further studies are required to understand the regulation of the 1S promoter.

Although the S-transcript was suggested to generate soluble b5R,15 until now, the translation products of the human M-and S-transcripts had not been directly characterized. The results of the present study demonstrate unequivocally that the M and S-transcripts, when expressed in HeLa cells, generate the membrane-bound and soluble forms of b5R, respectively. Considering also the results of RNase protection assays and of radioimmunoblotting experiments, which detected no S-transcript and soluble enzyme in nonerythroid cells, the S-transcript appears to be both necessary and sufficient for the generation of soluble b5R.

In the rat, the myristoylated product of the M-transcript specifically associates with two subcellular compartments: MOM and ER membranes.30 Immunofluorescence analysis of HeLa cells transfected with the human M-transcript showed that also the human enzyme is capable of associating with the ER and with mitochondria (presumably with MOM). One function of MOM-associated b5R is in ascorbate free radical reduction,2,3 but it probably has other roles, which remain to be investigated. The persistence of the M-transcript in reticulocytes may be related to the requirement for continued synthesis of MOM-associated b5R also at late stages of erythroid maturation.

The generation of the soluble form of b5R requires some further discussion. As explained earlier (Fig 1), in the rat, exon 1S has an in-frame initiation codon, which is used inefficiently. “Leaky scanning”¹¹ allows use of a downstream AUG, contained in the second exon, with generation of soluble b5R.10 In humans, the 1S exon does not contain an initiation codon, however, the first amino acid of the stretch of 14 uncharged amino acids encoded in exon 2 is a methionine (instead of valine in the rat, see Fig 1). Therefore, also in humans, the potential initiation codon for soluble b5R is preceded by an upstream AUG. The results of this study strongly suggest that this upstream AUG allows leaky scanning, resulting in the use of the downstream AUG to produce soluble b5R. Initiation from the upstream AUG should have resulted in a protein capable of associating with the phospholipid bilayer through its N-terminal hydrophobic domain. Indeed, the additional hydrophobicity conferred by myristoylation is not required for association of b5R with artificial or natural membranes.30,38 Another hypothesis is that the soluble b5R in the transfected cells was generated by proteolysis of the translation product initiating from the upstream AUG. This possibility is very unlikely, because the product of the M transcript would contain the same hypothetical cleavage site at the end of the hydrophobic domain, yet it generated no detectable soluble b5R.

The first AUG of the human S transcript, although not preceded by a purine in position -3, is followed by a G in position +4, a context often found for AUGs, which function well as initiation codons.11 A recent study, however, shows that the enhancing effect of G at +4 is strongly diminished by a U in position +5.39 This is exactly the context of the first AUG of the human S-transcript and probably explains why it is used inefficiently.

In conclusion, although the position of the first AUG is shifted in man, the basic features of the S transcript described for the rat, ie, tissue specificity and generation of soluble b5R from an internal initiation codon, appear to be conserved. Whereas the importance of soluble b5R in MetHb reduction is well established, the biologic significance of the conserved and unusual features of its mode of production in erythrocyte precursors remains to be investigated.

We are grateful to T. Yubisui for his gift of a human b5R cDNA clone, to K. Shirabe for methodologic advice and for providing us with unpublished data, as well as with the entire human b5R gene sequence. We thank E. Battaglioli for helpful suggestions, S. Ottolenghi for fruitful discussion, and D. Fornasari and F. Clementi for critically reading the manuscript.