In response to anemia, erythropoietin (Epo) gene transcription is markedly induced in the kidney and liver. To elucidate how Epo gene expression is regulated in vivo, we established transgenic mouse lines expressing green fluorescent protein (GFP) under the control of a 180-kb mouse Epo gene locus. GFP expression was induced by anemia or hypoxia specifically in peritubular interstitial cells of the kidney and hepatocytes surrounding the central vein. Surprisingly, renal Epo-producing cells had a neuronlike morphology and expressed neuronal marker genes. Furthermore, the regulatory mechanisms of Epo gene expression were explored using transgenes containing mutations in the GATA motif of the promoter region. A single nucleotide mutation in this motif resulted in constitutive ectopic expression of transgenic GFP in renal distal tubules, collecting ducts, and certain populations of epithelial cells in other tissues. Since both GATA-2 and GATA-3 bind to the GATA box in distal tubular cells, both factors are likely to repress constitutively ectopic Epo gene expression in these cells. Thus, GATA-based repression is essential for the inducible and cell type–specific expression of the Epo gene.


To maintain proper homeostasis during growth and development, the regulation of gene expression needs to be strictly controlled, especially at the transcriptional level. For instance, in response to anemia and hypoxia, erythropoietin (Epo) gene transcription is activated in the kidney and liver (reviewed in Ebert and Bunn1). Epo is then secreted from these tissues to stimulate erythropoiesis in hematopoietic tissues. The kidney produces approximately 90% of the total Epo during the response to anemia,2 indicating that Epo gene expression is under strict tissue-specific regulation. Thus, the Epo gene provides an intriguing model for understanding gene expression in higher eukaryotes. Furthermore, dysfunctional Epo gene regulation has been shown to underlie the pathogenesis of renal anemia.3,4 As such, recombinant Epo has been used widely as a powerful drug for treating patients with anemia (reviewed in Jelkmann5).

Epo gene expression in response to anemia and/or hypoxia has been vigorously analyzed by exploiting the human hepatoma cell lines Hep3B and HepG2 that produce Epo in a hypoxia-inducible manner.6 An important finding from these studies was that Epo gene expression is regulated by an enhancer located 3′ to the transcriptional termination site.7 This 3′ enhancer contains a hypoxia response element (HRE) that has been shown to bind hypoxia-inducible transcription factors (HIFs).7 A binding sequence for nuclear receptor also resides in the enhancer.1,8 Thus, these 2 cis-acting elements may control Epo gene expression in a hypoxia-inducible manner (reviewed in Koury9).

A GATA factor–binding motif (GATA box) has been identified in the core promoter region of the Epo gene, where a TATA box normally resides.10 We have found that this GATA box actively participates in Epo gene regulation. The GATA box acts as a negative regulatory element in the hepatoma cell lines.10 During normoxic conditions, GATA transcription factors bind to the GATA box and repress Epo gene transcription, but when exposed to hypoxia, GATA binding markedly decreases, with a marked increase in Epo gene expression.10,11

Although the mechanisms underlying inducible gene expression of the Epo gene are generally understood, the basis of tissue-specific Epo gene regulation is largely unknown. Constitutive expression, as well as ectopic transgene expression, has been frequently detected in several transgenic mouse reporter studies using constructs containing less than 20 kb of the Epo gene flanking region.1215 In agreement, our own transgenic reporter gene expression analysis demonstrated that an 11-kb mouse Epo gene fragment, which incorporates the 2 important regulatory elements mentioned above (the promoter GATA motif and 3′ enhancer), was unable to fully recapitulate Epo gene expression (N.S. and M.Y., unpublished data, March 25, 2002). We therefore realized that regulatory element(s) critical for Epo gene expression must be contained in further upstream or downstream regions surrounding the Epo locus.

An unsolved but important issue in Epo gene regulation is the identification of precisely which kidney cells actively produce Epo. This topic is controversial; some reports claim that proximal tubular cells produce Epo,14,16 whereas others present evidence that the Epo-producing cell population might be glomerular,17 mesangial,18 or interstitial cells of the renal cortex.13,1923 We surmise that these discrepancies are most likely due to a low sensitivity in the detection of Epo.

To elucidate in this study how inducible and tissue-specific transcription of the Epo gene is attained, we established bacterial artificial chromosome (BAC) transgenic mouse lines expressing green fluorescent protein (GFP) as a reporter under the control of a 180-kb mouse Epo gene locus.24 This 180-kb region can sufficiently recapitulate inducible transgene expression in a tissue-specific manner, including expression in the kidney and liver. The GFP labeling clearly identified the renal Epo-producing (REP) population as interstitial cells. Surprisingly, we discovered that REP cells have a neuronlike shape and express marker antigens found in neuronal cells. Furthermore, we also found that a single nucleotide mutation in the promoter GATA box can cause constitutive ectopic expression of GFP in the renal distal tubules, collecting ducts, and certain epithelial cells of other tissues. Thus, GATA box–based repression is essential for inducible and cell type–specific Epo gene expression.


