Motoneurons and oligodendrocytes are sequentially generated from neural stem cells but do not appear to share common lineage-restricted progenitors in vivo

One of the major goals of developmental biology is to elucidate the lineage relationships between populations of cells. To this end, different models have been proposed for the genesis of cells within the central nervous system (CNS)(Anderson, 2001; Briscoe et al., 2000; Noble et al., 2004; Stiles, 2003). It is generally thought that CNS development starts from neuroepithelial stem cells (NSCs),which ultimately give rise to the three major cell types (neurons,oligodendrocytes and astrocytes) of the CNS. However, the intermediate steps that NSCs take to generate the differentiated cell types in vivo have not been resolved. In vitro work showed that neuronal and glial restricted progenitors(i.e. NRPs and GRPs) can be isolated from embryonic spinal cord with the potential to generate only neurons or only glia, respectively(Mayer-Proschel et al., 1997; Rao and Mayer-Proschel, 1997; Rao et al., 1998). This led to a model (Fig. 1A) proposing that a homogeneous NSC population within the CNS first generates lineage-restricted progenitors, NRPs and GRPs, which then go on to generate neuronal and glial cell types, respectively. More recent studies, however,making use of grafting experiments and patterning analyses, have shown that the early neural tube is not homogeneous and is subdivided into several different domains, each of which develops into different differentiated cell types. Thus, in vivo lineage relationships may differ from those predicted by in vitro studies. In particular, it has been proposed that oligodendrocytes and astrocytes arise from bipotential progenitors in spatially distinct domains, which generate combinations of one neuronal and one glial cell type,rather than from lineage-restricted glial progenitors(Briscoe et al., 2000; Richardson et al., 2000; Stiles, 2003).

The differences between these two models are exemplified by recent studies on two basic helix-loop-helix genes, Olig1 and Olig2(Gabay et al., 2003; Lu et al., 2002; Zhou and Anderson, 2002; Zhou et al., 2001). These two genes are both expressed in cells of the motoneuron domain (pMN), a progenitor domain that has been shown to be involved in motoneuron and oligodendrocyte specification. Mice lacking Olig2 or Olig1/2 display an absence of both motoneurons and oligodendrocytes in the spinal cord, although excess numbers of V2 interneurons and astrocytes are generated(Lu et al., 2002; Takebayashi et al., 2002; Zhou and Anderson, 2002). In addition, lineage analysis using Olig1-Cre and a Rosa26-lacZ reporter has shown that motoneurons and oligodendrocytes, but not astrocytes, are included in the Olig1 lineage (Lu et al.,2002). One interpretation (Fig. 1B) of these data is that common motoneuron and oligodendrocyte progenitors (MNOP) exist that do not generate astrocytes(Lu et al., 2002; Noble et al., 2004; Rowitch et al., 2002). This interpretation suggests that GRPs, as initially defined from in vitro clonal analysis, may not be a major participant in this context in vivo(Fig. 1A). Others have argued,however, that Rosa26-lacZ is not easily detectable in astrocytes(Malatesta et al., 2003), and thus previous lineage analysis using Olig1-Cre and Rosa26-lacZ may be misleading. Furthermore, three recent papers have convincingly shown that Olig⁺ OPCs can be generated across much broader progenitor domains than previously thought, suggesting that the MNOP model may be oversimplified(Cai et al., 2005; Fogarty et al., 2005; Vallstedt et al., 2005). Alternatively, therefore, the data can be reasonably explained by the NRP/GRP model: while the continuous presence of Olig1/2 is required for NRPs and GRPs to generate motoneurons and oligodendrocytes, the downregulation of Olig1/2 in these progenitors will direct them to generate interneurons and astrocytes,respectively (Fig. 1A)(Liu and Rao, 2003; Liu and Rao, 2004). However,neither the MNOP nor the NRP/GRP model can readily accommodate observations from retroviral-marking experiments indicating that motoneurons and astrocytes derive from a shared lineage, even at late stages(Leber et al., 1990). Therefore, a third model (Fig. 1C) that appears to be consistent with existing data would be that NSCs in the pMN domain sequentially generate separate Olig⁺motoneuron and oligodendrocyte precursors, and probably Olig-negative astrocyte precursors, without generating intermediate progenitors (i.e. MNOPs or GRPs). In the absence of Olig1/2 function, motoneuron and oligodendrocyte precursors transform into V2 interneurons and astrocytes, respectively.

