Neomorphic effects of the neonatal anemia (Nan-Eklf) mutation contribute to deficits throughout development

Direct activation of tissue-restricted gene expression relies on transcription factors that faithfully recognize their cognate binding sequence in DNA and influence the transcriptional and epigenetic outcome (Novershtern et al., 2011). Of special importance for erythropoiesis is EKLF/KLF1, which exerts a global role in tissue-specific gene regulation by virtue of its three-finger recognition of the 9 bp 5′CCMCRCCCN sequence at target gene promoters and enhancers (reviewed by Siatecka and Bieker, 2011; Tallack and Perkins, 2010; Yien and Bieker, 2013). Its effects are circumscribed owing to its tissue-restricted expression in erythroid cells and its bipotent myeloid progenitor. Genetic ablation studies in the mouse show that EKLF is absolutely essential because death results from a profound β-thalassemia and the low to virtually non-existent expression of global erythroid genes of all categories.

Monoallelic mutation of KLF1 in humans can lead to a benign outcome that is nonetheless phenotypically important (Borg et al., 2011; Helias et al., 2013; Singleton et al., 2012; Tallack and Perkins, 2013; Waye and Eng, 2015), as haploinsufficient levels of KLF1 lead to altered β-globin switching, which can be advantageous in individuals with β-thalassemia (Liu et al., 2014). However, some mutations lead to anemias (Arnaud et al., 2010; Huang et al., 2015; Jaffray et al., 2013; Singleton et al., 2011; Viprakasit et al., 2014; reviewed by Perkins et al., 2016). The human KLF1 mutation (E325K) in congenital dyserythropoietic anemia (CDA) (Arnaud et al., 2010; Jaffray et al., 2013; Singleton et al., 2011) is at the same amino acid as that seen in the mouse Nan mutant (Heruth et al., 2010; Siatecka et al., 2010b), albeit a different substitution. Nan is inherited in a semi-dominant fashion: homozygotes die in utero at E10-11, while heterozygous Nan/+ mice exhibit lifelong severe anemia that is characterized by reticulocytosis, splenomegaly, altered globin expression and cell membrane defects (Lyon, 1983; Siatecka et al., 2010b; White et al., 2009). Transfer of Nan mouse anemic hematological properties after bone marrow transplantation of Nan/+ cells to a wild-type recipient shows the dominant, cell-autonomous nature of the effect (Lyon, 1983; White et al., 2009).

The mutation in zinc finger 2 at E339D, although conservative, is at a universally conserved amino acid and leads to altered recognition properties such that Nan-KLF1 recognizes only the 5′CCMCGCCCN subset of wild-type sequences and leads to a distorted genetic output (Siatecka et al., 2010b). Thus, there are two types of binding sites: those able to bind both wild-type KLF1 and Nan-KLF1 (category I) and those able to bind only wild-type KLF1 (category II) (Siatecka et al., 2010b). Paradoxically, the mutant Nan-KLF1 form interferes with the expression of target genes with category II sites, to which it can no longer bind. The molecular explanation behind this remains unclear. However, the change also confers recognition of an altered sequence that is not normally associated with KLF1 recognition (Gillinder et al., 2016), such that genes that are never seen in the erythroid cell are ectopically expressed in the Nan/+ mutant.

An incompletely explained aspect of the Nan/+ mouse is that its anemia is wide-ranging during development in both the embryo and the adult, and may not be completely accounted for by the distortion in the intrinsic pattern of expression in the Nan/+ erythroid cell. This is directly relevant to the humans with CDA, who also exhibit a range of effects that appear extrinsic to those caused by a tissue-restricted factor such as KLF1. We have addressed these issues by testing the hypothesis that ectopic expression of a specific subset of genes leads to effects that are superimposed on the red cell defects and thus plays a significant contributory role to the Nan anemia phenotype.

