Isolation and Characterization of Renal Erythropoietin-Producing Cells from Genetically Produced Anemia Mice

Understanding the nature of renal erythropoietin-producing cells (REPs) remains a central challenge for elucidating the mechanisms involved in hypoxia and/or anemia-induced erythropoietin (Epo) production in adult mammals. Previous studies have shown that REPs are renal peritubular cells, but further details are lacking. Here, we describe an approach to isolate and characterize REPs. We bred mice bearing an Epo gene allele to which green fluorescent protein (GFP) reporter cDNA was knocked-in (EpoGFP) with mice bearing an Epo gene allele lacking the 3′ enhancer (EpoΔ3′E). Mice harboring the mutant EpoGFP/Δ3′E gene exhibited anemia (average Hematocrit 18% at 4 to 6 days after birth), and this perinatal anemia enabled us to identify and purify REPs based on GFP expression from the kidney. Light and confocal microscopy revealed that GFP immunostaining was confined to fibroblastic cells that reside in the peritubular interstitial space, confirming our previous observation in Epo-GFP transgenic reporter assays. Flow cytometry analyses revealed that the GFP fraction constitutes approximately 0.2% of the whole kidney cells and 63% of GFP-positive cells co-express CD73 (a marker for cortical fibroblasts and Epo-expressing cells in the kidney). Quantitative RT-PCR analyses confirmed that Epo expression was increased by approximately 100-fold in the purified population of REPs compared with that of the unsorted cells or CD73-positive fraction. Gene expression analyses showed enrichment of Hif2α and Hif3α mRNA in the purified population of REPs. The genetic approach described here provides a means to isolate a pure population of REPs, allowing the analysis of gene expression of a defined population of cells essential for Epo production in the kidney. This has provided evidence that positive regulation by HIF2α and negative regulation by HIF3α might be necessary for correct renal Epo induction. (282 words)