To gain further insight into the lineage relationship between motoneurons and oligodendrocytes in vivo, we created a conditional cell-ablation mouse line in which the Diphtheria toxin gene has been targeted into the Rosa26 locus and is activated upon Cre-mediated recombination(Maxwell et al., 1987; Soriano, 1999). We demonstrate the usefulness of this genetic cell-ablation strategy to study the above discussed cell lineage relationships in vivo, by crossing it with the Olig1-Cre knock-in mouse (Lu et al.,2002). We find that ablation of Olig1-Cre lineage cells eliminates differentiated motoneurons and oligodendrocytes, a finding that recapitulates the phenotype of Olig1/2 double-knockout mice. However, at twelve days of gestation (E12), when motoneuron precursors are killed in our system,normal numbers of oligodendrocyte precursors are still being generated. Moreover, even after these oligodendrocyte precursors are subsequently killed,astrocytes are generated normally. This result is inconsistent with the MNOP model and strongly supports a sequential progenitor model(Fig. 1C)(Qian et al., 2000; Richardson et al., 2000). This model not only explains the results from Olig1-Cre-mediated ablation and the previous Olig1/2 knockout data, but can also accommodate in vivo the in vitro observations that led to the NRP/GRP model.

The DTA176 coding sequence was PCR amplified from pIBI30-176(Maxwell et al., 1987) and subcloned into the NheI and NotI sites of pBigT(Srinivas et al., 2001), which was subcloned into pRosa26PA to make the final targeting vector. The linearized targeting vector was electroporated into R1 embryonic stem cells. From 72 colonies analyzed by Southern hybridization, five correctly targeted clones were obtained. One of the positive ES cell lines was used for blastocyst injection to generate chimeric mice. Germline transmission from chimeric mice was confirmed by Southern transfer analysis. PCR was used for all later genotyping with three primers:5′-GTTATCAGTAAGGGAGCTGCAGTGG-3′,5′-AAGACCGCGAAGAGTTTGTCCTC-3′ and 5′-GGCGGATCACAAGCAATAATAACC-3′. The product is 302 bp for DTA and 415 bp for wild type, with PCR conditions of 94°C for 30 seconds,59.5°C for 30 seconds and 72°C for 30 seconds, for 32-35 cycles.

Noon of the plug day was considered as embryonic day 0.5 (E0.5). Mouse embryos were dissected and fixed in 4% paraformaldehyde in PBS for 45 minutes to overnight depending on the age. Cryosections were cut at 12 μm and collected on superfrost plus slides (Fisher). For cell counts at E12.0, more than six embryos were used for each genotype, and Chx10⁺ or Olig2⁺ cells in five 12 μm sections (60 μm between two sections), spanning 240 μm of the brachial region, were counted. The following primary antibodies were used for immunostaining. Sheep anti-Chx10(1:1000, Exalpha), monoclonal anti-CNP (1:250, Sigma), rabbit anti-Cre(1:3000, Covance), monoclonal anti-Cre (1:500, Sigma), rabbit anti-Gfap(1:1000, Dako), rabbit anti-Hb9 (1:8000, gift from Samuel Pfaff, The Salk Institute), rabbit anti-Irx3 (1:4000, gift from Thomas Jessell, Columbia University), rabbit anti-Isl1 (1:2500, gift from Samuel Pfaff), rabbit anti-MBP (1:200, Chemicon), rabbit anti-Nkx2.2 (1:4000, gift from Thomas Jessell), guinea pig anti-Nkx6.1 (1:2500, gift from Thomas Jessell), rabbit anti-Olig2 (1:8000, gift from Charles Stiles, Harvard Medical School), rat anti-Pdgfrα (1:200, Pharmingen). Monoclonal antibodies against Hnf3β, Nkx2.2, Pax6, RC2, and Shh were obtained from the Developmental Studies Hybridoma Bank.

Immunohistochemistry was performed first to detect Cre or Olig2 antigens,followed by the TUNEL assay using a Fluorescein In Situ Cell Death Detection Kit from Roche, performed according to the manufacturer's instructions.

