Genetic lineage tracing defines distinct neurogenic and gliogenic stages of ventral telencephalic radial glial development
© Anthony and Heintz. 2008
Received: 08 July 2008
Accepted: 05 November 2008
Published: 05 November 2008
Radial glia comprise a molecularly defined neural progenitor population but their role in neurogenesis has remained contested due to the lack of a single universally accepted genetic tool for tracing their progeny and the inability to distinguish functionally distinct developmental stages.
By direct comparisons of Cre/loxP lineage tracing results obtained using three different radial glial promoters (Blbp, Glast, and hGFAP), we show that most neurons in the brain are derived from radial glia. Further, we show that hGFAP promoter induction occurs in ventral telencephalic radial glia only after they have largely completed neurogenesis.
These data establish the major neurogenic role of radial glia in the developing central nervous system and genetically distinguish an early neurogenic Blbp + Glast + hGFAP - stage from a later gliogenic Blbp + Glast + hGFAP + stage in the ventral telencephalon.
- BAC: :
bacterial artificial chromosome;
- BLBP: :
brain lipid binding protein;
- CNS: :
central nervous system;
- E: :
- eGFP: :
enhanced green fluorescent protein;
- GLAST: :
glial high affinity glutamate transporter;
- hGFAP: :
human glial fibrillary acidic protein;
- ISH: :
in situ hybridization;
- NE: :
neuroepithelial stem cell;
- PBS: :
- PH3: :
phosphorylated histone H3;
- RG: :
- VZ: :
Radial glia (RG) comprise a molecularly defined cellular population, playing critical roles during central nervous system (CNS) development as both neural progenitors and as a scaffolding for migrating neurons [1, 2]. Despite their established role as the major neuronal precursors in the cerebral cortex [3–8], two fundamental aspects of RG development and function remain contested. The first regards the question of whether, similar to the situation in cortex, RG function as the principal neuronal progenitors in all CNS regions. This question has been addressed with Cre/loxP lineage tracing using two different RG-specific promoters, the mouse brain lipid binding protein (Blbp) promoter and the human glial fibrillary acidic protein (hGFAP) promoter. However, sharply different patterns of recombination were driven by each promoter, prompting two distinct models for the RG role in neurogenesis. As significant neuronal recombination was detected in only a subset of brain regions when the hGFAP promoter was used to drive Cre [5, 9], it was proposed that RG differ regionally in terms of their neurogenic potential and generate neurons only in certain brain regions [5, 10, 11]. In contrast, the extensive recombination driven in all brain regions by the Blbp promoter supported an alternative model in which essentially all RG populations are neurogenic and generate most neurons in the CNS . Each model has different implications not only for RG function but also the larger issue of the mechanisms used to generate neurons during development. An accurate picture of lineal relationships within the developing brain is, therefore, essential but the controversy surrounding the RG contribution to neurogenesis has remained unresolved.
A second related issue of contention concerns the question of when cells with the molecular properties of RG first appear in the CNS. Resolution of this question is necessary as elucidation of the mechanisms responsible for driving RG differentiation requires analysis of the relevant pathways when they first become active, not at later stages when the end products of these programs become morphologically evident. Two different estimates for the time point when induction of RG differentiation occurs in the murine forebrain have been proposed based on Cre/loxP lineage tracing results (used to indicate onset of transcription). Whereas the mouse Blbp promoter drove recombination from embryonic day (E)10 , the hGFAP promoter is not active until after E12 . These contrasting results have been alternatively proposed to reflect either precocious or delayed activity of the Blbp or hGFAP promoters, respectively [8, 10, 11], but neither direct evidence supporting either model nor data obtained using an independent RG promoter has yet been reported.
A third deficiency in the current understanding of RG development is the inability to distinguish RG at distinct stages of maturation. Multiple lines of evidence indicate that RG populations progress stepwise through distinct developmental stages during which particular classes of neurons or glia are generated, and that progression to successively later stages is accompanied by restricted potential to produce earlier cell types. For example, cortical RG stop generating neurons at late stages of embryogenesis and transform into astrocytes postnatally . Determination of the precise time points when such transitions occur using in vivo genetic lineage tracing is a prerequisite to the identification of mechanisms regulating shifts in neurogenic potential; to date, few such studies have been conducted focusing on RG.