Bacterial artificial chromosome transgenes

A bacterial artifical chromosome (BAC) clone (Epo-60K/BAC, RP27826) containing the Epo gene and flanking regions was obtained from a C57black/6 mouse genomic library (Incyte Genomics, San Diego, CA). To make the wt-Epo-GFP construct, GFP cDNA was integrated into Epo-60K/BAC in Escherichia coli EL250 as described previously.25,26 For construction of the mutant Epo-GFP transgenes (m1-Epo-GFP and m2-Epo-GFP), the polymerase chain reaction (PCR)–based mutant 5′ arms of the targeting vectors were used.11,27 Recombinant BAC clones were identified through sequencing, restriction enzyme analysis, and Southern blotting. BAC transgenic constructs were purified from a bacterial cell suspension cultured in LB/chloramphenicol (25 μg/mL) for 24 hours at 32°C using a Nucleobond BAC DNA preparation kit (Macherey-Nagel, Düren, Germany). The purified DNA was linearized by digestion with λ-terminase (Epicentre Technologies, Omaha, NE) and purified by pulse-field gel electrophoresis.28


Purified linear transgenes (5 ng/μL) were microinjected into the pronuclei of fertilized eggs from BDF1 parents. Transgenic mice were screened by PCR genotyping of tail DNA using the following 3 pairs of primers. BAC1F (5′-GTGCGGGCCTCTTCGCTATT-3′) with BAC1R (5′-CAGGTCGACTCTAGAGGATC-3′) and BAC2F (5′-AGTGTCACCTAAATAGCTTG-3′) with BAC2AR (5′-CAGTACTGCGATGAGTGGCA-3′) were used to amplify the BAC vector sequences at the 3′and 5′ inserts, respectively. The primer set GFPs4 (5′-CTGAAGTTCATCTGCACCACC-3′) and GFPas4 (5′-GAAGTTGTACTCCAGCTTGTGC-3′) was used to amplify the GFP cDNA.29 Southern blotting analyses were carried out to determine the transgene copy numbers. Genomic tail DNA was prepared and digested with ApaI. Southern membranes were hybridized with a 32P-labeled probe. The copy numbers of the transgenes were determined by the intensities of the bands. To induce profound anemia (Hct: < .15 [ 15%]), 6- to 8-week-old mice were bled from the retro-orbital plexus (0.3-0.4 mL) at 48, 36, 24, 12, and 6 hours before analysis. To induce moderate anemia (Hct: .20-.35 [20%-35%]), the amount of bleeding was reduced to 0.2 mL. The plasma Epo concentrations were measured using a photometric enzyme-linked immunoabsorbent assay (Epo ELISA kit; Roche, Indianapolis, IN). The Gata2-GFP30 and Gata3-LacZ knock-in31 lines of mice have been described previously. All mice were kept strictly in specific pathogen-free conditions, and were treated according to the regulations of The Standards for Human Care and Use of Laboratory Animals of the University of Tsukuba.32


Tissues were fixed in 4% paraformaldehyde for 30 minutes and embedded in OCT compound (Sakura-Finetechnical, Tokyo, Japan). Sections 10 μm in thickness were incubated with rabbit anti-GFP polyclonal antibody (diluted 1:1000; Molecular Probes, Eugene, OR) or anti–GATA-4 antibody (diluted 1:500; Santa Cruz Biotechnology, Santa Cruz, CA) at 4°C overnight. After treatment with hydrogen peroxide, sections were incubated with horseradish peroxidase–conjugated anti–rabbit IgG secondary antibody. Color detection was performed using diaminobenzidine as a chromogen (brown staining). Hematoxylin was used as a counterstain. To detect the expressions of CD31 and Mac1, frozen sections were incubated with anti-CD31 and Mac1 antibodies (BD Pharmingen, San Diego, CA) conjugated with phycoerythrin (PE) for 2 hours at room temperature. To detect the expressions of α-fetoprotein (AFP), microtubule-associated protein 2 (MAP2), and neurofilament light polypeptide (NFL), sections were incubated with primary antibodies (Santa Cruz Biotechnology) against AFP (diluted 1:200), MAP2 (diluted 1:200), and NFL (diluted 1:200), respectively, followed by incubation with TRITC-conjugated anti–goat IgG secondary antibody (ZyMed, South San Francisco, CA). Fluorescent images were observed using the LSM510 confocal imaging system (Carl Zeiss, Heidelberg, Germany).