To better understand the lineage relationships of motoneurons and oligodendrocytes, we first extended previous work on the analysis of Olig1 and Olig2 expression in the developing spinal cord(Lu et al., 2000; Novitch et al., 2001; Zhou et al., 2000). Olig1 and Olig2 are first expressed at embryonic day 8.5 (E8.5) (somites 8-10) in the ventral part of the rostral spinal cord, just dorsal to the floor plate(Fig. 2A,I). Olig2 expression can be detected slightly earlier than Olig1 expression. By E10.0, both genes reach peak expression in the broadest area, and all E10.0 Olig2⁺cells express Olig1, and vice versa (Fig. 2B,J,K). However, from E10.5 to E11.5, the number of Olig1⁺ cells gradually decreases whereas the number of Olig2⁺ cells appears relatively unchanged(Fig. 2C-E,L-N). Beginning at E12.0, the number of Olig2⁺ cells increases dramatically(Fig. 2O-Q), and a small number of Olig1⁺ cells re-appear at ∼E11.75(Fig. 2F-H, and data not shown). More and more Olig1/2⁺ cells can be seen away from the ventricular zone from ∼E12.5 and express the oligodendrocyte precursor cell (OPC) marker platelet-derived growth factor receptor α(Pdgfrα). Double staining of Olig1 or Olig2 with Pdgfrα indicates that almost of all these Olig2⁺ cells are Pdgfrα⁺/Olig1⁺(Fig. 2H,Q).

We also performed in vivo cell lineage analysis using the Olig1-Cre and Rosa26-eYFP mouse lines, which constitutively express eYFP once activated by Cre. From E10.5 to E12.0 and beyond, all Olig2⁺ cells are YFP⁺ (see Fig. S1 in the supplementary material, and data not shown), indicating that they or their precursors must have expressed Cre under Olig1 control. Therefore, a simple conclusion from the Olig1 and Olig2 expression data is that these two genes are expressed in the cells of the same lineage in the spinal cord, although Olig2 appears to be expressed slightly earlier. Because all Olig2⁺ cells are included in the Olig1 lineage, Olig2 expression can serve as a useful marker for monitoring the ablation of Olig1-Cre-expressing cells.

We reasoned that the use of diphtheria toxin to kill Olig-expressing cells should help to distinguish between the different lineage models in vivo for the genesis of cells in the CNS (Fig. 1). In their simplest form, these three models offer different predictions that can be tested using the Rosa26-DTA176 allele in combination with the Olig1-Cre allele. For example, in compound heterozygotes containing Olig1-Cre and Rosa26-DTA176 (Olig1-DTA), all three models would predict an absence of motoneurons and oligodendrocytes owing to Olig1 expression in their progenitors. However, assuming a single pathway for formation of GRP cells that are themselves Olig⁺, the NRP/GRP model would predict a dramatic decrease of astrocytes, whereas the MNOP and sequential models would predict normal astrocyte formation. To conditionally kill Olig1-Cre expressing cells, we created a conditional cell-ablation mouse line that should be useful for a wide range of cell ablation studies. The ubiquitously expressed Rosa26 locus has been used extensively for the generation of Cre and Flp reporter strains (Awatramani et al.,2001; Soriano,1999; Srinivas et al.,2001). Diphtheria toxin polypeptide has a toxic fragment A (DTA)and a ligand fragment B that is required for the transfer of DTA into cells. Once inside a cell, one molecule of DTA appears to be sufficient to kill cells(Yamaizumi et al., 1978). To reduce potential problems associated with leaky DTA expression from the Rosa26 locus prior to Cre-mediated activation, we used an attenuated form of fragment A (DTA176) (Maxwell et al.,1987), which is toxic at ∼100-200 molecules per cell. With the attenuated DTA, leaky, Cre-independent expression is less likely to result in cell lethality, but following Cre recombination more than enough DTA176 should be produced to yield the desired cell killing. Unlike most other mammalian cells, mouse cells have no receptor for diphtheria toxin(Mitamura et al., 1995);therefore, strict cell-autonomous ablation is ensured even upon possible release of active DTA176 molecules from dead cells. By gene targeting, we successfully generated heterozygous and homozygous Rosa26-DTA176 mice(Fig. 3), all of which in the absence of Cre appeared normal and healthy and were obtained at the expected Mendelian ratios.

