Xenograft models for normal and malignant stem cells

Since immunodeficient mice were first used in biomedical research, there has been a continual effort to improve their utility and expand their applicability to more areas of research. Over the last decade, with the advent of widespread transgenic approaches and the understanding by researchers of the potential for these models in moving our understanding of biology forward, a number of genetically engineered animals have become available. For a comprehensive overview of the history and use of xenografting, the reader is directed to a number of recent reviews.^1-5  In this review, we will focus on the models and approaches of most relevance to researchers studying normal and malignant hematopoiesis and discuss future avenues to take that will address current limitations.

Of the different strains available to the research community, the most popular strains are the NOD/SCID-IL2RG^−/− (NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ: NSG and NOD.Cg-Prkdc^scid Il2rg^tm1Sug/Jic: NOG) mice.^6,7  Variations of this strain are becoming available through breeding strategies and genomic engineering techniques. It is clear that different strains will be optimal for different types of studies. For example, the humanized mouse, in which components of human bone marrow (BM), liver, and thymus (BLT) are grafted in immunodeficient mice, is most popular with investigators studying infectious disease due to the faithful development of mature, properly educated human T cells and a more complete, functional human immune system.^5,8  The BLT approach is powerful but requires specialized surgical expertise and additional time, as well as tissue from human fetuses, raising feasibility issues. For studying hematologic malignancies, many researchers use the NSG mouse expressing cytokines that support human myelopoiesis (NSG-hSCF, hGM-CSF, hIL3, all driven from the cytomegalovirus promoter; NSGS), for their superiority in promoting robust engraftment of a wide range of patient samples.^9,10  Designer mice are now becoming available through the efforts of a number of laboratories around the world, with Drs Mamoru Ito (NOG), Leonard Shultz (NSG), and Richard Flavell and Markus Manz (Rag2^−/−IL2RG^−/−) leading the way on improvements in these mouse strains. In a relatively short period of time, numerous genetically modified substrains will be available that seek to address the cross-species incompatibility of growth factors, receptors, adhesion molecules, and histocompatibility antigens. Engraftment levels in these 3 mouse strains are relatively equal now that the Rag2^−/−IL2RG^−/− mice express the human SIRPα molecule^11,12  (Figure 1).

For studies focused on human HSC self-renewal in vivo, the NSG/NOG mice are preferred due to their widespread availability and excellent engraftment rate. Although high levels of sustained engraftment can be obtained on transplant of human CD34⁺ cells into primary mice, secondary transplants are rarely performed, and for those reported, most show low levels of engrafted human cells. In 1 study, the authors demonstrated that the primitive stem/progenitor cells remain in active cycle for up to 8 weeks after engraftment, possibly implicating a lack of quiescence as the driving factor in the limited self-renewal.¹³  This could also explain the unusually high frequency of human CD34⁺ cells detected in the BM of xenografted mice (10-20% of the human graft) compared with a normal human BM containing 1% to 2% CD34⁺ cells at steady state.^14,15  It is possible the signals that normally induce transplanted HSCs to enter quiescence are absent and/or non-cross-reactive in the immunodeficient mouse. To address this, Flavell’s group used Rag2^−/−IL2RG^−/− mice expressing human thrombopoietin (TPO) in place of murine. However, only a slight enhancement in serial transplantation was observed.¹⁶  It is likely that multiple signals are aberrant in the murine microenvironment, and a complex array of genetic changes will be needed to mimic the human BM niche. It is also possible that the presence of facilitating cells during transplant is critical for proper function of the HSCs. We recently described a simple procedure using total BM or cord blood samples in xenotransplants. The addition of the OKT3 monoclonal antibody to a cell mixture readily addressed the graft-versus-host disease that ensues on transplanting samples containing mature T cells into immunodeficient mice. This process also more closely mimics the stem cell transplant procedure as it is performed in humans.¹⁷  Limiting dilution analysis showed the stem cell frequency was as good or better than approaches using purified human CD34⁺ cells. Secondary transplant was very robust in a small pilot experiment, ranging between 10% and 50% multilineage human engraftment in the BM at 16 weeks, which is dramatically better than found in most published studies. More work is needed to determine whether this approach will contribute to solving the problems associated with analysis of human HSC self-renewal in xenotransplants.