Introduction
Erythropoietin (Epo) governs mammalian erythropoiesis. Epo is a glycoprotein hormone mainly produced in the kidney and liver in response to changes in tissue oxygen tension. Epo regulates erythropoiesis by supporting the survival of erythroid progenitors and stimulating their differentiation and proliferation in bone marrow, hence increasing the oxygen-carrying capacity of blood [1]. Lack of Epo during mouse development leads to lethality at embryonic day 13.5 (E13.5) due to severe anemia [2] and over-or under-production of Epo results in polycythemia or anemia clinically [1]. Epo production is considered to be controlled primarily at the level of gene transcription and Epo gene expression is strictly regulated in a tissue/cell-specific and hypoxia/anemiainduced manner [3][4][5][6][7].
Several tissues have been reported to express the Epo gene; but the ability to produce substantial amounts of Epo during hypoxia/ anemia is restricted to the fetal liver and adult kidney [4][5][6][7][8]. The kidney plays a major role in oxygen sensing and contributes ,90% of plasma Epo in adult animals [9]. However, difficulties in identification and purification of the renal Epo-producing cells (REPs) have limited the understanding of the mechanism for controlling Epo production in kidney. REPs are frequently reported to be peritubular fibroblast-like cells in kidney [6,10,11]; and a hypoxia-dependent Epo-producing cell line derived from human renal cancer was also described recently to exhibit fibroblast-like phenotype [12]. However, further details remain to be elucidated [5,7,13].
Current knowledge of the molecular mechanisms of oxygensensing and renal Epo gene expression has been extrapolated mostly from in vitro studies in hepatoma cell lines [14][15][16]. These studies have suggested that hypoxia responsiveness of the Epo gene depends on an enhancer containing hypoxia-responsive elements (HREs) located in the 39 flanking region of the gene (39 enhancer), to which the hypoxia-inducible transcription factor (HIF) 1 binds. HIF1 is composed of two subunits, HIF1a and HIF1b. HIF1b is constitutively expressed, but HIF1a expression, almost absent in normoxia, is increased during hypoxia. Under normoxic conditions, HIF1a is hydroxylated at two proline residues by specific prolyl-4-hydroxylases (PHD1-3) that allow the E3 ubiquitin ligase von Hippel-Lindau (pVHL) to bind to HIF1a and mark it for proteasomal degradation. In addition, HIF1a is regulated by the aspargine hydroxylase factor inhibiting HIF1 (FIH1), which inhibits p300/CBP (CREB-Binding Protein) binding to HIF1a. The activities of PHD and FIH1 are basically dependent on cellular oxygen concentration and thus qualify as cellular oxygen sensors. Low oxygen tension causes inactivation of PHDs and FIH1, allows HIFa to accumulate, forms active transcription factor-complex HIF with HIF1b, recruits transcriptional cofactors, and initiates the transcription of hypoxia responsive genes including the Epo gene. Thus the PHD/pVHL/HIF system likes to be the oxygen-sensing pathway regulating Epo gene transcription [17].
However, recent clinical and in vivo studies have suggested a new layer of complexity to the mechanisms involved in the cellular response to hypoxia/anemia. Evidence from mouse models and hereditary erythrocytosis in humans has revealed that HIF2a rather than HIF1a plays a vital role in oxygenregulated erythropoiesis and renal Epo production is probably regulated by PHD2/pVHL/HIF2a pathway [13,18,19]. There are three HIFa family members: HIF1a, HIF2a, and HIF3a, which share a number of similarities e.g. DNA-binding sequence, oxygen-dependent hydroxylation. Unlike ubiquitously expressed HIF1a, expression of HIF2a and HIF3a is limited to several tissues [20]. Both HIF1a and HIF2a activate transcription, while HIF3a negatively regulates HIF1a and HIF2a activity [20][21][22]. There is no literature on HIF3a's role in hematopoiesis thus far.
In order to clarify the whole picture of Epo gene regulation, we have generated a panel of mouse lines. First, we genetically deleted the 39 enhancer (referred to as the Epo D39E allele) and showed that this enhancer is necessary for hepatic Epo expression during the perinatal period {E17-postnatal day 13 (P13)} but dispensable for renal Epo expression after birth. Mice homozygous for the targeted allele (Epo D39E/D39E ) are viable and fertile, but exhibit anemia during late-embryonic and newborn stages [23]. Then, using a 180-kb Epo transgene with a green fluorescent protein (GFP) reporter (Epo-GFP), we recapitulated tissue-specific, hypoxiainducible GFP expression in kidney and liver tissue of mouse. Mutation studies on the transgene indicated that GATA factors are required for suppression of ectopic expression of the gene, but not essential for the Epo gene induction in REPs [6]. Also, we developed GFP knock-in mice (Epo GFP/wt ) by homologous recombination in mouse embryonic stem cells (NS and MY, unpublished data). By examining these mouse lines, we identified GFP-labeled REPs as a population of peritubular interstitial cells in the kidney after birth.
Taken together, all these data in vivo strongly imply novel mechanism(s) and necessitate detailed studies on REPs to explore a specific oxygen-sensing pathway underlying the hypoxia-induced Epo production in the kidney.
Fluorescence activated cell sorting (FACS) of GFP expressing cells has been widely used for the isolation of hematopoietic stem cells in our laboratory [24,25]. To make the link between the molecular and cellular mechanisms of hypoxia-induced Epo expression, in this paper, we have addressed controversial issues in REPs using cell-sorting techniques. Our Epo-GFP transgenic mice are a source of REPs isolation, but a pretreatment to induce anemia is not always successful for stable GFP expression in the kidney [6]. We therefore generated Epo GFP/D39E mice, in which REPs were labeled with GFP as a result of neonatal anemia caused by genetic modifications. Taking advantage of the strong GFP expression in anemic newborns carrying the 39 enhancer deletion, we have purified by FACS a cell population responsible for anemia/hypoxia induced Epo expression in the kidney.

