Fetal expression of a human Aγ globin transgene rescues globin chain imbalance but not hemolysis in EKLF null mouse embryos

EKLF^+/− mice were bred with mice containing a single copy of a human ^Aγ transgene linked to the micro locus control region, μLCR^−201Aγ.12 These transgenic mice express ^Aγ globin at high levels during all 3 waves of hematopoiesis—in the yolk sac, the fetal liver, and the bone marrow. EKLF^+/−, μLCR^−201Aγ⁺ mice were identified by Southern blotting of HindIII-digested genomic tail DNA. Presence of the mutant and wild-type EKLF alleles was determined as described.5 Presence of the μLCR^−201Aγ transgene was determined by hybridization with a 722-bp Asp718-HindIII human HS-2 probe derived from pUC19-HSII^1.9β.13 In most cases, the presence of 1 versus 2 copies of the μLCR^−201Aγ transgene could not be determined with certainty. EKLF^+/−μLCR^−201Aγ⁺ mice were interbred, and, in some cases, EKLF^+/−μLCR^−201Aγ⁺ mice were bred with EKLF^+/−μLCR^−201Aγ⁻ mice. Staged litters were killed at E12 to E17 to examine definitive hematopoiesis. The morning of vaginal plug discovery was designated E0.

Total RNA was prepared from fetal livers,14 and RNase protection analyses were performed as described.15 One microgram total RNA was hybridized simultaneously with murine α-globin and human γ-globin riboprobe, generated as described.15 The γ-globin probe was generated with 5-fold less cold rCTP than the α-globin probe, so that the specific activity, and therefore the signal, was 5-fold greater. The intensity of bands corresponding to the protected globin mRNA was quantitated using a Molecular Dynamics PhosphorImager and ImageQuant software (Amersham Pharmacia Biotech, Uppsala, Sweden).

Embryos were carefully dissected from the uterus to maintain the integrity of the uterine and vitelline circulations. The umbilical and vitelline vessels were clamped, the yolk sac was punctured, and whole blood was immediately collected from embryos by direct cardiac puncture of the beating heart. Twenty-five microliters whole blood was diluted immediately into 75 μL acid-citrate-dextrose and analyzed on a Technicon H-3 automated blood analyzer.16Values for hematocrit and hemoglobin were multiplied by the dilution factor and reported as the mean ± SEM from embryos of equivalent genotype. Because Aγ^+/− and Aγ^+/+ could not be reliably distinguished, they have been reported together as Aγ⁺ animals.

Fetal livers were surgically resected, and single-cell suspensions were made in phosphate-buffered saline (PBS) by passage through a 21-gauge needle. Cells (1 ×  10⁵) were cytocentrifuged at 500g and fixed in methanol:acetone (1:1). Specimens were stained with a fluorescein isothiocyanate (FITC)-conjugated monoclonal antibody specific for human γ-globin chains [9C3, a kind gift from Dr Thomas Campbell] as described.10 Specimens were simultaneously stained with 0.01% 4′-6-diamidino-2-phenylindole HCl (DAPI; Sigma, St. Louis, MO) to identify all cell nuclei in the field. Fresh fetal liver cells (10⁵ cells in 100μL PBS) were stained for hemoglobin by incubation in 0.2% o-dianisidine (D-9143; Sigma) in 0.3% glacial acetic acid/3% H₂O₂ for 5 minutes. Cells were subsequently cytocentrifuged (as above) and counterstained in Harris' hematoxylin for 30 seconds.

To analyze the component hemoglobins in blood, circulating red cells were isolated by bleeding E14 to E17 embryos into 1.5 mL PBS. Hemolysates were prepared from the packed red cells by freeze-thawing in water. Hemoglobins were separated by isoelectric focusing and visualized after staining in o-dianisidine. Individual hemoglobin bands were excised, extracted with water, and subjected to electrospray mass spectroscopy to identify constituent globin chains according to their precise average molecular weights as described before.17Selected separated hemoglobin species were analyzed by analytical reverse-phase high-performance liquid chromatography (HPLC) using the system previously described.18 The elution gradient was based on a method of Shelton et al,19 and it was optimized to afford separation of murine adult and embryonic globins and human fetal globins. It consisted of 3 linear steps, from 58%A/42%B to 56%A/44%B in 20 minutes, then to 44% A in 60 minutes, and then to 15% A in 40 minutes, where A was 20% acetonitrile/0.1% trifluoroacetic acid and B was 60% acetonitrile/0.1% trifluoroacetic acid. Identity and N-terminal processing of the murine embryonic ζ-globin were confirmed by observing its isolation from hemolysate by analytical reverse-phase HPLC, tryptic digestion, and LC/MS analysis of proteolytic fragments.18 