Compound heterozygous Olig1^Cre/+; Rosa26^DTA176/+(Olig1-DTA) mice were able to survive up to E18.0, but were not recovered as live newborns. Because motoneuron generation precedes oligodendrocyte generation, we first examined the genesis of motoneurons in Olig1-DTA embryos using the motoneuron markers Isl1 and Hb9. Embryos heterozygous for only Rosa26-DTA176 or Olig1-Cre showed no discernible phenotypic differences from wild type and are collectively referred to as controls. Normally motoneurons are generated from the ventral spinal cord during the period from E9.5 to E12.0. At E10.0, very few Isl1⁺ or Hb9⁺ somatic motoneurons could be detected in Olig1-DTA compound heterozygous animals when compared with controls from the same litter (compare Fig. 4A,B with 4C,D). We examined later stages of development for motoneuron markers, and obtained similar results from E10.5-E14.0 (Fig. 4E-P, and data not shown). Interestingly, although Isl1/Hb9-positive cells were mostly missing in the ventral horns of Olig1-DTA embryos from E11.0 to E12.0, some were still detected around the ventricular zone(Fig. 4G,H,K,L, arrowheads),albeit in lower numbers than in controls(Fig. 4E,F,I,J, arrowheads). These residual Isl1⁺/Hb9⁺ cells appeared to be newly generated, and were killed at later stages(Fig. 4O,P) when DTA had sufficient time to kill them. The continuous killing of motoneurons therefore closely parallels the normal continuous generation of motoneurons during this developmental period. In summary, ablation of the motoneuron lineage in the developing spinal cord is very efficient in our cell ablation system, and the ablation of early-born motoneuron precursors does not appear to lead to a compensatory generation of later-born motoneuron precursors.

To determine whether ablation is specific to Olig1-Cre-expressing cells, we took advantage of homeodomain markers with well-characterized expression patterns in the spinal cord (Briscoe et al., 2000). If cell ablation by Olig1-DTA is specific, only the Olig1-Cre-expressing pMN domain should be affected, leaving all other dorsal and ventral progenitor domains that do not express Olig1 unaffected. In E10.5 control embryos, Irx3 expression marks the dorsal boundary of pMN(Fig. 5A); Pax6 is strongly expressed throughout the p2 and more dorsal domains, and weakly expressed in the pMN domain (Fig. 5B);Nkx2.2 marks the p3 domain (Fig. 5D); Nkx6.1 is expressed in the p2, pMN and p3 domains(Fig. 5A,C); and Hnf3β and Shh are specific to the floor plate (Fig. 5E, and data not shown). No differences in Irx3, Hnf3β or Shh expression were found between Olig1-DTA and control embryos(Fig. 5G,K, and data not shown). Pax6 expression in the p2 and more dorsal domains of Olig1-DTA embryos was unchanged, but normally weak expression in the pMN domain was not detected in Olig1-DTA embryos (Fig. 5B,H).

Surprisingly, the p3 domain defined by expression of both Nkx2.2 and Nkx6.1 was not present in Olig1-DTA embryos (Fig. 5G,I,J), and Olig2⁺ cells spread just dorsal to the floor plate (compare Fig. 5E to 5K). This prompted us to re-examine the early expression of Olig1 in Olig1-Cre embryos because it was assumed that Olig1 is not expressed in the p3 domain. Using a Cre antibody, we confirmed that Olig1 is transiently expressed in the Nkx2.2⁺ domain from E8.5 to E9.0 in Olig1-Cre animals (Fig. 2I, and data not shown). Lineage analysis of E10.5 Olig1-Cre/Rosa26-eYFP mice also verified early Olig1 expression in this domain (Fig. 5M-O), thereby confirming that specific Olig1-DTA-mediated ablation was responsible for the absence of the p3 domain. These results are consistent with a previous report that early expression of the Oliggene occurs just dorsal to the floor plate in chick(Novitch et al., 2001).