Inspection of globally expressed genes whose levels are altered in Nan/+ E13.5 fetal livers compared with their wild-type littermates (Gillinder et al., 2016) support the initial genotypic analysis of the Nan mouse (Siatecka et al., 2010b), showing a dysregulated erythroid transcript pattern in spite of the heterozygosity of the KLF1 mutation. For example (Fig. 1A), expression of dematin (Epb4.9, also known as Dmtn) and E2f2 are virtually absent in KLF1 knockout cells, and this is mimicked in Nan/+ cells; these contain category II elements that are affected by the presence of Nan-KLF1. As a control, Alas2, which contains a category I-binding element, is not altered in Nan/+ cells compared with wild type. Finally, hemoglobin switching is altered, as embryonic Hbb-bh1 globin expression is retained at high levels in both knockout and Nan/+ E13.5 fetal liver, whereas adult Hbb-b1 and Hbb-b2 globin expression are 50-70% less (Fig. 1B), as previously observed (Siatecka et al., 2010b). Although many of these changes arise from the more-limited target DNA sequence recognized by Nan-KLF1 protein (i.e. 5′CCMCGCCCN) compared with wild type (i.e. 5′CCMCRCCCN) (Siatecka et al., 2010b), our recent studies show that there is a second consequence of the E339D mutation in Nan-KLF1, namely that a different recognition sequence (5′CCMNGCCCN) is newly recognized (Gillinder et al., 2016). We considered whether this might play an additional role in the anemia and the phenotypic properties observed throughout development in Nan/+ embryos and adults, specifically whether the new targets would give rise to aberrant expression of unusual/unexpected proteins in Nan erythroid cells that yield deleterious effects.

We began by focusing on the relatively small number (∼80) of genes that are newly and most highly expressed in Nan/+ cells compared with wild type (Gillinder et al., 2016), and further parsed this to genes that are secreted and thus could systemically affect normal erythropoiesis via the circulation. Two genes in particular caught our attention. Hepcidin is a regulator of cellular iron use by virtue of its interaction with ferroportin (Ganz, 2011), and is primarily expressed in the adult liver. Interferon regulatory factor 7 (IRF7) is a transcription factor primarily expressed by macrophages that mediates inflammation by virtue of its activation of interferon β (IFNβ) (Ford and Thanos, 2010; Honda et al., 2005). Misexpression of either of these would be detrimental to effective erythropoiesis (Ganz and Nemeth, 2012; Kim and Nemeth, 2015; Nagata, 2007; Yoshida et al., 2005a). Our data show little expression in the wild-type or knockout cells or in the wild-type cells from the Nan strain; however, their levels are dramatically increased in the Nan/+ cells (Fig. 1C) [also observed in an independent experiment (Fig. S1A), thus providing us with data from a total of 12 biological replicates]. These are not the result of global or regional changes; in each case, expression of the adjacent genes remain unaffected irrespective of the genotype (Fig. 1C). Total KLF1 levels are only slightly lower in Nan/+ compared with wild type (Fig. 1D and Fig. S1B).

Although EKLF expression is restricted to erythroid progenitors and their progeny, we verified that the ectopic expression results from the erythroid compartment by depleting adherent cells and expanding E13.5 fetal liver cells in erythropoietin. qRT-PCR analysis of RNA shows that these erythroid-enriched cells express KLF1 at similar levels, that Nan/+ cells exhibit the dramatic decrease in dematin (Epb4.9) expression previously seen with total fetal liver material, and that there is dysregulation of hepcicin (Hamp) and Irf7 (Fig. 1E). We conclude that Hamp and Irf7 are mis-expressed in the erythroid cell by virtue of the presence of the Nan mutant variant of KLF1.

Hepcidin levels are suppressed by erythroferrone (Fam132b) (Kautz et al., 2014), a gene whose expression in the erythroid cell is stimulated by erythropoietin when iron mobilization is desired. We found that erythroferrone is regulated by KLF1 levels in the erythroid cell, as its expression is virtually absent in the knockout (Fig. 2A). Of relevance here, its levels are also dramatically deficient in Nan/+ cells (Fig. 2A,B; Fig. S1C); as with other similarly affected genes (Siatecka et al., 2010b), this is not due to haploinsufficient levels of wild-type EKLF (not shown). The two likely KLF1-binding sites in this gene are category II (Siatecka et al., 2010b), and thus are predicted to be negatively affected by the presence of the Nan variant.

The implication that erythroferrone is a KLF1 target was directly tested using the KLF1-null rescue system with which stably expressed cytoplasmic ER/KLF1 or ER/ Nan-KLF1 can be induced to translocate to the nucleus by treatment of the cells with tamoxifen (Coghill et al., 2001; Hodge et al., 2006). Although the treatment protocol is limited to 1 h, these rescued cells show robust chromatin association by ER/KLF1 to one of these sites (a category II site in intron 1) but not by ER/Nan-KLF1 (Fig. 2C). In addition, RNA begins to accumulate within this short time period only in the wild-type rescue (Fig. 2D).