Generation of Epo GFP/D39E Mice
We first tried to define REPs by GFP fluorescence, which truthfully reflects endogenous Epo expression. We have established Epo GFP/wt mice by homologous recombination. In our Epo GFP/wt mice, the Epo locus was targeted by replacing the 39 part of exon 2 through exon 4 with GFP ( Figure 1A, middle panel). The mice homozygous for Epo GFP/GFP , which lacked a functional Epo gene, died around E13.5 from anemia, corresponding with the previous report on a Epo gene knockout mouse [2]. The heterozygous animals (Epo GFP/wt ) were healthy and fertile; and distinct GFP expression, mimicking the endogenous Epo expression pattern, was observed in the kidney and liver under hypoxia/anemia conditions (NS and MY, unpublished data).
For steady Epo induction, we set out to genetically produce an anemia model by taking advantage of our established Epo D39E/D39E mice ( Figure 1A, lower panel). As we previously reported in Epo D39E/D39E mice, deletion of the 39 enhancer provokes transient anemia at late embryonic and neonatal stages due to defect in hepatic Epo production and erythropoiesis. This anemic phenotype is recovered in 2 weeks after birth when major Epo production site switches from the liver to kidney [23]. Crossing an Epo GFP/wt heterozygote with a mouse homozygous for Epo D39E/D39E allele, we generated Epo GFP/D39E compound offspring. The compound mice basically showed a similar phenotype to that of their Epo D39E/D39E parents: anemia persisting after birth and recovered in the juvenile stage. Epo GFP/D39E newborns were severely pale compared with Epo D39E/wt littermates ( Figure 1B). At P4-6, the Hematocrit (Hct) of the Epo D39E/D39E and Epo GFP/D39E newborns were 22.765.3 and 18.062.0 (%), respectively ( Figure 1C). At the same stage, the Hct of the Epo D39E/wt control was 32.064.1%; this value is comparable with that of wild type (data not shown). Increased Epo mRNA could be detected in the kidney of the Epo GFP/D39E newborns by quantitative (q) RT-PCR (see below).

Cellular distribution of GFP expression in anemic neonatal kidneys
We then examined newborn (P4-6) kidneys to see GFP expression by immunostaining studies. In microscopic observation of the kidney section with anti-GFP immunostaining, fluorescence was minimal in control kidney samples from P4-6 Epo D39E/wt littermates, but was prominent in the kidneys from the Epo GFP/D39E mice (Figure 2A,B). GFP fluorescence was also directly detected in fixed Epo GFP/D39E kidney slices by confocal microscopy (data not shown). GFP signals were confined to the cortex-medulla junction and focally distributed along the curve of the kidney at P4-6 ( Figure 2B). GFP-expressing cells were stellate-shaped, nesting around proximal tubules in the deep cortical labyrinth and outer strip area of the kidney ( Figure 2C,D). These findings are consistent with previous descriptions [5,6,26,27] and indicate that the GFP expression reflects the endogenous expression of the Epo gene in kidneys of anemic Epo GFP/D39E newborns.
Cell type of the GFP-expressing kidney cells in Epo GFP/D39E newborns Kidney interstitium contains mainly fibroblastic, dendritic cells [28], and vascular endothelial as well as tubular cells have also been identified as the site of Epo gene expression in the kidney [5]. We therefore carried out marker studies on GFP-expressing kidney cells of P4-6 anemic newborns by dual immunostaining under confocal microscopy and FACS detection ( Table 1). Expression of platelet-derived growth factor receptor b (PDGFRb) ( Figure 3C) and Ecto-59-nucleotidase/CD73 ( Figure 4), but not CD31 ( Figure 3A), major histocompatibility complex class II (MHCII) ( Figure 3B), or E-cadherin (Movie S1), indicated that REPs are cortical fibroblast-like, but not tubular, endothelial, or dendritic cells. Our three-dimensional (3D) movie reveals that REPs are tightly packed around tubules, but not tubular cells themselves (Movie S1). We also examined the expression of alpha-smooth muscle actin (a-SMA), a marker for the subtype of renal interstitial fibroblasts [28]. Staining with a-SMA antibody was observed in the medullar and cortical cells near the capsule, but did not overlap with GFP-positive cells ( Figure 3D). Taken together, most neonatal REPs at P4-6 stage, are CD73-, PDGFRb-positive, a-SMA-negative peritubular fibroblasts (Table 1). Thus, neonatal REPs have the typical characteristics of adult cortical fibroblasts previously reported in healthy rat kidney [29].
Unlike GFP signals, which were restricted to peritubular interstitium, CD73 stains were widely spread in the renal cortex. Besides fibroblasts, proximal tubule (brush boarder), glomeruli (mesangial cells) and the cells in the medullar rays were also observed to express CD73 in sections. The proportion of CD73 expressing cells and their staining intensity in non-fibroblast cells were similar in both anemic and non-anemic kidney sections ( Figure 4C, D), i.e. among various types of CD73-expressing kidney cells, only cortical fibroblasts showed a hypoxia/anemiaresponsive tendency.