Liver sections were microdissected with the Leica laser microdissection system (LMD; Leica, Heidelberg, Germany) and RNAs were purified using RNeasy (Qiagen, Hilden, Germany). Total RNAs were purified using ISOGEN (Nippon-Gene, Tokyo, Japan). Reverse transcription was performed with SenscriptRT (Qiagen). Epo mRNA levels were measured quantitatively using the ABI PRISM 7700 (Perkin-Elmer, Waltham, MA) and the primer pair 5′-GAGGCAGAAAATGTCACGATG-3′ and 5′-CTTCCACCTCCATTCTTTTCC-3′ with FAM-labeled oligo-DNA probe (5′-TGCAGAAGGTCCCAGACTGAGTGAAAATA-3′).33 Glyceraldehyde phosphate dehydrogenase (GAPDH) was used as an internal standard.29 Semiquantitative reverse-transcriptase (RT)–PCR analysis of Epo-GFP transgene expression was performed using the primer pair 5′-AAGGATGAAGACTTGCAGCG-3′ and 5′-GAAGTTGTACTCCAGCTTGTGC-3′. Hypoxanthine guanine phosphoribosyl transferase (HPRT) was used as an internal control using 5′-GCTGGTGAAAAGGACCTCT-3′ and 5′-CACAGGACTAGAACACCTGC-3′ primers.29

Chromatin immunoprecipitation

Normal renal medulla (10 mg) was minced by needling and fixed with 1% formaldehyde for 10 minutes. Sample in a 200-μL aliquot was sonicated for 10 seconds on ice 12 times by a 2-mm microtip sonicator (Branson, Danbury, CT). One half of the sonicated sample was used as preimmunoprecipitated control (input), while the other half was incubated with antibody against GATA-2, GATA-3, or normal rabbit IgG (Santa Cruz Biotechnology). Antibody-bound chromatin complexes were precipitated with an Immunoprecipitation Kit (Upstate Biotechnology, Lake Placid, NY). To detect the promoter region in the precipitated chromatin complexes, PCR was performed using the following primers: 5′-TCACGCACACACAGCTTCAC-3′ and 5′-CACGCTGCAA-GTCTTCATCC-3′ for the Epo promoter; 5′-TTATCTGGAGTCCATTAATGAGG-3′ and 5′-CTGCTCGGCCTTCTGAGCGCTG-3′ for the Aquaporin-2 (Aqp2) gene promoter; and 5′-TATGGCGGGCAAGAAGTTGA-3′ and 5′-GTACTAGGCCAGGACTAGTG-3′ for the Gata1 gene promoter.34 The Aqp2 and Gata1 promoters were used as positive and negative controls, respectively.


Regulatory region for tissue-specific and inducible Epo gene expression

To obtain a genomic DNA fragment suitable for our in vivo reporter expression analysis, we screened several BAC clones containing the mouse Epo gene and finally selected a 180-kb BAC clone (Epo-60K/BAC; Figure 1A). The region spanning exon II to intron IV of the Epo gene in the BAC clone was replaced with a GFP reporter gene through standard homologous recombination.2426 The GFP gene was inserted in frame in such a manner that GFP expression would mimic Epo gene expression (Figure 1A). We confirmed the authenticity of the recombinant BAC clone by Southern blotting and DNA sequencing (data not shown).24 The BAC-derived transgene (wt-Epo-GFP) was microinjected into fertilized eggs and 3 transgenic mouse lines were established. Transgene expression could not be detected under normoxic conditions, yet was unequivocally detected in the kidneys and livers of the wt-Epo-GFP mice when bleeding-induced anemia was carried out (Figure 1B panels). These results were quite reproducible; anemia-induced expression of the GFP gene was shown in 3 (lines WA, WB, and WC) of 4 lines of mice by semiquantitative reverse transcriptase–PCR (RT-PCR) analysis. This was in excellent agreement with the endogenous Epo gene expression detected by quantitative RT-PCR analysis (Figure 1B bar chart). The Epo-GFP transgene expression was detected exclusively in both anemic liver and kidney. Therefore, we concluded that this 180-kb Epo-GFP transgene contained the regulatory regions sufficient for tissue-specific and inducible Epo gene expression in vivo.

Figure 1

Structure and tissue distribution of the Epo-GFP transgene. (A) Strategy for constructing the Epo-GFP transgenes in the BAC recombination system. Epo-60K/BAC (top), including the 60-kb 5′ upstream and 120-kb 3′ downstream regions of the mouse Epo gene, was isolated. Exons I to V of the Epo gene are depicted by black boxes. The targeting vector contained GFP cDNA (white box) and a polyadenylation signal (pA; hatched box) and a Neo cassette (speckled box) between the FRT sequences (white ovals). The vector was homologously recombined with Epo-60K/BAC within the 5′ (1.1 kb) and 3′ (0.9 kb) homologous arms. The Neo cassette in the targeted BAC clone was excised using the FLP-FRT system in bacteria. To verify integration of the intact transgene into the mouse chromosome, 5′ and 3′ fragments derived from the BAC vector (a and b) and GFP cDNA (c) were amplified by PCR using primers indicated by the double arrowheads. (B) Expression of the endogenous Epo gene (bar chart) and wt-Epo-GFP transgene (panels) in adult mice. Epo mRNA levels under normal (gray bars) and anemic (black bars) conditions were measured by quantitative RT-PCR for the organs indicated and normalized to the level of GAPDH mRNA (bar chart). Data are the means (± SD) of at least 3 independent mice. Semiquantitative RT-PCR analysis of transgene expression in the wt-Epo-GFP transgenic mouse (line WA) under normal and anemic conditions was performed for the organs indicated (panels). HPRT was used as an internal control. (C) Southern blotting analysis of ApaI-digested genomic DNA revealed the copy numbers of the transgenes. Tail DNA from each transgenic line was digested with ApaI (Ap; indicated in A) and the endogenous Epo gene (5.2 kb) and transgene (4.4 kb) were hybridized with a radiolabeled probe (indicated in A).