We report here that: similar to what was observed using the Blbp promoter for lineage tracing, RG expressing the glial high affinity glutamate transporter (Glast, a universally accepted RG marker [10–12]) are molecularly detectable in the forebrain by E10 and generate most neurons in the brain; hGFAP promoter activity is poorly correlated with RG differentiation in multiple brain regions, failing to drive recombination until days after RG first appear; and in ventral telencephalon, hGFAP activity commences after RG are predominantly gliogenic, genetically distinguishing an early neurogenic Blbp + Glast + hGFAP - stage from a later gliogenic Blbp + Glast + hGFAP + stage in this region.
Endogenous GLAST expression and recombination in Glast::Cre;R26R mice are detected in the forebrain from E10.5
Due to the discrepant results previously obtained using the hGFAP and Blbp promoters, we performed Cre/loxP lineage tracing using the Glast promoter to drive Cre expression. Like Blbp, Glast expression is restricted to RG and astrocytes and molecularly distinguishes them from developmentally earlier neuroepithelial stem cells [5, 12, 13]. To generate mice expressing Cre under the regulation of the Glast promoter, we inserted a Cre-polyA cassette into the second exon of the Glast gene in a bacterial artificial chromosome (BAC) and generated BAC transgenic mice. Five separate Glast::Cre founder lines were obtained, all of which yielded similar results when used for lineage tracing; representative results from one line are presented.
To determine when recombination occurs and how well correlated it is with endogenous GLAST, we crossed Glast::Cre mice to the Rosa26LacZ and Z/EG  Cre reporter lines, which express β-galactosidase (R26R) or eGFP (Z/EG) after Cre-mediated recombination. Histochemical staining of Glast::Cre;R26R embryos demonstrated a tight concordance between the onset of recombination and onset of endogenous GLAST expression, with β-galactosidase activity and GLAST protein both becoming detectable in the forebrain at E10.5 (Figure 2a,b,g–j). Importantly, this spatiotemporal pattern of recombination also matches what was observed in Blbp::Cre;R26R embryos . The fact that both endogenous expression and recombination driven by the promoters of two different RG genes are so similar strongly suggests that the NE→RG transition occurs between E9.5 and E10.5 in the forebrain. It should also be noted that the timing of recombination in Glast::Cre;R26R and Blbp::Cre;R26R embryos sharply contrasts with that driven by the neuroepithelial cell marker Nestin, as recombination in Nestin::Cre;R26R embryos is readily detectable throughout the CNS by E8.5 . Thus, the data together provide both molecular and genetic evidence for the existence of a developmental transition that occurs by E10.5 in the forebrain, consisting of an earlier Nestin + Blbp - Glast - NE stage and a later Nestin + Blbp + Glast + RG stage. Analysis of recombination in Glast::Cre;R26R embryos at later time points showed a similarly tight concordance between GLAST expression and recombination (Figure 2k–m), providing further support for the specificity of the Glast::Cre transgene and for the validity of the lineage tracing results.
Glast + RG generate neurons throughout the brain
hGFAP promoter activity is poorly correlated with RG differentiation in multiple brain regions
Why are the hGFAP lineage tracing results different from those using the Blbp and Glast promoters if all three promoters are cell-specifically expressed in RG? As neurogenesis precedes gliogenesis during development, one possibility is that the hGFAP promoter does not become active in some brain regions until after RG neurogenesis is complete. This idea is supported by two observations: eGFP expression in hGFAP::eGFP mice is undetectable in the GE until E14.5 ; whereas most Blbp + RG isolated from the E11.5 GE generated neurons in clonal cultures, most clones generated from cells isolated from E14.5 GE lacked neurons . This latter result indicated that GE RG neurogenesis may be largely completed by E14.5.
Analysis of the GE at E14.5 showed that some recombination had occurred medially, but the vast majority of VZ cells remained unrecombined (Figure 4m,n) and appreciable recombination could not be observed here until E15.5 (Figure 4o). Taken together, these results establish that hGFAP promoter activity is poorly correlated with RG differentiation in multiple brain regions, requiring several days after endogenous RG markers are first induced before becoming active. Direct examination of hGFAP promoter activity further illustrated this point: even though some expression can be detected in the E14.5 GE, significantly higher expression was observed in the cortex (Figure 4p,q). Moreover, weak ventral expression persisted as late as E16.5 (Figure 4r). This is the opposite of the endogenous RG markers BLBP and GLAST, which are more highly expressed ventrally at E14.5 and roughly equivalent at E16.5  and further demonstrates that the hGFAP promoter does not accurately reflect normal characteristics of RG development in mice. Lastly, we note that previous work showed that all GE hGFAP + RG are also GLAST+  and that at E14.5 (when GE hGFAP-mediated recombination first occurs), all BLBP+ RG are also GLAST+ . Thus, hGFAP does not define a unique BLBP-GLAST- population of radial glia.