We found additional evidence for disarray in iron utilization components within the Nan/+ erythroid cell (Gillinder et al., 2016). For example, when comparing Nan/+ cells with wild type we see a slight decrease in transferrin receptor (Tfrc) levels, and a slight increase in ferroportin (Slc40a1) in spite of the increase in hepcidin (Fig. 3A). This yields a lower level of iron in the red cell (Ganz and Nemeth, 2012; Tolosano, 2015). Coupled to this, heme oxygenase (Hmox1) levels increase while heme transporter HRG1 (Slc48a1) levels decrease in the Nan/+ cells compared with wild type (Fig. 3B), yielding less heme in the erythroid cell (Korolnek and Hamza, 2015; Narla and Mohandas, 2015).

In light of this, we examined expression patterns of other genes involved in iron utilization or in the inflammation pathway (reviewed by Ganz, 2011; Ginzburg and Rivella, 2011; Guida et al., 2015; Kim and Nemeth, 2015; Li et al., 2014; Nagata, 2007; Rivella, 2015; Yoshida et al., 2005b) to address indirect or Nan-KLF1-independent causes for the inefficient erythropoiesis (Gillinder et al., 2016). Expression is either unchanged [hemochromatosis (Hfe), IRP1 (Aco1), aconitase (Aco2), Flvcr1 (Mfsd7b); Fig. S2A] or remains insignificantly expressed [Bmp6, hemojuvelin (Hfe2), Gdf11, Pigf; not shown]. Genes involved in regulation of hepcidin, such as Tmprss6, are not altered (Fig. S2B) or not expressed (Gdf15, Atoh8 and Sox2; not shown), although the level of Stat3, an upstream regulator of hepcidin, is increased in Nan/+ cells (Fig. S2B) in spite of undetectable Il6 (not shown). Levels of erythroid regulators such as Twsg1, HIF2α (Epas1), or IRP2 (Ireb2) are not altered (Fig. S2C). With respect to genes involved in a predicted inflammatory response due to Irf7 expression, we see no change in expression of Irf3, Irf2 or Tlr2 (Fig. S2D), and no expression of Tlr6 (not shown). We conclude that the misexpression of hepcidin and IRF7, and the depletion of erythroferrone expression in the red blood cell are not the result of a generalized response to external stimuli but rather are caused by the intrinsic presence and inappropriate binding of Nan-KLF1.

Embryos that express only the Nan-EKLF variant (i.e. Nan/Nan or Nan/−) are lethal by E10-11 and display a striking dysmorphology that extends beyond the fetal erythroid cell (Heruth et al., 2010). To model these early stages, we established embryonic stem cells(ESCs) from Nan/Nan and +/+ littermates and differentiated them to embryoid bodies (EBs). EBs synchronously recapitulate erythropoiesis (Choi et al., 2005; Kennedy and Keller, 2003); in our hands, the onset of Klf1 and primitive erythropoiesis begins at day 4 and peaks by day 6, when definitive erythropoiesis is well underway to eventual dominance (Adelman et al., 2002; Bruce et al., 2007; Lohmann and Bieker, 2008). Nan/Nan EBs demonstrate morphological defects by day 6 of differentiation, as they are smaller in comparison with +/+ EBs (Fig. 4A). Although the onset of the CD44 early definitive erythroid cell surface marker (Chen et al., 2009; Liu et al., 2013) is normal in both sets by day 6 (Fig. 4B, left), Nan/Nan EBs are defective in generating late (primitive or definitive) erythroid cells, as judged by the absence of Ter119 expression in these cells at day 6 or day 7 (Fig. 4B, left), which attains normal levels (Lohmann and Bieker, 2008; Soni et al., 2014) in the wild-type EBs.

Although this may result from an intrinsic erythroid deficit, we also tested whether the Nan cells could exert an extrinsic effect on wild-type EB differentiation by mixing equivalent numbers of +/+ and Nan/Nan cells at d0 and then monitoring erythroid status during differentiation via cell surface marker analysis as before. Again, although CD44 levels are not affected, no evidence of Ter119 expression is evident even though the number of +/+ cells included in the culture was no different from before (Fig. 4B, right). We conclude that the dysmorphology of the non-erythroid tissues seen in the Nan embryo may well result from the ectopic expression pattern and resultant secretory proteins seen within the Nan-expressing cell.