Unfortunately, there is currently no standard in the field regarding which mice, approaches, and time points are best for measuring human HSC activity in vivo. A wide range of engraftment levels and durations have been used as experimental end points, and few researchers use secondary transplant as a readout due to poor engraftment. It is important that researchers adopt an accepted standard for measuring human HSC and hematopoietic stem and progenitor cell (HSPC) function as has been done for murine HSPC studies. This will allow a straightforward comparison of different approaches and models using a standard methodology.

Xenograft models have been used to determine immunophenotypes of human HSCs. In contrast to murine HSCs that are enriched in the CD34-negative fraction, CD34 has long been used as a positive marker of human HSCs not only in xenograft repopulation assays but also in clinical HSC transplantation.^18,19  Additional HSC markers have been identified, and it is now possible to isolate human HSCs at the single cell level.^20-28  Some primitive human HSCs may reside in a CD34-negative fraction similar to murine HSCs, but the frequency of CD34-negative SCID-repopulating cells (SRCs) is very low even with the use of additional HSC markers^29-31  (Table 1). Interestingly, current evidence suggests that human HSCs have little correspondence with murine HSCs in terms of surface marker expression. Stem cell antigen-1 (Sca-1) is used extensively to enrich for murine HSCs, but it does not have a human homolog.³²  Human HSCs express FLT3, whereas mouse HSCs do not.³³  CD38 is not expressed on human HSCs but is expressed on murine HSCs.^34-36  CD150, a member of the SLAM family receptors, has been widely used to isolate murine HSCs.³⁷  However, human HSCs cannot be purified based only on SLAM markers.³⁸  In addition, human HSCs expressing high levels of KIT (KIT-hi) contain more potent repopulating activity compared with HSCs expressing intermediate levels (KIT-int),³⁹  whereas Kit-int murine HSCs are the best at repopulation and Kit-hi defines murine HSCs initiating the process of differentiation.⁴⁰  These discrepancies likely indicate species differences but may partially be explained by different experimental conditions (syngeneic transplantation vs xenotransplantation). It should also be noted that many previous xenograft studies assessed engraftment of human HSCs at a relatively early time point (6-12 weeks after transplant; Table 1). Given that long-term HSCs show a delayed engraftment pattern,⁴¹  a longer period (or serial transplant) may be needed in future studies.

Studies for ex vivo HSC expansion combined with xenograft assays have revealed mechanisms governing self-renewal of human HSCs. Many studies have suggested that Notch signaling plays important roles in the regulation of human HSCs. Activation of Notch receptors by ligands (JAG1, DLL1, or DLL4) has been shown to promote ex vivo expansion of SRCs.^42-45  In line with this, enforced expressions of HES1 (a Notch target gene) and NOV (an extracellular activator of Notch signaling) in human CB cells confer enhanced in vivo reconstitution ability in NOD/SCID mice.^46,47  Notch signaling may also regulate the quiescent state of CD34⁻ HSCs together with transforming growth factor β to repress the Wnt pathway.³⁰  The effect of Notch on human HSCs appears to be dosage dependent, because low doses of DLL1 expand CD34⁺ cells, whereas higher doses induce apoptosis.⁴⁸  Early studies also identified WNT and hedgehog pathways to promote HSC expansion,^49-51  and several inhibitory pathways, including transforming growth factor β, tumor necrosis factor α, and chemokines, to suppress HSC proliferation.⁵²  Recent studies have identified new pathways involved in human HSC regulation. Angptl5, a member of angiopoietin-like proteins, promotes 20-fold expansion of SRCs when used in serum-free culture media containing stem cell factor (SCF), TPO, fibroblast growth factor 1, and insulin-like growth factor binding protein 2 (IGFBP2).⁵³  A neurite outgrowth factor Pleiotrophin also increases SRC counts in culture, and Notch and phosphatidylinositol 3-kinase pathways mediate the response to Pleiotrophin.⁵⁴  A zebrafish screen and subsequent reports have shown that prostaglandin E2 (PGE2) enhances emergence and repopulating ability of HSCs through activation of WNT signaling and upregulation of CXCR4 and Survivin.^55-58  Conversely, inhibition of endogenous PGE2 by nonsteroidal anti-inflammatory drug treatment promotes HSC egress from the BM to the circulating blood not only in mice but also in healthy human volunteers.⁵⁹  More recently, an unbiased drug screening identified an aryl hydrocarbon receptor (AhR) antagonist SR1 that promotes ex vivo expansion of human CD34⁺ cells that retain the ability to engraft NSG mice.⁶⁰  AhR requires ARNT hypoxia inducible factor (HIF1B) to regulate gene expression, and interestingly, ARNT is also required by HIF1A to enhance gene expression in response to hypoxia,⁶¹  suggesting the involvement of the HIF pathway in the regulation of human HSCs. Indeed, a study showed that knockdown of HIF2A, and to a lesser extent HIF1A, impedes the long-term repopulating ability of human cord blood (CB) CD34⁺ cells through increased reactive oxygen species (ROS) production and endoplasmic reticulum stress.⁶²  In addition, treatment of recipient mice with the ROS scavenger N-acetyl-l-cysteine significantly improves engraftment of human HSCs in NOD/SCID and NSG mice.⁶³  Another recent screening confirmed that multiple compounds suppressing the AhR pathway can promote HSC expansion ex vivo, although it appears that SR1 mainly expands short-term HSCs with limited regenerative potential. Interestingly, the study also identified a compound UM171 that efficiently promotes a robust ex vivo expansion of long-term human HSCs through AhR-independent mechanisms.⁶⁴  As for the inhibitory signaling for human HSCs, short hairpin RNA library screening identified mitogen-activated protein kinase 14 (p38α) as a negative HSC regulator. Pharmacologic inhibition of p38 dramatically enhances the multilineage repopulation of human CB cells in NSG mice presumably by reducing levels of ROS.⁶⁵ 