In this study we have clarified and defined two major aspects of RG development and function. First, we have shown that cells with the molecular features of RG are present in the forebrain from E10.5, establishing that the NE→RG transition commences in this region between E9.5 and E10.5. Second, we have solidified the concept that RG are the primary neuronal progenitors in the developing brain by showing that: most neurons in the brain are derived from Glast + RG; and in multiple brain regions, the heterologous hGFAP promoter is not active until several days after morphologically and molecularly defined RG first appear. Thus, the limited neuronal recombination driven by the hGFAP promoter is not indicative of regionally restricted potential as proposed [5, 10, 11] but rather a result of delayed hGFAP activity in certain RG populations.
As the recombination observed in Cre/loxP tracing reflects transcriptional activity of the promoter driving Cre, lineages can be traced from the time points when the molecular programs responsible for inducing promoter activity first become active. Results using endogenous mouse RG promoters such as Blbp and Glast therefore record lineages from the time when cells first become molecularly distinguishable as RG. The validity of this approach has been questioned by the suggestion that the presence of both molecular and morphological features are required in order to define cell types . However, this idea does not take into account the fact that molecular changes necessarily precede and drive subsequent morphological changes. For example, the transcription factor Hb9 is required for normal specification of spinal cord motoneurons  and ectopic Hb9 expression is sufficient to induce motoneuron fate . However, axon extension and innervation of skeletal muscle are not observed until at least one day after Hb9 transcription is induced . Thus, although the presence of distinct morphology can allow cell types to be positively identified, their absence does not exclude a given identity during development since many morphological characteristics of differentiated cells appear only after the required molecular changes take place.
In addition to such reasoning, several lines of evidence support the idea that the NE→RG transition occurs in the forebrain by E10. First, expression of both Blbp and Glast becomes detectable in the forebrain by this time (Figure 2); thus, two different RG genes are induced at the same time. Second, Notch signaling promotes the RG phenotype [21, 22] and Blbp is a direct Notch target gene ; therefore, induction of Blbp expression indicates that Notch-dependent RG differentiation has occurred. Finally, mice lacking ErbB2, a gene required for RG differentiation, begin showing impaired RC2 labeling in the telencephalon from as early as E9.5  (some recombination restricted to the most rostral portion of the forebrain was detectable in E9.5 Blbp::Cre;R26R embryos ). Taken together, these data demonstrate that the molecular programs responsible for triggering RG differentiation are already active in some forebrain regions by E9.5 and induce broad expression of RG target genes by E10.5. They further suggest that an understanding of the mechanisms regulating the NE→RG transition will require analysis of molecular events occurring between these time points.
An important technical issue regarding Cre/loxP lineage tracing is that recombination efficiency is directly dependent upon cellular recombinase concentrations; this point must be considered when comparing results obtained using distinct transgenic or targeted lines. For instance, when four different promoters of varying strengths were used to drive Cre and compared in embryonic stem cells, recombination efficiency was directly correlated with promoter activity: the strongest promoter drove the highest levels of recombination . Furthermore, even when the identical promoter is used, recombination efficiency can vary dramatically according to copy number. For example, merely doubling the copy number of a β-actin::Cre transgene (from one to two copies) resulted in a 10-fold increase in the number of recombined cells in the brain . For these reasons, caution must be used when interpreting negative results in lineage tracing experiments as a lack of recombination could reflect weak promoter activity rather than restricted progenitor potential. Indeed, differences in copy number likely explain why subsequently generated hGFAP::Cre transgenic lines yielded higher levels of neuronal recombination  than observed using the original line [5, 9]. It will be important to keep these issues in mind when comparing lineage tracing results in which the same genomic locus was either targeted (that is, knockin) or used as a transgene. Whereas knockins can, at most, provide two targeted alleles, tens to hundreds of copies are typical for transgenics . Therefore, in many cases, transgenics may be more likely than knockins to drive recombination at the time when transcription commences and consequently should give a more accurate picture of the progeny derived from a given molecularly defined progenitor population.