We next addressed whether the altered expression patterns of Hamp and Irf7 seen in the fetal liver are also manifested in the Nan/+ adult, and could explain part of the anemia exhibited by the Nan mouse. The adult Nan/+ mouse exhibits low mean corpuscular volume and mean corpuscular hemoglobin, reticulocytosis and splenomegaly (Siatecka et al., 2010b), suggesting defective erythropoiesis possibly due to iron deficiency (Andrews, 2009). Consistent with this, circulating erythropoietin levels are higher in Nan/+ adults than in matched littermates (Fig. 5A). Hamp and Irf7 expression levels are significantly higher in spleen and bone marrow RNA from the Nan/+ mouse than in the wild type (Fig. 5B) (although hepcidin expression is lower in bone marrow than in spleen). To probe the possibility of systemic changes, we monitored serum levels of hepcidin and found that its levels are significantly increased in the circulation in Nan/+ mice compared with wild type (Fig. 5A). Analysis of serum IFNβ levels show an even more dramatic change (Fig. 5A). Serum analyses in the adult reflect alterations in the iron use components we had observed in the Nan/+ fetal erythroid cell (Fig. 3), as there is a slight but significant increase in serum bilirubin and a decrease in serum iron (Fig. 5C). Together, these add up to an iron- and heme-deficient red cell environment that contributes to the overall anemia of the mouse.

Increased hepcidin production can be achieved by IL6-mediated stimulation of the liver, particularly as a result of an inflammatory response (Nemeth et al., 2004a). As a result, we checked for IL6 in the serum, but did not find any detectable levels in either the wild-type or Nan/+ adult (Fig. 5D). This removes inflammation as a causal effector and negates a possible liver contribution to the circulatory hepcidin levels we see in the Nan/+ adult.

Collectively, the developmental data suggest that the anemia in adult Nan/+ mice is exacerbated not only by the intrinsic defects in erythroid expression based on the category I versus category II distinction, but also by the neomorphic genetic output via the new recognition site. This leads to expression of secreted molecules that work directly against establishing a steady-state replenishment of erythroid cells to combat the anemia.

The DNA-recognition properties of the Nan-KLF1 protein variant (Gillinder et al., 2016) suggest a possible mechanism for the effect; i.e. by binding to novel target site(s) within the Hamp and Irf7 genes not normally recognized by the wild-type KLF1 protein. Inspection of the gene loci of Hamp and Irf7 reveal a number of potential Nan-KLF1 target sites (5′CCMNGCCCN) that meet this criterion (Fig. 6A). A subset were tested in vitro by quantitative gel shift assays for binding to wild-type or Nan-KLF1 proteins. As controls, we obtained dissociation constants for known category I (p21 site 3) and category II (dematin and E2F2 site 2) DNA sites binding to wild-type or Nan-KLF1 protein (Fig. 6B). Using this approach, we monitored binding to the predicted new sites in the hepcidin (Hamp) and Irf7 genes and found that binding affinities are substantially higher for Nan-KLF1 compared with wild type (Fig. 6C), supporting the predictions made from the global variant analysis.

As a final test, we checked for in vivo interaction of KLF1 or Nan-KLF1 to these new sites. As we do not have an antibody that can distinguish wild-type KLF1 from Nan-KLF1, we simply monitored whether the adult Nan/+ spleen shows a novel KLF1 ChIP compared with wild type at any of the potential sites of Hamp and Irf7. KLF1 is well expressed within the red pulp of the normal adult spleen (Miller and Bieker, 1993; Southwood et al., 1996). The analyses (Fig. 6D) show that the putative sites are indeed occupied only in the chromatin derived from the Nan/+ mouse. Collectively, the in vitro and in vivo data demonstrate that the variant DNA recognition property of Nan-KLF1 is responsible for altered binding and subsequent ectopic protein expression in the Nan/+ erythroid cell.

Our studies demonstrate that an apparently minor conservative change in a single amino acid (E to D) of a crucial erythroid transcription factor can result in significant phenotypic changes. In Nan-KLF1, the residue in question is absolutely conserved across the whole KLF family in all species examined. This leads to a limitation on its ability to recognize the consensus wild-type DNA element (Siatecka et al., 2010b). However, a remarkable finding from the global analyses was that a novel, atypical site is newly recognized (Gillinder et al., 2016), leading to a significant number of expressed genes that are not normally present in the mammalian red cell. We have shown how this can lead to defects of gene expression within and outside the cell.