FACS sorting of GFP + cells and Epo mRNA expression in the isolated cells
We isolated GFP + cells from the kidneys of P4-6 Epo GFP/D39E neonates by flow cytometry. Kidney cells from Epo D39E/wt littermates were used as negative controls ( Figure 5B, upper-left panel). The GFP + population was present grossly with a low to intermediate GFP intensity and constituted up to 0.2% of the total fresh kidney cells ( Figure 5B, upper-right panel). This yielded several thousands viable GFP + cells per Epo GFP/D39E mouse at P4-6 newborn stage, with a purity of greater than 75%. We also assessed the association of GFP-expressing cells with CD73. As shown in Figure 5B, lower-right panel, of GFP + cells from P4-6 anemic kidneys, 63% were CD73 + and 37% were CD73 2 . A confocal microscopic image of the sorted GFP + cells is shown with anti-GFP immunostaining ( Figure 5C).
Subsequently, we evaluated the expression of Epo mRNA in each sorted fraction from P4-6 Epo GFP/D39E kidneys, and in unsorted kidney cells of P4-6 Epo D39E/wt and Epo GFP/D39E newborns as well. The hypoxanthine-phosphoribosyl-transferase (Hprt) gene was used as a loading control, because expression of HPRT is less affected by hypoxia/anemia [31]. As shown in Figure 5D, qRT-PCR revealed high Epo mRNA expression exclusively in samples from Epo GFP/D39E animals including unsorted kidney cells, CD73 2 , CD73 + and GFP + subsets. In Epo GFP/D39E mice, compared with the unsorted kidney, Epo mRNA levels were ,100 fold enriched in the GFP + fraction, but not in the CD73 + fraction. Epo mRNA expression was low but detectable in Epo D39E/wt kidney cells at P4-6, but not in the GFP-negative fraction of Epo GFP/D39E kidney cells ( Figure 5D). We have reported that neuronal markers are expressed by Epo-producing cells in the adult kidney [6]. Consistently, transcripts for microtubuleassociated protein 2 (MAP2) and neurofilament light polypeptide (NFL) were detected in the GFP + fraction from P4-6 Epo GFP/D39E kidneys (data not shown). These results demonstrate our system for isolation of REPs is reliable and efficient.

Gene expression profile of oxygen-sensing and HIFs of the isolated REPs
We examined the expression of molecules, known to be involved in oxygen tension-dependent regulation [19], by qRT-PCR analysis. Compared to GFP-negative kidney cells, no enrichment of mRNA expression of oxygen sensor genes, Phd1-3 and Fih1 genes was found in the REPs fraction ( Figure 6A). Hif2a but not Hif1a ( Figure 6B, upper panel) mRNA expression was upregulated in the REPs. In this line, no enrichment of HIF1a target genes [4], Pgk1 and Phd2, Phd3 were found in the REPs. These are consisted with recent reports on the relationship of HIF2a to renal Epo production [12,29,32,33]. Hif3a mRNA expression was also enriched in the REPs (( Figure 6B, upper right panel). In the three alternatively spliced variants of Hif3a mRNA: Ipas, Nepas and Hif3a, Nepas is expressed during infantile and newborn stages, while Ipas is seen in adults. In accordance with a reported method [34,35], we examined the expression of these splice variants in REPs, and detected all three variant mRNAs in the fraction of REPs at P4-6 stage. Nepas mRNA expression was the highest, suggesting Nepas might be a dominant form among the three splicing variants of the Hif3a at P4-6 infantile stage ( Figure 6C).

Discussion
By generating Epo GFP/D39E mice, we isolated a specific type of renal cells namely REPs, which are responsible for Epo production after birth. REPs 1) are fibroblast-like interstitial cells residing in the tubulo-interstitial compartment of the kidney in anemic hosts (Hct 18%); 2) constitute up to 0.2% of whole kidney cells. About 63% of REPs also express CD73, a marker for cortical fibroblasts and Epo-expressing cells in kidney [6,10,28,30]; 3) highly express Hif2a, Hif3a, but not Hif1a mRNAs; 4) are efficiently isolated from naive kidney tissues as GFP-expressing cells in our mutant.