Although the onset of Blbp and Glast transcription distinguishes RG from nestin + Blbp - Glast - NE, the available evidence indicates that, as a population, Blbp + Glast + RG themselves transit through multiple distinct developmental stages. For example, the hGFAP promoter does not drive any recombination in cortex prior to E12.5 (data not shown). As rodent Gfap is not normally expressed so early in RG neurogenesis, it is possible that the timing of hGFAP activity in mice merely reflects the peculiarities of a heterologous promoter. However, it is alternatively possible that a fundamental transition occurring at E12.5 upregulates hGFAP activity; if so, the latter stage might represent the terminal phase of RG generation of layer VIb neurons as only about 30% of these neurons were recombined in hGFAP::Cre;R26R mice . It is also well known that RG ultimately cease generating neurons and switch to production of glia; although several pathways have been implicated in promoting gliogenesis (for example, Jak-STAT signaling  and NFI genes ), a comprehensive description of the mechanisms involved is lacking and requires transcriptional profiling and comparisons between neurogenic and gliogenic RG. Steps in this direction have been taken by transcriptome analyses of FACS purified RG whose neurogenic potential was determined by in vitro clonal analyses . However, given that progenitor neurogenic potential in vivo can differ from that observed in vitro , subsequent studies might yield additional insights by defining potential on the basis of either in vivo transplantation or genetic lineage tracing. Accordingly, our lineage tracing data provide in vivo evidence that hGFAP promoter activity commences in ventral telencephalic RG only after neurogenesis is largely completed and thus distinguishes a later Blbp + Glast + hGFAP + gliogenic stage from an earlier Blbp + Glast + hGFAP -neurogenic stage in this region. This is significant as it enables direct in vivo comparisons of neurogenic and gliogenic ventral telencephalic RG via FACS purification and microarray analyses using hGFAP::eGFP and Blbp::eGFP or Glast::eGFP transgenic mice. Such studies should facilitate efforts to more thoroughly determine the molecular mechanisms that control progenitor potential and fate specification.
These data establish the major neurogenic role of RG in the developing CNS and genetically distinguish an early neurogenic Blbp + Glast + hGFAP - stage from a later gliogenic Blbp + Glast + hGFAP + stage in the ventral telencephalon.
Materials and methods
Mice and BAC transgenesis
All protocols were approved by the Rockefeller University IACUC. Glast::Cre BAC transgenic mice were generated using homologous recombination to insert a Cre-SV40polyA cassette at the site of the endogenous ATG (exon II, BAC clone RP23-63O21). Transgenic FVB/N mice were generated using standard procedures . R26RlacZ, Z/EG, hGFAP::Cre, and hGFAP::eGFP mice were obtained from JAX (Bar Harbor, Maine, USA). Glast::eGFP mice were obtained from GENSAT (New York, NY, USA). The BAC clone used to generate Glast::eGFP mice is the same as that used for the Glast::Cre mice and the eGFP-polyA cassette was targeted to the identical locus in the BAC as the Cre-polyA cassette. Analysis of endogenous gene expression was done using C57BL/6J (JAX).
Tissue processing and analysis
Tissues were either immersion fixed (embryos) or perfused (postnatal/adult) with 4% paraformaldehyde, cryoprotected in 15% sucrose in phosphate-buffered saline (PBS), and cut on a cryostat (15 μm) or vibratome (75 μm). Cryosections were dried at room temperature for 2 hours, a step that improved signal (insufficient drying was responsible for the previous inability to detect BLBP protein in the E10.5 forebrain ). Primary antibodies used were: rabbit α-BLBP (1: 1500); rabbit α-β gal (1:500; ICN Santa Ana, CA, USA)); mouse α-Calbindin (1:500; Swant, Bellinzona (Switzerland)); rabbit α-Cre (1:5000 and TSA amplified; Novagen, San Diego, CA, USA); goat α-GFP (1:500; US Biological, Swampscott, MA, USA); guinea pig α-GLAST (1:5000; Millipore, Billerica, MA, USA); mouse α-NeuN (1:200; Millipore, Billerica, MA, USA); rabbit α-phosphorylated histone H3 (PH3, 1:250; Millipore, Billerica, MA, USA). Alexa 488-coupled secondary antibodies were obtained from Molecular Probes, and Cy3-coupled secondaries were from Jackson Immunoresearch (West Grove, PA, USA); both were used at 1:750. Sections were preblocked in 5% donkey serum, 0.1% Triton X-100 in PBS and incubated with primary antibodies overnight at 4°C. After washing, sections were incubated with secondary antibodies (generated in donkey) for 2 hours in the same solution at room temperature. Confocal imaging was done on an LSM 510 Axioplan (Zeiss). For X-gal histochemical staining, tissues were incubated at 37°C in X-Gal solution: 1 mg/ml X-gal (Sigma, St. Louis, MO, USA)), 20 mM K3Fe(CN)6, 20 mM K4Fe(CN)6, 2 mM MgCl2 and 0.02% NP-40 in PBS.