While the expression of GFP from the transgenic BAC clones was far more stable than that from the plasmid transgenes in mice in vivo, we also noticed a variation in the intensity of fluorescence between the lines. The green fluorescence intensities (data not shown) were not related to the transgene copy numbers (Figure 1C). These data suggest that the BAC transgene might be affected to some extent by the integration position effect. We confirmed integration of the intact BAC transgene for all the lines of transgenic mice by a series of PCR and Southern blotting analyses (data not shown).

Renal Epo-producing cells are peritubular interstitial cells expressing neuronal markers

An important prerequisite for the current study was to elucidate the mechanisms regulating Epo gene expression by identifying the precise cells in the kidney that produce Epo, since earlier conclusions were controversial.24 To this end, we exploited normal wt-Epo-GFP (Figure 2A) and wild-type (Figure 2C) mice and wt-Epo-GFP mice bled to induce anemia (Figure 2B,D) and stained for GFP expression in the kidneys. The anemic wt-Epo-GFP mice displayed a GFP expression that was detected reproducibly and uniquely in the peritubular interstitial cells of the border region of renal medulla and cortex. We referred to these cells as REP (renal Epo-producing) cells and performed further characterization using immunofluorescent techniques.

Figure 2

Morphology of renal Epo-producing (REP) cells expressing the wt-Epo-GFP transgene. Immunohistochemical staining was carried out with anti-GFP antibody of normal (A) and anemic (B-D) mouse kidneys from transgenic line WA (A,B,D) and wild-type (C) mice. GFP expression (brown) was detected in the peritubular interstitial cells (arrows in B,D) with long projections (arrowheads in D) between the proximal tubules (PT) and vessels (V). Fluorescent images of REP cells from anemic kidneys of transgenic mice (line WA) taken with the confocal laser scanning microscope (E-I). When costained with cell lineage markers (red), REP cells (green) were negative for CD31 (F) and Mac1 (G), but positive for MAP2 (H) and NFL (I). Scale bars represent 50 μm (A-C) and 10 μm (D-I).

The REP cells displayed a unique stellar or arboroid shape with projections extending in various directions (Figure 2D,E; Video S1A,B, available on the Blood website; see the Supplemental Materials link at the top of the online article). These cells were located between the proximal tubular cells and vascular endothelial cells and were tightly attached to the renal tubular cells (Figure 2D). When we stained vascular endothelial cells with anti-CD31, REP cells did not overlap with the red staining representing CD31 (Figure 2F). As the dendritelike processes of the REP cells suggested a neuronal or macrophage origin, we examined whether REP cells express the macrophage marker Mac1 or neural-specific markers such as microtubule-associated protein 2 (MAP2) and neurofilament light polypeptide (NFL). We found that REP cells are Mac1 negative (Figure 2G), but positive for MAP2 and NFL (Figure 2H and I, respectively). We found that not all interstitial cells positive for MAP2 or NFL express GFP. In our preliminary analysis, 5% to 10% of those neural marker–positive cells expressed wt-Epo-GFP transgene when Hct was reduced to .15 (15%). Concurring with the previous finding that Epo is produced in CD73 (ecto-5′-nucleotidase)–positive peritubular cells in the kidney,22,23 REP cells were found to express CD73 (data not shown). While we used mainly line WA mice in these analyses, we confirmed that the other 2 lines of wt-Epo-GFP mice showed reproducible results (data not shown). Taking advantage of the GFP labeling, we executed 3-dimensional imaging of REP cells by confocal laser microscopy (Video S1A,B). Each REP cell had 4 to 6 projections stretching in various directions in the cavity between the tubules, and several REP cells formed a cluster (Video S1B). Taken together, REP cells with long processes seem to form a reticular network between renal tubules and capillaries.24

The number of Epo-producing cells correlates with the plasma Epo concentration

Upon induction of bleeding anemia, the number of GFP-positive cells in the kidney increased exponentially, in excellent agreement with both increasing plasma Epo concentration and decreasing hematocrit (Hct) score (Figure 3). This analysis revealed that renal Epo production is regulated mainly by the number of REP cells, rather than by the expression levels of the Epo gene in individual REP cells. Indeed, there was no significant difference between mildly anemic and severely anemic animals in the fluorescence intensity of individual GFP-positive cells (data not shown). Since Epo production was also activated in mice made anemic by the cell cycle inhibitor 5-fluorouracil or γ-ray irradiation (data not shown), the increase in GFP-positive cells is likely due to the onset of Epo-producing ability, but not due to the proliferation of Epo-expressing cells. This result supports previous a hypothesis by Koury et al35 that Epo production is regulated by a small subset of peritubular interstitial cells with on-off mode.