We previously suggested that the fatal anemia in EKLF^−/− embryos is primarily caused by β-thalassemia.5 To improve globin chain balance in EKLF^−/− fetal liver erythrocytes and thereby improve the anemia, EKLF^+/− animals were bred with a mouse strain that contains a single copy of a μLCR^−201Aγ-globin transgene.12EKLF ± μLCR^−201Aγ⁺ mice were identified by Southern blotting of tail DNA (see “Materials and methods”) and interbred. We could not be certain whether embryos harbored 1 or 2^Aγ transgene alleles by Southern blotting, so the genotype has been reported as + or − to reflect the presence (+/+ or +/−) or absence (−/−) of the Aγ transgene. Expression of the human ^Aγ transgene in the fetal livers of embryos was 30% or greater than that of the endogenous murine α-globin gene. This was determined by PhosphorImager quantitation of RNase protection analyses of the α-globin and γ-globin mRNA after a correction was made for the 5-fold greater specific activity of the γ-globin riboprobe (Figure1A). There was no alteration in γ-globin mRNA levels in the fetal livers of EKLF^−/−versus EKLF^+/− embryos. Thus, EKLF was not required for γ-globin promoter function or for LCR function in its capacity to interact with the γ-globin promoter. Significantly, our experimental objective, which was to generate high-level expression of β-like mRNA (in this case, ^Aγ-globin) in the fetal liver of EKLF^−/− embryos, was achieved.

To confirm that γ-globin was present in EKLF^−/− Aγ⁺ fetal liver erythrocytes at the protein level, we performed immunofluorescence analysis for human γ-globin (see “Materials and methods”). Most EKLF^−/− Aγ⁺ fetal liver cells expressed cytoplasmic human γ-globin (Figure 1B), whereas there was no detectable green fluorescence in EKLF^−/−fetal liver cells that harbored no transgene (not shown).

E15 hemolysates contained 6 different hemoglobin bands (Hb), as determined by isoelectric focusing (labeled 1-6, from anode to cathode, in Figure 2A). EKLF^−/− embryos contained less of hemoglobin bands 4 and 5 than EKLF^+/− litter mates. These were isolated from control hemolysates and subjected to electrospray mass spectroscopy to confirm the identity of the component globin chains by determination of their precise molecular masses. They were murine α₂β^maj₂ and murine α₂β^min₂, respectively (data not shown). This confirmed that the murine β^min gene and the β^maj gene are EKLF dependent in vivo, as expected from the sequence similarity and the relative position of the CACC box elements within the 2 promoters.

Hb 2 contained predominant globin chains of molecular mass 16 006 and 16 146 kd. The MWt of the first, 16 006 kd, was very close to the mass expected for murine ε-y globin (MWt 16 005.5); this identification was further confirmed by analytical reverse-phase HPLC (data not shown). The MWt of the second, 16 146 kd, did not correspond to the size calculated for murine ζ-globin according to its cDNA-derived protein sequence. However, after peptide mapping and partial sequencing of the N-terminal peptide (Ac-Ser-Leu-Met-Lys, MWt 519.3 kd), this species was authenticated as the N-terminally processed murine ζ-globin (removal of initiator Met and acetylation of the N-terminus, MWt 16 145.9, data not shown). Hb 6 was murine α₂ε₂. Because murine ε-y is only expressed in the yolk sac, the presence of this band reflected the persistence of some circulating yolk sac–derived erythroid cells at E15. The presence of the Aγ transgene had no effect on the level of α₂ε₂ (Figure 2A).