Isolation system of REPs
Previously, difficulties in the isolation of REPs prevented better understanding of the mechanisms of Epo regulation in response to hypoxia. We developed an isolation system that phenotypically labeled REPs in the kidney, by way of a GFP knock-in (Epo GFP ) combined with a 39 enhancer (Epo D39E ). This enabled us to purify REPs, a rare cell population, from kidney by FACS-sorting.
Our GFP knock-in strategy facilitates the capacity to express GFP based on endogeneous Epo gene expression confined to a rare population, without worries about aberrant or ectopic transgene expression. As a result, we identified REPs as peritubular fibroblast-like interstitial cells concentrated at the cortico-medullary junction, corresponding to our previous finding from the Epo-GFP transgenic mouse studies [5,6].
In the history of Epo research, in situ hybridization is a classic method to localize Epo mRNA; but extensive studies lead to confusion of REPs' whereabouts: peritubular interstitium, or a tubular site [5]. Transgenic mice, created by integrating a marker gene with regulatory sequences of the Epo gene, have provided a powerful tool for accurate localization of REPs in the kidney [6,8,11]. In mice bearing an Epo/SV40 T antigen transgene, REPs were successfully identified but attempts to isolate REPs in vitro failed [8,11]. Importantly, later transgenic mouse studies suggested that a much wider (20-kb,) flanking region of the Epo gene was needed for adequate levels of transgene expression in kidney [5,6,36].
To induce Epo gene expression, pre-treatment to induce hypoxia/anemia is usually required. These procedures, such as bleeding or phenylhydrazine injection, are not always successful in inducing stable anemia. For instance, we have tried isolation of REPs using our Epo-GFP transgenic mouse [6]. The GFP + population of the kidney cells from the transgenic mouse were not distinct under FACS detection, despite that severe anemia was induced by phlebotomy. Epo GFP/D39E mice provide a handy and important source: lacking the 39 enhancer demonstrated impaired hepatic Epo expression and profound anemia (Hct value was about 18% in newborn stage P4-6) and allowed us to directly sort REPs by FACS, which are constantly and stably labeled with GFP fluorescence in newborn kidneys. Moreover, considering the loose connections of renal tissues and the decreased interstitial volume of the cortex, renal tissues from newborns promised to be a better source of this rare cell population than adult kidney [37]. What is important that newborn REPs are fully functional with regard to Epo secretion, soon after birth [9,23]. Indeed, newborn REPs (P4-6) displayed adult REPs phenotypes: fibroblast-like (CD73 + /a-SMA 2 ) [28,29] with neuronal marker expression (MAP2 + /NFL + ) [6].  Cellular characters of the GFP-labeled REPs GFP coupled with marker molecule analyses revealed that REPs are localized in the deep renal cortex, form a network around tubules and adjacent capillaries with their processes, and express fibroblast markers CD73, PDGFRb and soluble guanylyl cyclase (sGC) (data not shown) but not other cell markers CD31, MHCII or E-cadherin. Surface expression of these molecules on REPs was verified by both confocal microscopy and flow cytometry. Therefore, all of these data support a widely accepted notion that REPs are peritubular fibroblast-like interstitial cells [6,10], and resolves earlier conflicts on the cell identity in the literature [5].
Fibroblasts are considered to be an easy cell type to cultivate. However our attempts to culture REPs were not successful. It is interesting to consider that REPs are a type of fibroblasts in a resting state, and when induced to proliferate (signified by a-SMA expression), they lose their ability to produce Epo [38,39], such as the case for renal fibrosis and renal anemia. Concurring with our previous finding that adult REPs express neuronal markers [6], we confirmed Map2 and Nfl mRNA expression in FACS-sorted REPs from P4-6 newborns. These observations further support the notion that REPs are unique fibroblast-like cells.