Several transcription factors possess the ability to promote HSC expansion ex vivo. HOX proteins, HOXB4 and HOXC4, moderately (2- to 6-fold) improve the level of in vivo engraftment of human CD34⁺ cells.^66,67  Similarly, enforced HLF expression confers increased repopulation potential to human HSCs.⁴⁶  RUNX1 has isoform-specific activity to regulate self-renewal and differentiation of HSCs. Forced expression of RUNX1a, a short isoform of RUNX1, increases SRC activity of human HSCs and facilitates emergence of definitive human HSCs from human embryonic stem cells.^68,69  In contrast, long isoforms of RUNX1 (RUNX1b and RUNX1c) inhibit repopulation of human/mouse HSCs in recipient mice by promoting myeloid differentiation^68,70  and/or increasing quiescence of HSCs.⁷¹  GATA2 is highly expressed in quiescent HSC fractions, and enforcing GATA2 expression increases quiescence of human CB cells.⁷²  DNA damage response is essential for maintaining HSC function, and p53 plays a key role in this process. In contrast to mouse HSCs with decreased sensitivity to cytotoxic agents, human HSCs exhibit enhanced p53-dependent apoptosis after irradiation. Inactivation of p53 reduces apoptosis and partially rescues the repopulating ability of the irradiated HSCs in primary recipient mice. However, p53-depleted HSCs show diminished SRC activity in secondary recipients probably due to accumulated DNA damage, suggesting that intact p53 is important to maintain human HSCs in the serial transplantation assay.⁷³  In addition to these transcription factors, epigenetic modifiers also regulate human HSC function. The polycomb complex protein BMI1 was shown to promote the repopulation potential of human HSCs.^74,75  Furthermore, addition of epigenetic drugs can increase the self-renewal of human HSCs ex vivo. Human CB cells treated with decitabine (DNA methylation inhibitor) and trichostatin A (histone deacetylase inhibitor) in the presence of hematopoietic cytokines resulted in significant expansion of CD34⁺CD90⁺ cells, including transplantable HSCs.^76,77  Another histone deacetylase inhibitor, valproic acid, also stimulated proliferation and self-renewal of human HSCs accompanied by p21^cip1/waf1 downregulation, WNT pathway activation, and HOXB4 upregulation.⁷⁸  Finally, increasing evidence suggests the importance of miRNAs for human HSC regulation.^79,80  Among the miRNAs highly expressed in human HSCs, miR-125 was shown to regulate HSC function positively,⁸¹  whereas miR-126 was shown to be a negative regulator of HSC proliferation.⁸² 

Although these studies have revealed key players to maintain/expand human HSCs (Figure 2), it should be noted that signals needed for HSC expansion ex vivo may be different from those to maintain HSCs in human bodies. Indeed, a recent study showed that Notch signaling might be dispensable for self-renewal of human HSCs in vivo. Inhibition of Notch activity using a dominant-negative inhibitor of Mastermind-like (dnMAML) did not change SRCs when the dnMAML-transduced HSCs were directly transplanted to NSG mice, whereas dnMAML-transduced HSCs were not maintained ex vivo.⁸³  Developing more humanized mice will be necessary to decipher actual mechanisms regulating human HSCs in vivo.