Two novel hemoglobins were detectable in all Aγ⁺ embryos but not in Aγ⁻ litter mates (bands 1 and 3 in Figure 2A). Hb 3 contained peaks corresponding to MWt 14 981 kd, 14 996 kd, and 16 009 kd, which identified them as murine α1, murine α520, and human ^Aγ chains (Figure2C). Thus, Hb 3 is a hybrid murine α₂-human^Aγ₂ hemoglobin (mα₂h^Aγ₂). It was the predominant hemoglobin present in EKLF^−/− Aγ⁺ embryos. The amount of Aγ chains normalized to murine α-globin chains in the hemolysate was 31% in the EKLF^+/+ Aγ⁺ embryo (Figure2A, lane 3), as measured by reverse-phase HPLC (Figure 2D). Thus, the strategy to ameliorate the globin chain imbalance was successful.

Hb 1 contained a predominant MWt species of 16 009 kd, which identified it as human ^Aγ. Murine ζ-globin (see above), with a MWt of 16 146 kd (Figure 2B), was also detectable in isolated Hb 1 but only at 10% to 20% of ^Aγ chain levels. Again, Hb 1 was only present in embryos that were subsequently genotyped as Aγ⁺. Thus, Hb 1 consisted primarily of^Aγ₄ (HbBarts), with some comigration of a hybrid mζ₂^Aγ₂ hemoglobin. HbBarts accounted for less than 15% of the total hemoglobin in E15 EKLF^−/− Aγ⁺ embryos (see Figure2A, lane 4), suggesting that the interaction between human^Aγ chains and murine α-chains within fetal liver erythrocytes was efficient but incomplete. Low-level amounts of embryonic βh1 globin was observed (by electrospray mass spectroscopy and reverse-phase HPLC) in E15 hemolysates of all embryos, but a mobility of the βh1-containing hemoglobin in isoelectric-focusing gels was not established.

The EKLF null phenotype is highly consistent. Embryos killed at E11 are indistinguishable from wild-type litter mates, embryos killed at E12 have slight pallor, and the severity of pallor increases until E15 when the embryos are severely anemic.5 We have never detected a living EKLF^−/− embryo at E16. This time course correlated precisely with the progressive switch from circulating embryonic to fetal liver–derived red blood cells.

Surprisingly, there were no live-born EKLF^−/−Aγ⁺ animals of the 98 live-born mice generated from an F2 cross of EKLF^+/− Aγ⁺ animals. Therefore, litters were analyzed at E11 to E17 to determine whether there was any improvement in the severity of anemia in EKLF^−/− embryos afforded by the Aγ transgene. At E15 there were 2 apparent degrees of pallor in the litters. One set of animals displayed pallor typical of EKLF^−/−embryos, and the other set had slightly less pallor (Figure3A). Southern blot analysis of carcass DNA subsequently revealed that the pinker animals contained the Aγ transgene. The fetal livers of these EKLF^−/−Aγ⁺ embryos were also slightly more crimson than those of the EKLF^−/− Aγ⁻ litter mates, but not as crimson as those of the wild-type embryos (Figure3B). Furthermore, o-dianisidine staining of fetal liver erythrocytes revealed a slight improvement in the presence of the Aγ transgene (Figure 3C). Taken together, these results indicated that there was slight improvement in the production of hemoglobin in EKLF^−/− fetal liver erythrocytes that expressed high levels of γ-globin; this was consistent with the formation of hybrid mouse–human hemoglobin.

Despite improvement in hemoglobinization, there was minimal, if any, improvement in the survival of EKLF^−/−Aγ⁺ embryos compared with their EKLF^−/− Aγ⁻ litter mates. At E16, all EKLF^−/− Aγ⁺ embryos (n = 2) and EKLF^−/− Aγ⁻embryos (n = 3) were dead, whereas all (n = 13) EKLF^+/+and EKLF^+/− litter mates were alive irrespective of the presence or absence of the transgene (Table1).

In addition, the peripheral blood of EKLF^−/−Aγ⁺ embryos was similarly abnormal compared with that of EKLF^−/− Aγ⁻ litter mates. There were few enucleated fetal liver–derived erythrocytes in EKLF^−/− Aγ⁺ embryos, whereas these account for more than 80% of circulating cells in EKLF^+/− litter mates by E15.5 Rather, the circulating cells were predominantly nucleated with dyserythropoietic, poorly hemoglobinated cytoplasm, though a few cells had pinker cytoplasm than cells in EKLF^−/−Aγ⁻ litter mates (Figure 3D). Furthermore, there was still a marked increase in fetal liver iron deposition (data not shown), consistent with persistent red cell destruction or ineffective erythropoiesis in the fetal livers of EKLF^−/−Aγ⁺ embryos.