Within the wild-type erythroid compartment, neither hepcidin nor IRF7 is expressed. However, unanticipated phenotypic changes arise from two physiologically nonproductive cycles generated by the Nan/+ red cell (Fig. 7). In one case, misexpression of hepcidin and deficient expression of erythroferrone lead to a decrease in iron availability and lower red cell numbers. In the second case, misexpression of IRF7 induces IFNβ expression, which has a repressive effect on erythroid development. The organism responds to the defects by inducing erythropoietin to replenish erythropoiesis, leading to splenomegaly. However, this is futile, as the very cells being expanded are those that are secreting hepcidin and IFNβ, leading to an ill-designed feedback inhibition. This cycle is superimposed upon the intrinsic erythroid deficits, leading to a worsening of the red cell anemia. The increase in hepcidin is both direct (via binding of Nan-EKLF to its atypical site) and indirect (via low erythroferrone expression in the Nan/+ environment), as ablation of EKLF (and thus low erythroferrone) is not sufficient to substantially increase hepcidin levels. Remarkably, these deficits occur even in the presence of a normal wild-type copy of KLF1.

More globally, iron is indispensable for all tissues, being a crucial component of essential enzymes and multiprotein complexes that contain heme and iron-sulfur clusters (Zhang et al., 2014). As a result, heme-containing proteins play crucial roles in all cells (Sawicki et al., 2015); however, heme can also play lineage-specific roles in differentiation (Philip et al., 2015). Proper heme levels are also crucial for normal fetal development, as defects can lead to limb deformities in addition to abnormal erythropoiesis (Keel et al., 2008). Not surprisingly then, increased circulatory hepcidin can have systemic effects; for example, decreasing heme- and nonheme-iron absorption to the duodenum (Cao et al., 2014). In the present case, the levels of hepcidin in the Nan/+ mouse are high enough to exert a physiological effect (Nemeth et al., 2004b).

As a result, we suggest that Nan-EKLF expression contributes to the dysmorphology of Nan embryos seen during early stages of development and during ES cell differentiation, to the fetal liver erythroid misregulation, and to the chronic anemia of the adult (Fig. 7). In this context, it is of interest that a DAVID analysis of the most significantly upregulated gene categories in Nan/+ RNA shows that those in the ‘secreted’ and ‘signal’ groupings are the most numerous (Fig. S3).

Intriguingly, low iron availability via hepcidin expression coupled with expression of IFNβ resembles the anemia of chronic inflammation (Goodnough et al., 2010), where iron restriction sensitizes cells to the inhibitory effects of inflammatory cytokines (Richardson et al., 2013). However, we do not detect any IL6, the key inducer of hepcidin and mediator of inflammation (Hom et al., 2015; Wang and Babitt, 2016).

To our knowledge, finding genetic effects of a tissue-restricted factor that occur outside of its normal expression realm is novel, and thus may be an unrecognized contribution to human disease. In the present case, the most direct connection is with CDA type IV (Iolascon et al., 2012), which is caused by a mutation in human KLF1 at the same amino acid (Arnaud et al., 2010; Jaffray et al., 2013; Singleton et al., 2011). The change is non-conservative (E to K), so although not all characteristics of the Nan mouse may apply, individuals with CDA exhibit altered globin switching and membrane deficits that are highly reminiscent of the present model (discussed by Siatecka et al., 2010b). Of greater interest are the non-erythroid aspects of a subset of these individuals, such as short stature, high serum iron and bilirubin levels, and altered gonads. These suggest that there may very well be secreted factors that are aberrantly/ectopically expressed in these human cells as a direct result of the single amino acid change seen in the human KLF1 CDA mutant.

It may appear surprising that Nan-KLF1 is able to recognize its atypical site within genes that are presumably not easily accessible outside their normal cell of expression. However, recent studies with KLF4, which is highly related to KLF1 (Bieker, 2001), suggest that zinc fingers 2 and 3 play a necessary and dominant role in targeting its recognition site within nucleosomes (Soufi et al., 2015). In the present case, the amino acid change is located within zinc finger 2 and thus could enable Nan-KLF1 to play a role as a pioneer factor at ectopic sites in spite of their lower accessibility in the red cell.