Xenotransplant has proven highly successful for acute leukemia and to a lesser degree for the nonacute hematologic malignancies such as myelodysplastic syndrome (MDS), myeloproliferative neoplasms, and chronic lymphocytic leukemia (CLL).⁴  This indicates that the xenograft environment is suitable for the self-renewal of engrafting leukemia cells (SCID-leukemia-initiating cells [SL-IC]⁸⁴ ). SL-IC is a functional description of those leukemia cells that possess engrafting potential, whereas LSC is a conceptual description of the self-renewing leukemia cell that can propagate the tumor. The connection between SL-IC and LSC is complex and beyond the scope of this review, and we will use the term LSCs throughout most of this review. Even in the context of the transgenic cytokine mice (eg, NSGS mice), SL-ICs do not exhaust and are readily transferred to secondary and tertiary mice.⁹  In contrast, HSCs are mobilized from the BM in NSGS mice, the human grafts are less durable than those in nontransgenic NSG mice, and HSCs do not efficiently transplant to secondary mice⁸⁵  (and our unpublished data, January 2012). These data demonstrate a differential requirement for microenvironmental cues between HSCs and SL-ICs (and possibly LSCs). This may be due to the block in differentiation associated with leukemia, allowing unlimited self-renewal divisions in the absence of the normal niche signals needed by HSCs. These differences may also reflect discrepancies in the intrinsic nature of leukemia vs normal stem cells such that LSCs do not exhaust their self-renewal ability when mobilized into cycle, whereas HSC do (Figure 3).

Although transfer of leukemia samples to immune deficient hosts has greatly improved, a substantial number of samples still fail to engraft even the most optimized host mice.^9,86  Together, these data indicate that significant interspecies barriers to engraftment and expansion of certain LSCs remain, particularly for less aggressive hematologic malignancies. Specific subtypes of leukemia are significantly more problematic to engraft, for example, samples from patients with t(8;21). It may be that some subtypes of AML have low progenitor cell frequency, or, some samples may be particularly sensitive to the lack of a factor or cell type from the mouse BM. In the case of t(8;21), we and others have shown that signaling through TPO/mpl is particularly important for the growth of cells expressing AML1-ETO.^87,88  Thus, hTPO knockin mice may improve engraftment of t(8;21) AML, as appears to be the case according to a presentation at the 2014 American Society of Hematology annual meeting.^16,89,90  It should also be noted that engraftment itself does not indicate the detection of LSCs because residual normal HSCs in patients can repopulate mice. It is therefore important to confirm genetic and phenotypic abnormalities of the engrafted human cells in studies for human LSC. Interestingly, a recent xenograft study demonstrated the existence of preleukemic HSCs with a DNMT3A mutation. The preleukemic HSCs did not show leukemic engraftment characterized by a dominant myeloid graft but showed a multilineage repopulation advantage over nonmutated HSCs.⁹¹  Xenograft models combined with genetic analyses will enable the identification of preleukemic and leukemic stem cells in patients.

There is a growing list of studies that examine the effects of individual leukemia-associated oncogenes on human hematopoietic stem and progenitor cells (HSPCs), in effect building human leukemia using primary human HSPCs (Table 2). Although most of these studies report only partial phenotypes, consistent with a model of stepwise progression to transformation, there are now several examples of ex vivo generation of human leukemia from primary human cells as assayed by xenografts using defined genetic elements.^92-95  These successes are limited to a select few recurrent translocations, likely indicating that most oncogenes require cooperating genetic mutations to elicit full-blown transformation. These studies also reiterate the finding that human cells are more refractory to transformation than are murine cells, as shown experimentally.⁹⁶  It is also likely that BM microenvironmental cues are critical for preleukemic HSPC initiation and leukemia progression, similar to the situation for normal human HSCs, in contrast to a fully transformed LSCs.