Association with CD73 during anemia
Our histological and flow-cytometrical examinations revealed that REPs constitute grossly 0.2% of total kidney cells, with 63% co-expressing CD73 in the case of our Epo GFP/D39E mice (Hct 18% Figure 5. Isolation of REPs from the P4-6 Epo GFP/D39E anemic kidneys. A Cell purification protocol for a rare population of REPs from kidney. B, upper panel Representative FACS scatter plot of kidney cells from pooled P4-6 Epo GFP/D39E neonates demonstrating GFP + cells above the gate (set using Epo D39E/wt control cells). B, lower panel Assessment by FACS of percentage of GFP + and CD73 + fractions in the kidney cells from P4-6 anemic Epo GFP/D39E neonates. Pooled results were from three independent experiments. Note CD73 expression divides GFP + cells into two parts: CD73 + and CD73 2 . C Representative confocal microscopy of the FACS-sorted GFP + cells (REPs) with anti-GFP immunostaining. Scale bar: 5 mm. D Analysis of relative Epo mRNA levels by qRT-PCR (Hprt as a loading control) in FACS-purified GFP + or the remaining GFP 2 kidney cells from P4-6 Epo GFP/D39E mice; CD73 + vs. CD73 2 fractionated cells were also evaluated. The data shown are from four experiments, each performed in duplicate. nd: not detectable; *p,0.05. doi:10.1371/journal.pone.0025839.g005 at P4-6 newborn stage). In the non-anemic kidneys, only a few REPs, composed of both CD73 + and CD73 2 , can be observed in the juxtamedullary layer of the cortex. These cells probably represent a basal level of Epo production under normal conditions, which is required for daily production of red blood cells in normal individuals. In anemic kidneys both CD73 + and CD73 2 REPs are robustly increased in a pattern that spreads outward from the deep cortex toward the capsule and the inner medullar.
qRT-PCR revealed a 100-fold enrichment of Epo mRNA in the GFP + fraction, but no enrichment in the CD73 + fraction, compared with the unsorted total kidney cells. CD73 has a wide expression profile in renal cortex, occupying roughly 3% of the total kidney cells at P4-6 newborn stage based on our FACS study. In addition to the interstitial cells, such as fibroblasts, T-, Blymphocytes, many parenchymal cells e.g. glomorular mesangial, proximate tubular (brush board), collecting duct cells etc. also Figure 6. Expression profile of cellular oxygen-sensing and hypoxia-inducible molecules in the isolated REPs. qRT-PCR analysis of the genes related to hypoxia response using RNA extracted from the sorted GFP 2 and GFP + renal cells from P4-6 Epo GFP/D39E mice. Hprt as a loading control *p,0.05. Data are for three experiments performed in duplicate. A Four genes related to cellular oxygen-sensing; B three Hifa isoforms. Also note that HIF1a targets: Pgk1, Phd2 are not enriched in the GFP + fraction (REPs). C Three transcript variants of Hif3a in the purified GFP + fraction (REPs), using a method for exponential PCR amplification at a fixed threshold (see also Materials and Methods). doi:10.1371/journal.pone.0025839.g006 express CD73 [28]. Comparing the pattern of CD73 staining in the anemic kidney section with the non-anemic one, it seems that anemia increases the number of CD73 + cortical interstitial fibroblasts but not other types of CD73 + cells. Because cortical fibroblasts are rare population, among all of the CD73 + cells, the percentage of CD73 fraction did not change much in anemic kidneys compared with non-anemic kidneys. The heterogeneity of CD73 expression in REPs may reflect different functional or matured stages during anemia [39]]. Recently, it has been reported that loss-of-CD73 does not affect the expression of Epo [40]. Our data that only a part of REPs express CD73 in anemic kidneys seem to conform this finding that the loss of CD73 has not impact on renal erythropoietin induction under hypoxia.