Finally, there was no significant improvement in hemoglobin and hematocrit values of EKLF^−/− Aγ⁺embryos compared with their EKLF^−/−Aγ⁻ litter mates (Table2). We conclude that there was no significant improvement in hemolysis or survival of EKLF^−/− embryos afforded by the Aγ transgene. In view of the marked benefit even the moderate expression of γ-globin produced in humans with β-thalassemia, we propose that the reduced expression of nonglobin EKLF target genes must play a major role in the lethal EKLF^−/− phenotype.

We found no significant improvement in red blood cell morphology, level of anemia, or survival in EKLF^−/− mice in which the globin chain imbalance was significantly reversed by transgene-derived γ-globin chain synthesis. Thus, we concluded that EKLF target genes other than β-globin alone must play a crucial role in fetal liver erythropoiesis. At first glance, this conclusion appears to oppose that of Lim et al,21 who found that reintroduction of an LCR-γ-globin gene into EKLF^−/− ES cells rescues contribution to circulating erythroid cells in chimeric animals. A close examination of the published data suggested that the level of EKLF^−/− ES cell contribution to the blood was still significantly less than the contribution to other tissues. Thus, EKLF^−/− Aγ-globin–expressing erythroid cells may also be partially defective in these chimeric mice. The 2 attempts to rescue the EKLF null phenotype (that described here and that of Lim et al21) have some important experimental design differences. First, there may be certain EKLF target genes that have critical role(s) only at the fetal liver stage of erythropoiesis. The chimeras could survive this developmental phase by virtue of blood production from non–ES cell-derived fetal liver stem cells, whereas the EKLF^−/− embryos described in this article could not. Alternatively, wild-type stem cells, microenvironment components, or both within the adult bone marrow of chimeric animals may somehow nurture EKLF^−/− erythroid cells. That is, there may be a non-cell-autonomous component to the observed survival of EKLF^−/− Aγ⁺ erythroid cells in chimeric animals21 unavailable to erythroid cells within the fetal livers of the EKLF^−/−Aγ⁺ mice.

The nature of alternative EKLF target genes remains undetermined. Genes encoding other globin chains, heme biosynthetic enzymes, glycolysis pathway enzymes, other transcription factors, transmembrane proteins, and cytoskeletal proteins all have CACC box elements in their promoters.9,22 Many fall within the consensus site for EKLF (NCN-CNC-CCN) predicted by the crystal structure of the related protein, Zif268, bound to DNA.3 However, it remains to be tested whether EKLF binds equally well to all such CACC box elements or just to a subset of them. A detailed examination of the spectrum of CACC sequences able to bind purified recombinant EKLF would be helpful in the resolution of the precise EKLF binding-site preferences. In short, it is difficult to be sure whether any of the erythroid promoters with important CACC box elements actually binds EKLF in vivo.

One approach to the problem of alternative EKLF target genes is to examine the expression of candidate genes in fetal liver cells derived from EKLF^−/− embryos. We have previously examined the expression of GATA-1, EpoR, PBGD, βh1, and murine ε-y globin genes in EKLF^−/− fetal liver cells and found them all to be EKLF independent.5 A possible explanation is that other Kruppel-like factors are able to function at many of these promoters in vivo. This may reflect the inability of EKLF to bind to these sites in vivo (a promoter context argument), or it may indicate that similar Kruppel-like factors, such as basic Kruppel-like factor (BKLF),23 can bind these sites and compensate for the loss of EKLF (a redundancy argument). These alternatives might be resolved by studying mice null for both EKLF and BKLF. A similar dilemma exists for GATA-1 null embryos, which also die from anemia.24 It has been suggested that GATA-2 may substitute for GATA-1 at many erythroid gene promoters.