We also examined dorsoventral patterning at earlier and later developmental stages in Olig1-DTA embryos (data not shown), and found that the p2 and more dorsal progenitor domains were intact, similar to those observed at E10.5. In addition, we asked whether p2 domain-derived V2 interneurons are altered by following Chx10 expression during E11.0 to E14.0(Fig. 5P,Q, and data not shown). V2 interneurons were unaffected, as Chx10 expression patterns remained essentially unaltered in the Olig1-DTA embryos when compared with control. Further confirming these immunostaining data, cell counts of E12.0 Chx10⁺ cells revealed no difference between Olig1-DTA and control animals (Fig. 5R). Taken together, cell ablation in the Olig1-DTA system correlates with Olig1 (Cre)expression and appears strictly cell autonomous. The killing of Olig1-expressing cells does not appear to have any bystander effect on neighboring cells, and the integrity of non-Olig1 expression domains is well maintained.

Because oligodendrocyte generation starts after motoneuron generation is complete, we next examined the status of oligodendrocytes in those animals. Pdgfrα is one of the earliest expressed markers known for OPCs, with expression commencing at ∼E12.5(Pringle and Richardson, 1993)(Fig. 6A-C). Mature oligodendrocytes can be identified from ∼E16.0 onward by expression of the markers myelin basic protein (MBP) and 2′,3′-cyclic nucleotide 3′-phosphodiesterase (CNP) (Liu et al., 2002; Zhou and Anderson,2002; Zhou et al.,2001) (Fig. 6M,N,and data not shown). These oligodendrocyte markers were not detected in Olig1-DTA embryos at any stage examined from E12.5 to E18.0(Fig. 6D-F,O,P, and data not shown), indicating that Olig1-DTA-mediated killing of oligodendrocytes is extremely efficient. Olig2 expression, which at E12.0 and later stages marks potential oligodendrocyte precursors, was mostly absent from E14.0 embryos; a few Olig2⁺ cells could be detected in a focal area of the ventricular zone (compare Fig. 6G with 6J). Interestingly, a few Olig2⁺ cells could be detected in the p2 and more dorsal domains at E16.5 (Fig. 6K). Even fewer Olig2⁺ cells were found in these regions at E18.0 (Fig. 6L) in Olig1-DTA embryos. These Olig2⁺ cells also expressed Olig1-Cre(data not shown), and are presumed to be destined for cell death, as none of these Olig2⁺ cells progressed to Pdgfrα-expressing OPCs(Fig. 6E,F). This finding is consistent with three recent studies suggesting that Olig2⁺ OPCs can be generated after ∼E14.5 from more dorsal progenitor domains(Cai et al., 2005; Fogarty et al., 2005; Vallstedt et al., 2005). These results indicate that oligodendrocytes are specifically ablated in the Olig1-DTA system.

Astrocyte generation begins about three days after OPCs first appear. In Olig2 or Olig1/2 double knockout animals, more astrocytes are generated at the expense of oligodendrocyte formation(Lu et al., 2002; Takebayashi et al., 2002; Zhou and Anderson, 2002). If most astrocytes are generated through an Olig-positive progenitor, Olig1-DTA animals should exhibit a dramatic decrease in, or complete absence of,astrocytes. To visualize astrocytes, we examined the expression of the astrocyte marker, glial fibrillary acidic protein (Gfap)(Bignami et al., 1972). In control embryos, Gfap is first expressed at ∼E15.5 from the ventrolateral portion of the spinal cord. No obvious difference was seen between E16.5 Olig1-DTA and control mice with respect to the Gfap expression pattern(Fig. 6Q,S). At E18.0, the Gfap staining in the ventral horns of Olig1-DTA embryos appeared to be slightly increased (Fig. 6R,T); this spread of astrocytes may reflect the lack of motoneurons in this space. The p3 domain has recently been shown to normally generate astrocytes in vivo(Pringle et al., 2003), and it was essentially eliminated in Olig1-DTA embryos owing to its early expression of Olig1. Therefore, the astrocytes generated from the p3 domain should presumably be missing in Olig1-DTA animals. However, because the filamentous staining pattern of Gfap precluded an accurate cell count, we were not able to observe a clear reduction of Gfap staining. In summary, the generation of astrocytes does not appear to be significantly affected by Olig1-DTA-mediated cell ablation.