Expression profile of oxygen sensor molecules and transcriptional determinants in REPs
Clinically, congenital defects of the oxygen-sensing pathway have been reported including VHL, PHD2 and HIF2A mutations that cause secondary erythrocytosis through the EPO gene overexpression [1,7]. PHD2 inactivation is sufficient to induce near max. renal Epo production [41,42]; and recent RNAi-based studies confirmed the major role of PHD2 in Epo regulation in vitro as well as in vivo [43]. Human and rat PHD2 mRNA are hypoxically induced by HIFs for negative feedback regulation [44,45]. In the carbon monoxide exposed rat, PHD3 protein was detected and co-localized with HIF2a in cortical interstitial cells of the kidneys [46]. Phd2, 3 are the targets of HIF1a [4]. In this study, we examined four genes encoding oxygen-sensor molecules (PHD1-3 and FIH1) and did not observe any enrichment in REPs compared with other cells of the anemic kidneys in our gene expression profiling. This may be because all of these four genes are ubiquitously expressed in kidney, with respect to various cell types of the kidney.
As described above, Hif2a, rather than Hif1a shows highly REPs-specific expression patterns at the mRNA level. Interestingly, Hif2a mRNA levels are particularly high in tissues that are important for the systemic delivery of oxygen, for example the lung, heart, endothelium and the carotid body [47][48][49]. Quite recently, HIF2a protein expression has been shown in the peritubular fibroblasts that express Epo and CD73 in rat kidneys [30]. Preferential binding of HIF2a protein to the HRE within the native Epo gene 39 enhancer has been also confirmed in hepatocytes [50]. As renal Epo expression does not depend on the Epo gene 39 enhancer [23], the existence of a possible renal enhancer with a different HRE awaits investigation.
Enrichment of Hif3a mRNA was also observed in the REPs. Transcripts of all of the three splicing variants (Nepas, Ipas and Hif3a) of Hif3a could be detected in newborn REPs (P4-6), where Nepas seems to be the dominant form of the three. Nepas and Ipas have been demonstrated to be hypoxia-induced factors due to the presence of functional HREs upstream of Exon1a, and act as negative regulators of the HIF pathway. Both Ipas and Nepas show a cell-, and stage-specific expression pattern [21,22]. IPAS (inhibitory PAS protein) has already been reported to work as a negative feedback factor in a hypoxic condition in the cornea [21,51], but there is no literature on its roles in hematopoiesis so far. Our targeted Hif3a knockout mice (Hif3a 2/2 ) show an impaired cardiovascular formation around birth. This phenotype is possibly caused by over-expressed Endothelin-1 in pulmonary endothelium. HIF3a (Nepas) was suggested to suppress HIF2adriven transcription of Endothelin-1 according to the localization and reporter assays [22].
We were curious to see if a similar mechanism exists in Epo gene regulation. We are starting to explore the function of HIF3a in erythropoiesis by examination of our established Hif3a 2/2 mice [22]. Hif3a 2/2 mice were viable and fertile without abnormalities under normal conditions. Based upon our preliminary data in mouse hypoxia experiments, it appeared that Epo transcript showed up-regulated tendency in Hif3a 2/2 kidneys, in contrast to the wild type counterparts. In a recent review, McIntosh et al. also mentioned an erythropoietic phenotype in their independent Hif3a 2/2 mice [52]. We, therefore, hypothesized HIF3a-related negative regulation is also necessary in renal Epo production during hypoxia/anemia. By this, homeostasis of red blood cell mass might be maintained to prevent erythrocytosis and thrombosis occurring in animals and human beings. HIF response to hypoxia is complex. A recent report has demonstrated that human HIF3A gene expression is induced by hypoxia through activation of HIF1a but not HIF2a [53]. It raises the possibility that in REPs, Hif3a mRNA expression might be up-regulated by HIF2a, because REPs preferentially express Hif2a rather than Hif1a.
Recently a renal cell line producing Epo with a hypoxiadependent manner has been successfully established from a patient suffering from renal cancer [12]. Here, we report for the first time on isolation or purification of REPs in vivo. Our mouse enables the purification of a rare cell population specific for renal Epo expression during anemia and a detailed examination of the hypoxia-dependent aspect of the cells. Finally, we report the novel finding that Hif2a and Hif3a (but not Hif1a) mRNA are preferentially expressed in REPs. Combined with recent evidence in vivo about the role of HIF2a in erythropoiesis, we propose a hypothesis: positive regulation by HIF2a and negative regulation by HIF3a may be necessary for correct renal Epo induction during hypoxia/anemia.