Figure 3

The number of GFP-positive cells in the kidney of wt-Epo-GFP transgenic mouse is increased as the hematocrit is decreased. Various levels of anemia were induced in wt-Epo-GFP transgenic mice (line WA) by bleeding over 2 days and kidney sections were taken. The plasma Epo concentrations (Embedded Image) and the numbers of GFP-positive cells in the kidney sections (•) increased as the hematocrit (Hct) decreased.

Epo-GFP expression in hepatocytes

Under severe anemic conditions, Epo production also occurs in the liver, so we examined GFP expression in hepatocytes. Bleeding anemia gave rise to the emergence of GFP-positive hepatocytes, which rapidly increased in number as the Hct score fell (Figure 4A-C). Each GFP-positive region in the liver sections was a concentric circular ring surrounding a central vein (CV; Figure 4B,C). Since oxygen is supplied from the interlobular artery, the oxygen supply to hepatocytes surrounding the CV is low compared with other regions. In our microdissection analysis, following induction of anemia, the endogenous expression of Epo mRNA was detected specifically in cells surrounding the CV, but not in cells surrounding the interlobular triad (IL; Figure 4D). These results thus demonstrate that Epo gene expression is induced in hepatocytes through sensing the hypoxic threshold of the cellular oxygen tensions.

Figure 4

Induction of wt-Epo-GFP transgene expression by bleeding anemia in the liver. Anti-GFP immunohistochemistry of the livers from wt-Epo-GFP transgenic mice (line WA) under normal (A, Hct: .45 [45%]) and anemic (B, Hct: .30 [30%]; C, Hct: .15 [15%]) conditions is shown. At the lowest Hct (C), GFP-positive cell regions (brown) had expanded around the central vein (*), but hepatocytes surrounding the interlobular triad (sharp) did not express GFP. (D) Relative Epo mRNA levels in the kidneys and livers from line WA mice under normal (n, Embedded Image) and anemic (a, Embedded Image) conditions were measured by quantitative RT-PCR and normalized to the levels of GAPDH mRNA. Hepatocytes surrounding the interlobular triad (IL) and central vein (CV) under anemic conditions were collected using laser microdissection and the expression of Epo mRNA was examined by quantitative RT-PCR. Data are the means (± SD) of 3 independent mice. (E) GFP expression was also detected in E13.5 fetal liver from a wt-Epo-GFP transgenic embryo (line WA). The GFP-positive cells were fetal hepatocytes that were also positive for α-fetoprotein (AFP; red immunofluorescence in panel F). Panels E and F were merged in panel G. Scale bars are 300 μm (A-C) and 20 μm (E-G).

We next analyzed GFP expression in livers from wt-Epo-GFP transgenic mouse embryos. At embryonic day 13.5 (E13.5), GFP expression was detected in hepatocytes expressing α-fetoprotein (AFP) (Figure 4E-G). In contrast to REP cells, GFP-positive hepatocytes did not stain positive for neural markers (data not shown). Hematopoietic cells in the fetal liver made direct contact with the GFP-positive hepatocytes, suggesting that fetal liver erythropoiesis is supported by paracrine Epo production from the hepatocytes. Hepatic GFP expression was found in E9.5 embryos and this expression persisted until the neonatal stage (data not shown). GFP expression was slightly enhanced by the induction of bleeding anemia in the mother (data not shown).

When housed in a 6% oxygen chamber, GFP expression was induced in the livers and kidneys of adult wt-Epo-GFP transgenic mice, as was the case for bleeding anemia (data not shown). Throughout this analysis, we did not find GFP-expressing cells in any tissues other than the kidney and liver, irrespective of the normoxic or hypoxic/anemic condition. Based on these observations, we concluded with the following 3 points. First, in animals, Epo is produced in REP cells and hepatocytes through sensing the hypoxic threshold. Second, Epo production is controlled by the increase in the number of Epo-expressing cells, but not through the enhancement of Epo in individual cells. Third, the wt-Epo-GFP transgene contains the regulatory regions necessary for sufficient expression of Epo in physiologically important Epo-producing tissues.

A mutation in the GATA box results in constitutive GFP expression in epithelial cells

Our previous analyses showed that in Hep3B and HepG2 cells, GATA-2 represses Epo gene expression by binding to its GATA box located 30-bp 5′ to the transcription start site.10,11 To ascertain whether this repression occurs in vivo, we examined GATA box activity using the Epo-GFP BAC transgenic reporter system. For this purpose, we mutated the GATA box in the wt-Epo-GFP construct to TATA (Figure 5A m1-Epo-GFP), to which the GATA factors cannot bind.11 As there was the possibility that TATA-binding factor (TBP) may recognize the − 30 TATA box in the m1-Epo-GFP construct and activate Epo gene expression in place of the GATA factors,10,36 we prepared another mutant that is not recognizable by either GATA or TBP (TTTA mutant, m2-Epo-GFP transgene; Figure 5A).