One potentially important EKLF target gene is amino-levulinic acid synthetase (ALAS), the rate-limiting enzyme of the heme biosynthesis pathway. There are 2 overlapping CACC elements in the erythroid specific ALAS promoter at −49/−39,25 with an identical sequence (albeit on the reverse strand) to that found in the human β-globin promoter. This site binds EKLF in vitro, and the ALAS promoter is activated by EKLF in transient assays in a CACC element-dependent fashion.25 It remains to be determined whether EKLF is required for ALAS expression in fetal liver cells of developing embryos. Interestingly, ALAS null ES cells display a defect in erythroid cell maturation,26 and a defect in ALAS is responsible for the anemic zebrafish mutant,sautern.27 Nevertheless, it is likely that a significant amount of heme is produced in definitive and embryonic erythroid cells in the absence of EKLF. Although we did not make any direct measurements of heme, the positive staining for o-dianisidine suggests the presence of an intact heme moiety. Thus, we suggest that ALAS does not require EKLF for expression in vivo or that sufficient enzyme is produced in EKLF^−/− erythroid cells from reduced mRNA levels to permit adequate heme production rates.

Erythroid cytoskeletal genes may also be dependent on EKLF for expression. Moreover, their reduced expression may contribute to the EKLF null phenotype. The promoters for many of the erythroid cytoskeletal proteins, including band 4.1, band 3, ankyrin, α- and β-spectrin, and band 7.1 have been cloned and sequenced. Many have GC-rich elements in their promoters, but few fit perfectly with the proposed EKLF consensus. To date, none of these genes has been tested for their dependence on EKLF for expression in vivo or in transient assays. Many have been disrupted in ES cells by homologous recombination, and others occur as spontaneous mouse mutants.28 Mice null for band 4.1, ankyrin, β-spectrin, α-spectrin all have hemolytic anemia of varying severity (reviewed in27,28). The most severe anemia occurs in mice with defective β-spectrin (ja/ja mice), with approximately 90% dying in the first 2 weeks of life.29 In all cases the phenotypes are milder that the EKLF null phenotype. Nevertheless, it remains possible that defective expression of multiple red cell cytoskeletal proteins could result in a severely anemic phenotype such as that present in EKLF^−/− embryos.

When considering potential EKLF target genes, it is helpful to remember that EKLF^−/− embryonic red cells are morphologically normal or nearly so. Thus, it may be reasonable to consider those genes that are selectively expressed in definitive cells as more likely candidates for EKLF target genes. One of the major differences between embryonic and definitive red cells is that definitive red cells extrude their nuclei. It is striking that most EKLF^−/− erythroid fetal liver cells remain nucleated even in the presence of high levels of γ-globin (Figure 3). We originally suggested this was secondary to erythroid stress, but it remains possible that EKLF coordinates the process of enucleation itself.

Embryonic and definitive erythroid cells also differ in their metabolism. Definitive erythroid cells rely almost exclusively on anaerobic metabolism for the production of adenosine triphosphate from glucose. They are also dependent on the pentose-phosphate pathway for the generation of reducing power in the form of NADPH. Enzymes of the Embden–Myerhoff glycolysis pathway are certainly important for the metabolic function of definitive human erythroid cells. Defects in pyruvate kinase (PK), the rate-limiting enzyme of glycolysis, are common in humans and result in chronic hemolytic anemia. Like many erythroid genes, the PK gene contains closely associated CACC and GATA motifs in the proximal promoter. The CACC site is required for full expression in transient transfection assays and it could bind EKLF, though this remains to be tested. Additionally, an element 3.7 kb upstream of the transcriptional start site of the PK gene has erythroid-specific enhancer activity in transgenic mice and harbors a duplicated CACC element that fits well with the EKLF consensus.30 More work must be done to determine whether PK is truly EKLF dependent in vivo.

To make further progress in the identification of putative EKLF target genes, a cell-based functional assay would be helpful. For example, an immortalized EKLF^−/− erythroid cell line would allow conditional reintroduction of EKLF and subtractive approaches to target gene discovery. Candidate target genes could also be examined in such a system. Our results suggest that attempts to reactivate γ-globin in adults that harbor mutations in the β-globin gene through antagonism of EKLF carry a significant risk for inducing additional defects in red blood cell function.

We thank Carlo Brugnara and Vivian Smith for hemoglobin electrophoresis. We also thank Bing Chuen-Lau and John Kim for expert technical assistance, Klar Kleman for hemoglobin isolation by isoelectric focusing, and Cedric Shackleton for his support in mass spectrometric studies. We thank Y. Fujiwara and K. Cunniff for help with animal husbandry and Elise Coghill for critical appraisal of this manuscript.