Whether xenograft models can be used to predict response to leukemia therapy is an open question. Early studies showed a correlation between engraftment of AML samples in NOD/SCID mice and poor clinical outcome, and this finding has been replicated in NSG mice.^97,98  However, whether the phenotypically defined LSC frequency (simplistically characterized as CD34⁺CD38⁻) in AML is prognostic and whether this correlates with xenograft potential is controversial.^98-100  Interestingly, it was shown that poor clinical outcome correlated with the degree to which the LSCs matched normal HSC gene expression.¹⁰¹  Whether this means that the transformative event(s) occurs initially in true HSCs or rather it results from transformed progenitors acquiring stem cell activity and gene expression remains unclear. More work remains to be done before strong conclusions can be drawn, but it may be possible to use data derived from xenografted AML samples to identify biomarkers that may predict clinical outcome.

Xenograft models also provide powerful tools to perform preclinical testing of candidate drugs for hematologic malignancies. However, we should keep it in mind that the dominant clone present at relapse in a patient is not necessarily the clone emerging on xenotransplantation.¹⁰²  This may be due to different selective pressures in xenotransplant relative to what the AML clones encounter in patients undergoing induction and consolidation therapy. Xenomodels combined with therapeutic treatments may better mimic the situation of patients. We recently showed that NSGS mice can be used to mimic induction therapy during xenograft.¹⁰³  One drawback to this approach is the frailty of SCID-based mice, making it impossible to simultaneously use conditioning to ensure engraftment of AML samples and subsequently treating with induction chemotherapy to force outgrowth of a relapse clone. We have crossed the NSGS mice with the NRG mice and found these mice (NRGS) are robust enough to permit conditioning, induction, intensified induction, and consolidation without death of the host (unpublished observations). This is consistent with the demonstration of radioresistance of the NRG strain.¹⁰⁴  We expect the NRGS host combined with drug treatments will allow an examination of clonal evolution in leukemia under similar pressure to that in patients receiving therapies. These mice will be commercially available in 2015.

Xenograft studies have identified many useful markers to enrich LSCs, but it has become increasingly evident that LSC phenotype varies between individual patients. In addition, technical variation of xenograft assays also affects the results, with a trend that more immunodeficient strains allow the engraftment of variable LSC populations^{10,84,101,105-126}  (Table 3). Early pioneering studies using NOD/SCID mice showed that AML is a hierarchical disease that is initiated by a rare subset of immature CD34⁺CD38⁻ LSCs.⁸⁴  However, LSCs were found in other compartments including the CD34⁺CD38⁺ fraction in recent studies using NSG mice or NOD/SCID mice treated with anti-CD122 to inhibit natural killer cell function.^101,109,114  Furthermore, AMLs with NPM1 mutations were shown to contain LSCs in the CD34⁻ fraction,¹¹⁰  indicating the subtype-specific features of AML LSCs. Although human LSCs and HSCs share similar immunophenotypes, some markers such as CD123 (IL3-Rα), CD96, CD44, CD47, CD32, CD25, CLL-1, IL1RAP, and TIM3 appear to be expressed higher in LSCs than normal HSCs, providing potential therapeutic targets to selectively eradicate LSCs.^{105,107,108,111,112,127-132}  As another strategy to target LSCs in AML, a recent study showed that BCL2 inhibition reduced oxidative phosphorylation and selectively eradicated quiescent human LSCs.¹³³  For chronic myeloid leukemia (CML), DPPIV (CD26) was shown to be specifically expressed in CML LSCs that can be a therapeutic target.¹¹⁶ 

Studies using purified MDS patient samples have shown that MDS is a clonal disease of the HSCs.¹³⁴  MDS has been an extremely challenging disease to model in the immunodeficient mouse. The difficulty encountered in xenografting MDS samples may relate to special requirements of MDS cells for specific signaling events in the niche.¹³⁵  Recently, transplantation of matched mesenchymal stroma cells were shown to facilitate xenotransplant of MDS stem cells.¹⁰  This study showed that the MDS stem cell may be dependent on several differentially expressed genes from MDS stroma cells, some of which are induced by the MDS cells, including LIF, and may represent new targets for therapy. Additional xenograft studies have confirmed that only highly purified stem cells from multiple del(5q) MDS patients were able to engraft mice, showing definitively that MDS is driven by distinct identifiable malignant stem cells.¹²⁶  These MDS stem cells generated myeloid progenitor cells in xenografts but had no lymphoid potential. Another study engrafting del(5q) MDS stem cells also found a myeloid-restricted output with a reduction of granulocyte-monocyte progenitors, similar to what is observed in low risk MDS patients.¹²⁵  It is apparent the MDS stem cell has a skewed lineage potential in xenografts, similar to what is reported in patients.¹³⁶  These MDS xenograft models may prove particularly useful for testing new therapies for this difficult disease, as shown recently in a xenograft model of MDS using a lenalidomide-responsive MDS cell line.¹³⁷ 