Figure 5

Expression profiles of the Epo-GFP transgenes with mutations in the promoter GATA sequence. (A) Sequences near the GATA factor–binding site in the promoter region of the mouse Epo gene. The wild-type GATA-box in the wt-Epo-GFP transgene was mutated to create the transgenes m1-Epo-GFP and m2-Epo-GFP. Capital T indicates the point 30-bp upstream from the transcription initiation site. Sections of kidneys (B-G), livers (H-J), and lungs (K-M) from the m1-Epo-GFP transgenic mouse line 1A (B,C,E,F,H,I,K,L) and wt-Epo-GFP transgenic mouse line WA (D,G,J,M) under normal (B,E,H,K) or anemic (C,D,F,G,I,J,L,M) conditions were stained with anti-GFP antibody (brown). The distal tubules (d) constitutively expressed the mutant transgene (E,F), but not wt-Epo-GFP (G). Arrows indicate the kidney interstitial cells, which expressed GFP only after the induction of bleeding anemia in both m1-Epo-GFP (F) and wt-Epo-GFP (G) transgenic mice. Arrowheads indicate the bile ducts, which were constitutively positive for GFP in the mutant Epo-GFP transgenic mice (H-I), but negative for GFP staining in the wt-Epo-GFP transgenic mice (J). Hepatocytes surrounding the central vein (*) expressed GFP only in anemic conditions in both m1-Epo-GFP (I) and wt-Epo-GFP (J) transgenic mice. The bronchial epithelium (★) was also positive for GFP antibody staining only in the mutant Epo-GFP transgenic mice (K,L). # indicates, interlobular triad. Scale bars are 300 μm (B-D), 20 μm (E-G), and 100 μm (H-M).

We established 3 transgenic mouse lines each for the m1-Epo-GFP (lines 1A, 1B, and 1C) and m2-Epo-GFP (lines 2A, 2B, and 2C) constructs (Figure 1D) and evaluated GFP expression under anemic conditions. To our surprise, GFP was highly expressed in the distal tubules and collecting ducts of the m1-Epo-GFP mutant transgenic mice, even in nonanemic conditions (Figure 5B,C,E,F). This ectopic expression was reproducible in all 3 lines of m1-Epo-GFP mutant transgenic mice (data not shown). GFP was also expressed in both normal and anemic conditions in the bile duct epithelia, bronchial epithelia, and thymus epithelia of the m1-Epo-GFP mutant transgenic mice (Figure 5H,I,K,L; Table 1), however the intensity of expression varied among the lines. This ectopic expression in nonanemic conditions was also observed in m2-Epo-GFP mutant transgenic mice and the expression did not change substantially among the lines (data not shown).

View this table:
Table 1

Expression of Epo-GFP transgenes and GATA factors

Upon induction of bleeding anemia in the both mutant Epo-GFP transgenic mice, GFP expression was induced in REP cells and in hepatocytes surrounding the central vein (Figure 5F,I; Table 1), as was the case for wt-Epo-GFP transgenic mice (Figure 5G,J). These results demonstrate that the GATA box is essential for repression of the Epo gene in epithelial cells in both nonanemic and anemic/hypoxic conditions. It should be noted that the GATA box–based repression appears to be abrogated for expression of the Epo gene in the authentic Epo-producing cells (ie, REP cells and hepatocytes).

GATA-2 and GATA-3 in epithelial cells bind to the GATA box

Using Gata2-GFP knock-in and Gata3-LacZ knock-in mouse lines,30,31 we examined which GATA family members are expressed in the epithelial cells that constitutively expressed the mutant Epo-GFP transgenes. As described in the previous paragraph, constitutive ectopic expression of GFP from the mutant Epo-GFP transgenes was detected in epithelial cells of the kidney distal tubules and collecting ducts (Figure 5E,F; Table 1). In kidney, both Gata2 and Gata3 were expressed in the distal tubules and collecting ducts, as seen by GFP immunostaining of Gata2-GFP knock-in mouse kidney (Figure 6A,B) and LacZ staining of Gata3-LacZ knock-in mouse kidney (Figure 6C,D). Immunohistochemical staining with anti–GATA-4 antibody of wild-type mouse kidney showed that the distal tubules and collecting ducts were negative for GATA-4 expression. Both Gata2-GFP and GATA-4 were detected in REP cells (Figure 6B,F arrows; Table 1).

Figure 6

Expression of GATA factors in renal tubular cells and binding of the Epo gene to the GATA box promoter. (A,B) Immunohistochemical staining of GFP in the kidney of a Gata2-GFP knock-in mouse. (C,D) X-gal staining in the kidney of a Gata3-LacZ knock-in mouse. (E,F) Immunohistochemical staining with anti–GATA-4 antibody in the kidney of wild-type mouse. Note that the distal tubules (d) and collecting ducts are positive for both Gata2-GFP (brown) and Gata3-LacZ (blue), but negative for GATA-4. On the other hand, the expression of Gata2-GFP and GATA-4 was detected in REP cells (arrows in B,F). (G) Chromatin immunoprecipitation (ChIP) assays of GATA-2 and GATA-3 and the Epo promoter. Chromatin complexes from the normal renal medulla were immunoprecipitated with anti–GATA-2 or anti–GATA-3 antibodies and the presence of Epo, Aqp2, and Gata1 gene promoter fragments was examined by PCR. Preprecipitated samples (input) were used as the internal positive controls for PCR. Normal rabbit immunoglobulin G (IgG) was used as a negative control. Scale bars are 300 μm (A,C,E) and 20 μm (B,D,F).