Because DTA-mediated ablation occurs by cell-autonomous apoptosis(Morimoto and Bonavida, 1992),we were able to examine the time course of DTA176-mediated cell ablation in our system through TUNEL analysis on Olig1-DTA and control embryos from E8.5 to E16.5 (Fig. 7 and data not shown). It appears from TUNEL staining that DTA176-induced apoptosis began at E9.0 (∼14 somites), which is approximately 12 hours after the onset of Olig1-Cre expression at E8.5 (Fig. 7H,O). Massive cell death occurred from E9.0 to E11.5(Fig. 7H-L,O-S). Interestingly,a population of Olig2⁺ cells is continuously present in Olig1-DTA embryos from the time of motoneuron genesis through the time of OPC generation. In particular, at E12.0, cell counting indicated a similar number of Olig2⁺ cells in control and Olig1-DTA mice(Fig. 7F,M, Fig. 8A-C). Because Cre⁺ cells take less than 12 hours to become TUNEL positive, and Olig1 and Olig2 expression completely overlap at ∼E10.0, all of the E10.0 Olig2⁺ cells in Olig1-DTA embryos would be predicted to die by E10.5. Therefore, it is unlikely that a separate population of Olig2⁺ cells is generated at E10.0 and survive to E12.0. The persistence of Olig2⁺ cells in Olig1-DTA embryos is likely to refect their continuous generation, and suggests that E12.0 Olig2⁺cells are not derived from previous E10.0 Olig1/Olig2⁺ cells, but instead are newly born from pMN NSCs, or originate from a fate change in neighboring domains. Because the p3 domain is deleted by E9.0(Fig. 5G,I,J, and data not shown), it is an improbable source for Olig2⁺ cells. Also, the largely unaffected p2 and more dorsal domains, and the normal generation of V2 interneurons (Fig. 5, and data not shown) in Olig1-DTA spinal cords, suggest that the Olig2⁺ cells observed at E12.0 in the Olig1-DTA embryos originate from resident NSCs in the pMN domain.

The normal generation of Olig2⁺ cells in E12.0 Olig1-DTA embryos suggests a continuous presence of Olig-negative NSCs in the pMN domain. The unchanged p2 and more dorsal domains, and the normal generation of Chx10⁺ V2 interneurons from the p2 domain(Fig. 5), suggest a normal presence of NSCs in these domains as well. We have observed early expression of Olig1 in the Nkx2.2⁺ p3 domain of normal mice, and Nkx2.2⁺ cells are largely missing in Olig1-DTA embryos at E10.5(Fig. 5J). Therefore, we further examined Nkx2.2 expression at later developmental stages. Very few Nkx2.2⁺ cells in the p3 domain were detected in older Olig1-DTA embryos from E11.5 to E13.25 (compare Fig. 8D-F with 8G-I). This result suggests that early Olig1-Cre expression encompasses all p3 NSCs and that the killing results in an absence of NSCs in the p3 domain throughout later developmental stages. To further confirm this inference regarding NSCs in different domains, we assayed the presence of radial glia, which are proposed to be NSCs (Alvarez-Buylla et al.,2001; Lyons et al.,2003), using the radial glia-specific marker RC2 (Ifaprc2 - Mouse Genome Informatics). Within the Olig2⁺ pMN domain, compared with controls, RC2 expression was relatively unchanged in Olig1-DTA spinal cords. However, RC2 expression was largely eliminated in the p3 domain of the ablation system (compare Fig. 8J with 8K).

By using conditional cell ablation to dissect the lineage relationships of motoneurons and oligodendrocytes in the developing spinal cord, we have provided evidence in strong support of a sequential model for spinal cord neurogenesis and gliogenesis. The sequential model suggests that NSCs in the pMN domain persist throughout development of the spinal cord. These Olig-negative NSCs sequentially give rise to Olig⁺ MN precursors and Olig⁺ OPCs at different developmental stages. They might also give rise to astrocyte precursors (Olig^-?) and are likely to persist as adult stem cells (Fig. 1C). Although motoneuron precursors and OPCs probably share the same NSC pool in the pMN domain, they do not appear to share a common Olig1/2⁺ lineage-restricted progenitor. Early versions of this model under different names have been hypothesized previously(Alvarez-Buylla et al., 2001; Qian et al., 2000; Richardson et al., 2000). Such a sequential model bears resemblance to the one accepted for Drosophila CNS development, in which neuroblasts sequentially generate ganglion mother cells that produce two neurons, or one neuron and one glial cell (Pearson and Doe,2004). This model is also reminiscent of the competence model proposed for retinal development (Cepko et al., 1996), wherein multipotent retinal progenitors persist,undergoing progressive fate restriction and sequentially giving rise to all seven major cell types in the retina.