Materials and Methods
Generation of Epo GFP/D39E mice All mice used were from a C57BL/6 genetic background and were strictly kept in the specific-pathogen-free conditions. All experiments were conducted in accordance with the regulations of The Standards for Human Care and Use of Laboratory Animals of Tohoku University. The protocol was approved by the Committee on the Ethics of Animal Experiments of Tohoku University (Permit Number: 21-Idou-144 and 22-Idou-113).
Epo GFP/D39E mice were generated by mating mice heterozygous for Epo GFP/wt with mice homozygous for deletion of the 39enhancer (Epo D39E/D39E ) [23]. Genotyping was performed by polymerase chain reaction (PCR) with the primer sets listed in Table 2. From this mating, half of the offspring would be Epo GFP/D39E mice. These mice are genetically deficient in Epo gene 39enhancer activity and had anemia within two weeks after birth.

Hematological analysis
Whole blood was collected from the carotid arteries, and hematopoietic indices were measured using an automatic blood cell analyzer (Nihon Koden).

Immunostaining
Kidneys were immersion-fixed in 4% paraformaldehyde (Nakarai Tesque) for 3 hours at 4uC and embedded in OCT compound (Sakura Finetechnical). Frozen sections 20 mm in thickness were incubated with primary antibody for 16 hours at 4uC, and detected by Alexa Fluor 488 (Molecular Probes) or Alexa Fluor 555 (Molecular Probes) conjugated anti-IgG as second antibodies. Color detection was performed using diaminobenzidine as a chromogen (brown color staining). Nuclei were stained with 49-diamidino-2-phenylindole (DAPI). Fluorescent images were observed using the LSM510 confocal imaging system (Carl Zeiss).

FACS analysis and cell sorting
Epo GFP/D39E anemic newborns were sacrificed at P4-6. The kidneys were collected in PBS from several litters and teased away from their surrounding tissues. Single cell suspension was prepared using dispase (1.25 mg/mL; Invitrogen) in PBS-15% FCS for 60 min at 37uC followed by washing in DMEM-Ham's F-12-10% FCS and passing through a nylon mesh to remove any clumps. This cell preparation was approximately 85% single viable cells. Whole kidney was analyzed and sorted on the flow cytometer (FACSAria, Becton Dickinson). The effectiveness of each FACS separation was assessed by immediately resorting an aliquot of GFP + and GFP 2 cells (data not shown). Greater than 75% of the GFP + population resorted to the same gate used in the initial sort.

qRT-PCR analysis
Total RNA was extracted from FACS-purified cells using Isogen reagent (Nippon Gene), according to the manufacturer's protocol. RNA was then concentrated using RNeasy MinElute columns (Qiagen) and first strand cDNA synthesis was performed using the SuperScript III First Strand Synthesis System for RT-PCR (Invitrogen).
Primers for amplifying 100-300 bp of each PCR product were used ( Table 2). PCR reactions were SYBR Green programmed and carried out using qRT-PCR Mastermix (Takara). Each sample was analyzed in duplicate or triplicate. The data were normalized by subtracting the difference of the C T values between the target genes of interest (Tgene) and that of Hprt mRNA, thereby obtaining a DC T (Tgene C T 2HPRT C T ). Relative expression (fold induction) was calculated as 2 2(SDC T 2CDC T ) where SDC T 2CDC T is the difference between the sample DC T (GFP + cells) and the control DC T (GFP 2 cells). Both target gene and Hprt reactions approached 100% efficiency as determined by standard curves. PCR products were analyzed by dissociation curve and on agarose gels to check that a single band was amplified.
The molar ratio was calculated as previously described [33,34]: molar ratio = [L a 6(1+E a ) CTa ]/[L b 6(1+E b ) CTb ]. L a and L b indicate lengths of the amplicon for 1a and 1b transcripts, respectively. E a and E b indicate the amplification efficiency of a primer set for 1a and 1b transcripts, respectively. CT a and CT b indicate the numbers of threshold cycles for the 1a and 1b transcripts, respectively.

Statistics
Statistical analysis was performed between samples and controls using t-test (two tailed, unequal variance, p#0.05 cut-off).

Supporting Information
3D-movie of REPs and their spatial coordination, were made by compiling images collected using a Zeiss LSM 510 confocal microscope. One z-slice of the stack is shown in Movie S1.

Supporting Information
Movie S1 3D image of REPs by confocal laser-scanning microscopy. Kidney sections from P5 Epo GFP/D39E newborn were co-stained with anti-E-cadherin to label the tubular cells. GFP: green; E-cadherin: white; DAPI: blue/nucleus. Scale bar: 20 mm. (MOV) Author Contributions