The specific binding of these GATA factors to the GATA sequence in vivo was investigated by a chromatin immunoprecipitation (ChIP) assay, exploiting the samples from the normal renal medulla where most of the cells expressed the mutant Epo-GFP transgenes and Gata2 and Gata3 genes. We found that the promoter region of the Epo gene did indeed immunoprecipitate with anti–GATA-2 and anti–GATA-3 antibodies (Figure 6G top panel). The promoter region of the Aquaporin-2 (Aqp2) gene was also immunoprecipitated. The Aqp2 promoter served as a positive control, as it is known to be a target of the GATA factors in renal tubules (Figure 6G middle panel).37 Binding of GATA-2 and GATA-3 to the Epo promoter seemed to be specific, since we did not detect the Gata1 gene promoter containing GATA sequence in the ChIP assay (Figure 6G bottom panel).34 These results support our contention that GATA-2 and GATA-3 constitutively repress Epo gene expression in renal tubular cells through binding to the GATA element. As for the GATA factors in the other epithelial cell lineages that were positive for the mutant transgene expression, we suggest concomitant GATA factor expression (Table 1). While GATA-6 expression is not examined in this study, it has been reported that GATA-6 is expressed in the bile duct and bronchial epithelium.3840 Although GATA-3 expression was not detected in the thymic epithelium in this study (Table 1), expression and essential contribution of GATA-3 to the thymic epithelium has been found (T. Moriguchi and J.D. Engel, personal oral communication). Based on these observations, we conclude that the renal tubular cells and certain other epithelial cells have the potential to express Epo in the absence of anemic or hypoxic stresses, and that GATA factors constitutively inhibit Epo gene expression in these cells.


To address the mechanisms underlying the regulation of Epo gene expression in a tissue-specific manner and in response to anemia/hypoxia, we used a BAC-based GFP reporter transgenic mouse strategy, and the coding region of the Epo gene was replaced with GFP. GFP expression in wt-Epo-GFP BAC transgenic mice recapitulated the inducible and tissue-specific expression of the Epo gene. Importantly, our analysis identified REP cells in the kidney as the Epo-producing cells, which showed specific peritubular localization, a unique arboroid shape and neuronal marker expression. Furthermore, a novel regulatory mechanism of Epo gene regulation was discovered that used the GATA box.

RT-PCR analysis demonstrated that the wt-Epo-GFP transgene faithfully recapitulated endogenous Epo gene expression in mouse tissues under both normal and anemic/hypoxic conditions. Previous Epo transgenic mouse studies revealed that a 33-kb genomic fragment of the human Epo transgene gave rise to hypoxia-inducible expression of human Epo in the kidney and liver.12,13,15 In most of the mouse lines, the transgene expression was accompanied by ectopic expression and also resulted in polycythemia, indicating that the transgene used did not contain the full regulatory region to stably control Epo gene expression with cell type–specific manner in vivo. On the contrary, the BAC-based wt-Epo-GFP transgene used in our study not only reiterated the endogenous Epo gene expression, but avoided the ectopic expression of Epo observed in other tissues and cell types. We therefore concluded that the 180-kb genomic region in the BAC contained virtually all the necessary regulatory elements needed for Epo expression.

We identified and characterized REP cells in the kidney using wt-Epo-GFP transgenic mice. REP cells are found in the interstitial space between tubules and capillaries, showing good agreement with previous reports.13,1923 The stellar or arboroid shape of the REP cell expresses the neural cell marker genes, and forms clusters in the anemic kidney. These observations suggest that the reticular network of REP cells may contribute to the regulation of Epo production by sensing a low oxygen tension.

A widely believed idea in need of validation is that Epo must be produced by specifically sensing the cellular oxygen tension. Epo-producing cells are localized in the most hypoxic regions of the kidney and liver, supporting their ability to rapidly sense hypoxia.24 We found that the numbers of GFP-positive cells in the kidney and liver increased exponentially with escalating anemia, showing a counter-parallel relationship with the Hct. These data further implied that Epo production is controlled through the increase in the number of Epo-expressing cells rather than by the changes in the expression level of each Epo-producing cell.35 In contrast, the expression level of Epo appeared to be much higher in REP cells than in hepatocytes on the basis of GFP fluorescence intensity. This suggested a differential expression level of Epo dependent on the cell type. Indeed, most Epo in the adult body is produced by the kidney,2 even though the population of REP cells is minor compared with that of hepatocytes.