Our present studies, as with the earlier analyses of the Nan-KLF1 protein gene regulation (Siatecka et al., 2010b), re-emphasize that the effects of KLF1 mutation are not strictly due to a dominant-negative outcome. This is clearly different from the cell-intrinsic defects seen with the haploinsufficient or most of the other mutant KLF1 proteins that have been identified in humans (Borg et al., 2011; Helias et al., 2013; Perkins et al., 2016; Singleton et al., 2012; Tallack and Perkins, 2013; Waye and Eng, 2015). What is seen in the present case is a new variant with novel properties that dramatically changes Nan/+ red cell identity.

Establishment of the Nan mouse strain has been previously described (Heruth et al., 2010; Lyon, 1983; Siatecka et al., 2010b). All protocols have been reviewed and approved by the IACUC at Mount Sinai. RNA was isolated from E13.5 fetal livers or adult spleens using Trizol. Total RNA was purified with a Qiagen RNeasy kit. qRT-PCR was performed using published procedures (Siatecka et al., 2010b). Global RNA-seq was performed by the Mount Sinai Facility or by the JAX facility and analyzed as described (Gillinder et al., 2016). The DAVID analysis tool v6.7 (https://david-d.ncifcrf.gov/) was used as described (Huang et al., 2009).

K1-ER or K1-Nan-ER cells (originally described by Coghill et al., 2001, recently authenticated and reanalyzed by Gillinder et al., 2016) were grown and treated with doxycycline as described (Coghill et al., 2001; Hodge et al., 2006). Analysis is limited to 1 h post-treatment to avoid loss of cell number following KLF1 induction of CDK inhibitors and the ensuing cell cycle arrest (Gnanapragasam et al., 2016; Pilon et al., 2008; Siatecka et al., 2010a; Tallack et al., 2009, 2007). As a result, only immediate-early effects can be detected. Nascent RNA expression was monitored after a concurrent 1 h treatment of cells with 4-thiouridine, as described previously (Gillinder et al., 2016).

Mouse ESCs derived from wild-type and Nan/Nan blastocysts (littermates) were established as described previously with minor modifications (Meissner et al., 2009). These were maintained and differentiated to embryoid bodies as per established protocols (Choi et al., 2005; Kennedy and Keller, 2003) described previously (Manwani et al., 2007). Ter119-PE and CD44-FITC antibodies were from eBiosciences.

Erythroblasts were derived from wild-type and Nan/+ E12.5 fetal livers and depleted of adherent cells as previously described (England et al., 2011; Gnanapragasam et al., 2016). Briefly, cells were harvested, and plated on gelatin dishes for 2 days. This was followed by a 1-day culture in erythroid growth media prior to harvest for RNA analysis.

ChIP analysis of EKLF binding was performed using a modification of previous procedures (Lohmann and Bieker, 2008; Siatecka et al., 2015). Briefly, cells from fetal livers or spleens (7-10×10⁶ per immunoprecipitation) were crosslinked with 1.0% formaldehyde. Chromatin was sonicated, immunoprecipitated with antibodies against mouse IgG (557273; BD Pharmingen) or EKLF (7B2), and purified on magnetic protein G Dynabeads (10004D; Life Technologies). Eluted and precipitated genomic DNA was amplified and quantified by PCR.

Serum from adult mice was analyzed using ELISA kits for erythropoietin (R&D Systems), IL6 (Thermo Fisher), hepcidin (Intrinsic Lifesciences) or IFNβ (PBL Assay Science), following the manufacturers' instructions. Bilirubin and iron were measured on a Beckman Coulter AU680 Chemistry Analyzer; only samples with low hemolysis [i.e. hemoglobin value lower than 100 (Beckman-Coulter)] were used for these analyses.

To generate probes, one strand of DNA oligonucleotide was end-labeled with polynucleotide kinase and γ-P³² ATP, and then annealed to the unlabeled complimentary strand. For some experiments, only labeled DNA was used and concentrations were determined with Quant-it PicoGreen from Invitrogen. For other experiments, the labeled DNA was used a tracer with known quantities of unlabeled DNA. Similar results were obtained with these methods.

We thank S. Soni for providing purified recombinant protein, and N. Gnanapragasam and C. Deligianni for primary cell protocol advice. We thank T. Ganz, S. Rivella, N. Andrews, D. Higgs, H. Drakesmith, C. Roy and A. Goldfarb for discussion. We thank S. Ghaffari, J. Little and C. Deligianni for comments on the manuscript.