The basic assumption of the MNOP model(Lu et al., 2002; Noble et al., 2004; Rowitch et al., 2002) is that early (e.g. E10.0) Olig expression defines a population of common lineage-restricted progenitors for motoneurons and oligodendrocytes(Fig. 1B). In our Olig1-DTA ablation system, a clear prediction from the MNOP model is that Olig1-DTA should kill all MNOPs around the time of motoneuron generation, as a result of their expression of Olig genes. However, when almost all of the Olig⁺ progenitors for motoneurons were selectively killed in our ablation system, Olig2⁺ cells could still be detected at later time points; in particular, a normal number of Olig2⁺ cells were present in the mutant at E12.0.

Because the DTA-mediated cell ablation system appears to be extremely efficient, the normal number of Olig2⁺ cells present at E12.0 in Olig1-DTA embryos is not likely to be derived from escapers of earlier Olig⁺ precursors that survived the DTA-mediated cell killing. For example, virtually all of the motoneurons are killed by the Olig-DTA ablation system starting early in embryogenesis (E10.0) and continuing through the period of normal motoneuron generation. This suggests that, with respect to motoneuron precursors, this cell ablation system is very efficient. Also, from E12.0 onward, the killing of oligodendrocyte precursors by Olig1-DTA is also very efficient. Furthermore, we have tested this Cre-dependent conditional cell ablation system in a number of other embryological and post-embryological contexts, and in each case the system appears to be very efficient and specific in its elimination of the desired cell population (S.W. and M.R.C.,unpublished).

Therefore, the late-generated Olig2⁺ cells in Olig1-DTA embryos are likely to be generated from either the resident NSCs of the pMN domain, or from cells contributed by other domains through cell fate changes or migration induced by the Olig-DTA ablation system. For the following reasons, we propose that the former scenario is much more likely. (1) The generation and killing of motoneuron precursors is a continuous process in Olig1-DTA embryos, but the killing of early-born motoneuron precursors did not result in the compensatory generation of later-born motoneuron precursors through a fate change in other domains (Fig. 4). (2) The p2 and more dorsal domains did not show observable changes throughout development in Olig1-DTA mice, and Olig2⁺ cells did not co-label with markers that are specific for more dorsal domains, e.g. Irx3, or strong Pax6 expression (Fig. 5). (3) The number of p2 domain-derived Chx10⁺ V2 interneurons was normal in Olig1-DTA mice. It is difficult to imagine that the nearly normal numbers of Olig2⁺ cells in the pMN domain at E12.0 originate from the p2 domain, while at the same time a normal p2 domain and its derived V2 interneurons are maintained. (4) The Nkx2.2⁺ p3 domain was largely eliminated by Olig1-DTA-mediated killing(Fig. 5). If the early spinal cord were extremely plastic, the ablated p3 domain should be replaced by dorsal domains. However, the p3 domain is never regenerated in Olig1-DTA mice. Therefore, the normal numbers of Olig2⁺ cells in Olig1-DTA mice at E12.0 are not likely to arise as an artifact resulting from cell fate changes in neighboring domains, but rather are generated from the resident NSCs of the pMN domain. This idea does not support the MNOP model.

Furthermore, the Olig2⁺ cells present at E12.0 do not appear to be GRPs because, subsequently, the killing of these Olig2⁺ cells did result in the elimination of the majority of oligodendrocytes, but did not affect generation of the normal amounts of astrocytes. Therefore, it appears that strict interpretation of the GRP model is not consistent with our results, but variations of the GRP model are possible that may explain our data, and future experiments will address this. For example, the GRP model(Fig. 1A) could be rescued in vivo if GRPs are Olig^-, but oligodendrocyte precursors derived from GRPs are Olig⁺ and astrocyte precursors remain Olig^-. However, this rendition of the GRP model is equivalent, once given a developmental time parameter, to the sequential model as described herein.