Hierarchical organization is less strict in acute lymphoid leukemia (ALL). Some early studies showed the exclusive engraftment of primitive ALL LSCs (CD34⁺CD19⁻/CD10⁻ for B-ALL and CD34⁺CD4⁻/CD7⁻ for T-ALL) in NOD/SCID mice,^117,118,122  but many later studies demonstrated LSC potential of cells with phenotypic characteristics of differentiated progenitors (CD19⁺ cells for B-ALL and CD7⁺ cells for T-ALL).^119-121,123  Interestingly, a recent study suggests that CLL may be derived from primitive HSPCs. Purified HSCs from CLL patients could initiate xenografts that gave clonal expansion of B cells, whereas other isolated fractions did not have this ability.¹²⁴  Interestingly, leukemic clones derived from the same patient but expanded in different mice exhibited distinct VDJ recombination events, clearly suggesting a primitive preleukemic stem cell is driving a disease that continues to evolve in the xenografted mice. As immunodeficient models and techniques improve, more such studies will be possible using primary patient samples from a variety of hematologic malignancies.

Advances in the available mouse strains and in the techniques used to humanize the murine immune system and to model human hematologic malignancies in vivo have been dramatic over the last few years. Numerous important findings have been made with regard to the nature and function of human HSCs and LSCs. Nevertheless, some major hurdles remain to be overcome to broaden the use of this approach to studying human LT-HSCs and to model hematologic malignancies that are not currently amenable to xenograft.

The arm of the immune system that has proven most difficult to model in immunodeficient mice is the megakaryocytic and erythrocytic lineages. Numerous reasons have been proposed for the problems associated with development of these lineages, and it is likely that the problems are multifaceted and will not be solved with a single genetic fix.⁴  Phagocytosis of red blood cells and platelets by murine macrophages has been shown to play a role in the lack of these cell types in the periphery of humanized mice.^138,139  However, humanized mice also have a dramatic defect in the generation of the megakaryocytic/erythroid progenitor (MEP), and this deficiency is not corrected in Rag2^−/−IL2RG^−/− mice expressing human thrombopoietin.¹⁶  The lack of human megakaryocytes in the BM of humanized mice could play a role in the quiescence and self-renewal problems of the human HSCs, given the recent findings that mature megakaryocytes are essential niche cells for HSCs.^140,141  One possible contributing factor to both the HSC and MEP defects is the increased ROS levels found in NS and NSG mice.⁶³  ROS levels have been shown to be particularly low in MEP, and low levels of ROS in the common myeloid progenitor correlate with gene expression signatures that favor the MEP fate.¹⁴²  It is possible that correction of the high ROS levels in these immunodeficient mice may allow for development of a more complete immune system and simultaneously promote the quiescence and self-renewal of the human HSC through multiple mechanisms.

The development of genetically modified immunodeficient mice will continue, likely at an increased pace. These modifications will include not only substitutions of growth factors, receptors, and adhesion molecules but also major histocompatibility molecules to promote proper education and development of cells dependent on these molecules for function. In the near future, it is likely that an improved xenograft mouse will be available for most blood researchers, including specialists focused on human HSCs and LSCs. The in vivo approaches that are critical to a better understanding of stem cell function will be even more powerful than the excellent models that currently exist. In time, our scientific understanding of human stem cell function will rival that of the murine stem cell.

Contribution: S.G., M.W., and J.C.M. wrote the paper.

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

Correspondence: James C. Mulloy, Division of Experimental Hematology and Cancer Biology, Children's Hospital Medical Center, University of Cincinnati College of Medicine, Cincinnati, OH 45226; e-mail: james.mulloy@cchmc.org.