Riboprobes to detect Cre mRNA were generated using T7 polymerase off PCR amplified template DNA generated with the following primers: forward, CAAGCTCGAAATTAACCCTCACTAAAGGGgtccaatttactgaccgtacacc; reverse, AGAGAGCGGTAATACGACTCACTATAGGGCctaatcgccatcttccagcag. Probe hybridization and development was done using standard protocols.
We are grateful to Robert Machold (NYU) and Alison North (RU bioimaging facility) for assistance with imaging, to Mary-Beth Hatten for generous provision of reagents, and to Gord Fishell and David Anderson for critically reading drafts of this manuscript. NH is an HHMI Investigator and TEA was supported by HHMI, an NIH NRSA fellowship (grant #1F32HD055198-01), and a Beckman Fellowship.
- Rakic P: Elusive radial glial cells: historical and evolutionary perspective. Glia 2003, 43:19–32.View ArticlePubMed
- Kriegstein AR: Constructing circuits: neurogenesis and migration in the developing neocortex. Epilepsia 2005,46(Suppl 7):15–21.View ArticlePubMed
- Noctor SC, Flint AC, Weissman TA, Dammerman RS, Kriegstein AR: Neurons derived from radial glial cells establish radial units in neocortex. Nature 2001, 409:714–720.View ArticlePubMed
- Malatesta P, Hartfuss E, Gotz M: Isolation of radial glial cells by fluorescent-activated cell sorting reveals a neuronal lineage. Development 2000, 127:5253–5263.PubMed
- Malatesta P, Hack MA, Hartfuss E, Kettenmann H, Klinkert W, Kirchhoff F, Gotz M: Neuronal or glial progeny: regional differences in radial glia fate. Neuron 2003, 37:751–764.View ArticlePubMed
- Miyata T, Kawaguchi A, Okano H, Ogawa M: Asymmetric inheritance of radial glial fibers by cortical neurons. Neuron 2001, 31:727–741.View ArticlePubMed
- Tamamaki N, Nakamura K, Okamoto K, Kaneko T: Radial glia is a progenitor of neocortical neurons in the developing cerebral cortex. Neurosci Res 2001, 41:51–60.View ArticlePubMed
- Anthony TE, Klein C, Fishell G, Heintz N: Radial glia serve as neuronal progenitors in all regions of the central nervous system. Neuron 2004, 41:881–890.View ArticlePubMed
- Zhuo L, Theis M, Alvarez-Maya I, Brenner M, Willecke K, Messing A: hGFAP-cre transgenic mice for manipulation of glial and neuronal function in vivo. Genesis 2001, 31:85–94.View ArticlePubMed
- Gotz M, Huttner WB: The cell biology of neurogenesis. Nat Rev Mol Cell Biol 2005, 6:777–788.View ArticlePubMed
- Gotz M, Barde YA: Radial glial cells defined and major intermediates between embryonic stem cells and CNS neurons. Neuron 2005, 46:369–372.View ArticlePubMed
- Doetsch F: The glial identity of neural stem cells. Nat Neurosci 2003, 6:1127–1134.View ArticlePubMed
- Kriegstein AR, Gotz M: Radial glia diversity: a matter of cell fate. Glia 2003, 43:37–43.View ArticlePubMed
- Soriano P: Generalized lacZ expression with the ROSA26 Cre reporter strain. Nat Genet 1999, 21:70–71.View ArticlePubMed
- Novak A, Guo C, Yang W, Nagy A, Lobe CG: Z/EG, a double reporter mouse line that expresses enhanced green fluorescent protein upon Cre-mediated excision. Genesis 2000, 28:147–155.View ArticlePubMed
- Petersen PH, Zou K, Hwang JK, Jan YN, Zhong W: Progenitor cell maintenance requires numb and numblike during mouse neurogenesis. Nature 2002, 419:929–934.View ArticlePubMed