In this regard, it is interesting to note that Epo is also minimally expressed in several tissues besides kidney and liver. These Epo-producing tissues include the brain, heart, and testis (reviewed in Suzuki et al24 and Brines and Cerami41). However, we could not detect GFP expression from the wt-Epo-GFP transgene in any of these tissues. One plausible explanation for this discrepancy is that the expression level of Epo in these tissues was below the level of detection. That the 180-kb genomic fragment lacks specific enhancers important for Epo gene expression in these tissues seems unlikely, as Epo expression other than in the kidney and liver was not detected by our sensitive quantitative RT-PCR analysis. In fact, we previously demonstrated that restricted Epo receptor expression in hematopoietic lineages was sufficient to sustain mouse development.42

We previously reported that GATA factors repressed the production of Epo in cultured hepatoma cells under normoxic conditions through binding to the Epo gene promoter, and this repression decreased in response to hypoxic stress.3,10,11 To examine whether this GATA regulation is actually operating in vivo in response to hypoxia, a transgenic mouse study was conducted. Although an important feature of GATA-based repression of the Epo gene expression was also observed in the present study, this study further revealed the GATA box in the Epo gene promoter is required for the constitutive repression of gene expression in a set of epithelial cells, but it does not affect the hypoxia-inducible production of Epo in REP cells and hepatocytes. These differences in the results are most likely caused by the experimental systems exploited, and this argues that transgenic mouse analyses are important to clarify transcription regulation mechanisms operating in vivo.

As for the molecular mechanism underlying the function of GATA factors, we suggest one possibility, as depicted in Figure 7. In this scenario, we assume that there may be an enhancer in the vicinity of the Epo gene that is important for the expression of other genes in the epithelial cell lineage. GATA factors may act to insulate the regulatory influence of this presumptive epithelial enhancer by binding to the GATA box. In contrast, while we detected the expression of GATA-2 and GATA-4 in REP cells, these GATA factors did not appear to affect Epo gene expression through its promoter GATA box and GFP expression from the GATA mutant transgene was induced normally upon bleeding anemia. This observation indicates that the hypoxia-responsive enhancer can drive Epo gene expression regardless of GATA-based repression. In addition to the authentic Epo-producing cells (REP cells and hepatocytes; Figure 7A) and epithelial cells (Figure 7B), there is a third cell lineage in terms of Epo gene regulation. In these cells, Epo expression would be permanently silenced by epigenetic mechanisms that are independent of the GATA box and hypoxic threshold (Figure 7C).

Figure 7

Schematic model of cell type–specific and inducible Epo gene regulation by GATA factors. (A) Epo gene expression is controlled by cell type–specific and hypoxia-inducible enhancers. Both enhancers are required for inducible Epo gene expression in REP cells and hepatocytes. (B) In the epithelial cell lineage, the epithelial cell–specific enhancer constitutively stimulates Epo gene expression, but this is repressed through the GATA promoter motif. Therefore, epithelial cells do not express Epo, in spite of hypoxic conditions. (C) In other cell types, Epo expression may be permanently silenced by epigenetic mechanisms unrelated to GATA repression and the hypoxic threshold.

GATA factors have been found to activate transcription,43 but recently repressive gene regulation by GATA factors has also been discovered, including the observation that the PPARγ gene in preadipocytes is repressed by GATA-2 and GATA-3 in the immature stage, but that this repression is abrogated after cellular differentiation.44 In contrast, this study demonstrated that Epo gene expression in epithelial cells is constitutively repressed by GATA factors. This is a novel regulatory system exploiting the GATA box, and this element shows extremely high conservation among mammalian Epo gene sequences.45 We surmise that this novel mechanism may rely on the insulator-like functions of GATA factors.46

Finally, it should be noted that tumors can derive from renal tubular epithelia and are occasionally accompanied by polycythemia. Many of these tumors constitutively produce Epo.47,48 Since a single nucleotide mutation in the GATA box of the Epo-GFP transgene caused constitutive transgene expression in renal tubular cells, this study suggests that there may be some defects in the GATA signaling pathway in the epithelial tumor cells that leads to ectopic Epo gene expression. This observation led us to speculate that we may be able to develop an alternative treatment for Epo-dependent anemia by controlling GATA factor function in epithelial cells.

Supplementary PDF file available online.


Contribution: N.O. and N.S. designed the study, performed the research, generated transgenic mice, analyzed the data, and wrote the paper; K.K. assisted with RT-PCR analyses; T.N. analyzed data; S.I. analyzed the data and supervised the study; and M.Y. supervised the study, analyzed data, and wrote the paper.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Masayuki Yamamoto, Department of Medical Biochemistry, Tohoku University Graduate School of Medicine, 2-1 Seiryo-cho, Aoba-ku, Sendai 980-8575, Japan; e-mail: masi{at}


We thank Naomi Kaneko, Mitsuru Okano, Yuko Kikuchi, and Masako Yamagishi for their assistance on mouse maintenance, and Drs Doug Engel, Tania O'Connor, and Jon Maher for advice.

This work was supported by grants from the Ministry of Education, Science, Sports and Culture; the Ministry of Health, Labor and Welfare; and the ERATO project of JST Agency.


  • *N.O. and N.S. contributed equally to this work.

  • An Inside Blood analysis of this article appears at the front of this issue.

  • The online version of this article contains a data supplement.

  • The publication costs of this article were defrayed in part by page charge payment. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.

  • Submitted October 1, 2007.
  • Accepted January 14, 2008.


View Abstract