An interesting question is whether our data merely supports a modified MNOP model in which MNOPs are Olig-negative. In the sequential model, if NSCs in the pMN domain, after generating motoneuron precursors and oligodendrocyte precursors, do not give rise to astrocytes, they would represent Olig-negative MNOPs (Fig. 1C). However, no cells have been isolated in vitro that generate solely motoneurons and oligodendrocytes. The MNOP model also suggests that NSCs from the p2 and more dorsal progenitor domains similarly give rise to lineage-restricted common progenitors (NA in Fig. 1B) for interneurons and astrocytes (Lu et al.,2002; Zhou and Anderson,2002). However, three recent papers have convincingly demonstrated that Olig⁺ OPCs also originate from dorsal progenitor domains in addition to the pMN domain (Cai et al.,2005; Fogarty et al.,2005; Vallstedt et al.,2005). These findings are also not consistent with the MNOP model because they suggest that NSCs in the developing spinal cord are likely to be tripotential in vivo, a prediction consistent with the sequential model. Our ablation system not only killed all Olig2⁺ OPCs derived from the pMN domain but also those from more dorsal domains(Fig. 6K,L). In fact, not one Pdgfrα⁺ OPC was detected in Olig1-DTA animals. Because astrocyte generation did not appear to be obviously affected in these animals,this suggests that most astrocytes are not generated through an Olig⁺ progenitor in vivo.

The NRP/GRP model was primarily established to explain in vitro studies that are best suited to revealing the developmental potential of cells,whereas the sequential model focuses more on cell fates in vivo. It is possible, however, from the sequential model that progenitor cells generated at a similar developmental time may share a similar potential, although they end up having different in vivo cell fates. For example, although motoneuron precursors and V2 interneuron precursors are normally generated from NSCs in different domains, they may have a very similar potential when analyzed in vitro in the presence of appropriate signals. In this sense, they could constitute NRPs and could replace each other in an in vivo transplantation experiment. As another example, after motoneuron generation, cells can be isolated as GRPs from both dorsal and ventral progenitor domains that can become oligodendrocytes and astrocytes in vitro(Gregori et al., 2002),indicating they share a very similar potential. As such, the sequential model can be used to reconcile differences among previous reported in vivo and in vitro observations.

It is noteworthy that Olig1-DTA mice share a common phenotype with the Olig1/2 gene knockout: an absence of motoneurons and oligodendrocytes in the spinal cord. This suggests that Olig genes are not merely good markers for motoneuron and oligodendrocyte precursor cells, but that they participate in the guidance of fate choices. In fact, the breeding of Rosa26-DTA176 with Hb9Cre (Yang et al., 2001),ShhCre (Harfe et al., 2004)and other Cre drivers also recapitulated many of the phenotypes associated with the corresponding gene knockout (S.W. and M.R.C., unpublished). However,there are significant differences in the phenotypes resulting from genetic cell ablation versus gene knockout studies. For example, Olig1-DTA mice exhibit an absence of the Nkx2.2⁺ p3 domain, whereas the Olig1/2 knockout mouse has normal development in the p3 domain(Lu et al., 2002; Zhou and Anderson, 2002). This phenotype difference suggests that although Olig genes are expressed in all cells of the p3 domain they are not required for its development. By comparison, Olig genes are required for the specification of committed progenitors in the pMN domain. As shown in this paper, the combined use of genetic cell ablation and knockout approaches can provide new insight into the lineage relationships among early neural progenitor cells.

We thank Dr Charles Stiles and Dr Richard Qing Lu for providing the Olig1-Cre mouse line; Dr Frank Costantini and Dr Philippe Soriano for providing the Rosa26-related plasmids and the Rosa26-eYFP mouse line; Dr Ian Maxwell for providing the DTA176 plasmid; and Drs Thomas Jessell, Samuel Pfaff, Charles Stiles and Hirohide Takebayashi for providing antibodies. We thank Anne Boulet, Lara Carroll, Dennis Chesire, Richard Dorsky, Gary Gaufo,Eric Green, Matt Hockin, Richard Qing Lu, Charlie Murtaugh and Monica Vetter for critical reading of the manuscript. M.R.C. is an Investigator of the Howard Hughes Medical Institute.