- Hartfuss E, Galli R, Heins N, Gotz M: Characterization of CNS precursor subtypes and radial glia. Dev Biol 2001, 229:15–30.View ArticlePubMed
- Arber S, Han B, Mendelsohn M, Smith M, Jessell TM, Sockanathan S: Requirement for the homeobox gene Hb9 in the consolidation of motor neuron identity. Neuron 1999, 23:659–674.View ArticlePubMed
- Tanabe Y, William C, Jessell TM: Specification of motor neuron identity by the MNR2 homeodomain protein. Cell 1998, 95:67–80.View ArticlePubMed
- Wichterle H, Lieberam I, Porter JA, Jessell TM: Directed differentiation of embryonic stem cells into motor neurons. Cell 2002, 110:385–397.View ArticlePubMed
- Patten BA, Peyrin JM, Weinmaster G, Corfas G: Sequential signaling through Notch1 and erbB receptors mediates radial glia differentiation. J Neurosci 2003, 23:6132–6140.PubMed
- Gaiano N, Nye JS, Fishell G: Radial glial identity is promoted by Notch1 signaling in the murine forebrain. Neuron 2000, 26:395–404.View ArticlePubMed
- Anthony TE, Mason HA, Gridley T, Fishell G, Heintz N: Brain lipid-binding protein is a direct target of Notch signaling in radial glial cells. Genes Dev 2005, 19:1028–1033.View ArticlePubMed
- Schmid RS, McGrath B, Berechid BE, Boyles B, Marchionni M, Sestan N, Anton ES: Neuregulin 1-erbB2 signaling is required for the establishment of radial glia and their transformation into astrocytes in cerebral cortex. Proc Natl Acad Sci USA 2003, 100:4251–4256.View ArticlePubMed
- Araki K, Imaizumi T, Okuyama K, Oike Y, Yamamura K: Efficiency of recombination by Cre transient expression in embryonic stem cells: comparison of various promoters. J Biochem 1997,122(5):977–982.PubMed
- Zong H, Espinosa JS, Su HH, Muzumdar MD, Luo L: Mosaic analysis with double markers in mice. Cell 2005, 121:479–492.View ArticlePubMed
- Casper KB, McCarthy KD: GFAP-positive progenitor cells produce neurons and oligodendrocytes throughout the CNS. Mol Cell Neurosci 2006, 31:676–684.View ArticlePubMed
- Nagy A, Gertsenstein M, Vintersten K, Behringer R, Eds: Manipulating the Mouse Embryo. Third edition. Cold Spring Harbor: Cold Spring Harbor Laboratory Press; 2003.
- He F, Ge W, Martinowich K, Becker-Catania S, Coskun V, Zhu W, Wu H, Castro D, Guillemot F, Fan G, et al.: A positive autoregulatory loop of Jak-STAT signaling controls the onset of astrogliogenesis. Nat Neurosci 2005, 8:616–625.View ArticlePubMed
- Deneen B, Ho R, Lukaszewicz A, Hochstim CJ, Gronostajski RM, Anderson DJ: The transcription factor NFIA controls the onset of gliogenesis in the developing spinal cord. Neuron 2006, 52:953–968.View ArticlePubMed
- Pinto L, Mader MT, Irmler M, Gentilini M, Santoni F, Drechsel D, Blum R, Stahl R, Bulfone A, Malatesta P, et al.: Prospective isolation of functionally distinct radial glial subtypes – lineage and transcriptome analysis. Mol Cell Neurosci 2008, 38:15–42.View ArticlePubMed
- Mukouyama YS, Deneen B, Lukaszewicz A, Novitch BG, Wichterle H, Jessell TM, Anderson DJ: Olig2+ neuroepithelial motoneuron progenitors are not multipotent stem cells in vivo. Proc Natl Acad Sci USA 2006, 103:1551–1556.View ArticlePubMed
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.