|Home | About | Journals | Submit | Contact Us | Français|
The spinal cord contains several distinct classes of neurons but it is still unclear how many of the functional characteristics of these cells are specified. One of the most crucial functional characteristics of a neuron is its neurotransmitter fate. In this paper, we show that in zebrafish most glycinergic and many GABAergic spinal interneurons express Pax2a, Pax2b and Pax8 and that these transcription factors are redundantly required for the neurotransmitter fates of many of these cells. We also demonstrate that the function of these Pax2/8 transcription factors is very specific: in embryos in which Pax2a, Pax2b and Pax8 are simultaneously knocked-down, many neurons lose their glycinergic and/or GABAergic characteristics, but they do not become glutamatergic or cholinergic and their soma morphologies and axon trajectories are unchanged. In mouse, Pax2 is required for correct specification of GABAergic interneurons in the dorsal horn, but it is not required for the neurotransmitter fates of other Pax2-expressing spinal neurons. Our results suggest that this is probably due to redundancy with Pax8 and that the function of Pax2/8 in specifying GABAergic and glycinergic neuronal fates is much broader than was previously appreciated and is highly conserved between different vertebrates.
The spinal cord is a crucial part of the vertebrate central nervous system as its neuronal circuitry controls movements and receives sensory inputs from the trunk and limbs. For this circuitry to function, different classes of neurons have to be correctly specified in the developing embryo. These cells form at specific dorsal/ventral locations in the spinal cord and express particular combinations of transcription factors (Lewis, 2006 and references therein). Currently, we know a lot about how these cells become molecularly-distinct from each other and express particular combinations of transcription factors (Lewis, 2006 and references therein) but we know a lot less about how the expression of particular transcription factors relates to the development of specific functional neuronal characteristics.
One crucial identifying characteristic of a neuron is the neurotransmitter(s) that it releases. How this is determined is still not very well understood and it is unclear whether similar or different mechanisms operate in distinct classes of neurons. Currently, specification of neurotransmitter fate is best understood in the mouse dorsal spinal cord, where Lbx1, Ptf1a, Lhx1 and Lhx5 transcription factors act upstream of Pax2 to specify GABAergic cells and Tlx1/3 antagonise the effects of Lbx1 to specify glutamatergic cells (Cheng et al., 2004, 2005; Glasgow et al., 2005; Pillai et al., 2007). In the mouse, Pax2 is also expressed in more ventral interneurons, but the neurotransmitter fates of these cells are unchanged in Pax2 knock-out mice (Cheng et al., 2004; Pillai et al., 2007), suggesting that Pax2 has distinct functions in different interneurons.
In the zebrafish, Danio rerio, there are two pax2 genes (pax2a and pax2b) and these both have a very similar spinal cord expression pattern to the highly-related pax gene, pax8 (Figs. 1A, B and E; Pfeffer et al., 1998). In this paper, we show that in zebrafish the vast majority of glycinergic spinal interneurons express Pax2/8 as do most GABAergic neurons within the spinal cord region where pax2/8 genes are expressed. When we knock-down just Pax2a, Pax2b or Pax8, there is very little effect on these glycinergic and GABAergic interneurons, but if we knock-down all three of these Pax2/8 proteins many interneurons lose their glycinergic and/or GABAergic fates, including the vast majority of CiAs, which are the most ventral population of Pax2/8 expressing spinal cells. We also establish that this function of Pax2/8 in spinal interneurons is very specific: cells lose their glycinergic and/or GABAergic fates in triple knock-down embryos, but they do not become glutamatergic or cholinergic. In addition, the soma morphozlogies and axon trajectories of CiAs are unchanged.
Our results suggest that the limited phenotype in the Pax2 mutant mice may be due to redundancy with Pax8 and consistent with this, in these mouse mutants Pax8 continues to be expressed in the ventral spinal cord, but it is lost from the Pax2-expressing cells that migrate into the dorsal horn (Pillai et al., 2007). This suggests that the function of Pax2/8 in specifying glycinergic and GABAergic neuronal fates is much broader than was previously appreciated and is highly conserved between different vertebrates.
Zebrafish (D. rerio) embryos were obtained from wild-type (AB, TL, or AB/TL hybrids) or Tg(pax2a:GFP) adults (Picker et al., 2002) or identified carriers heterozygous for noitu29a, a null allele of pax2a (Lun and Brand, 1998). Embryos were staged by hours post-fertilisation at 28.5 °C (h) and/or appropriate morphological criteria as in Kimmel et al. (1995). noi mutants were identified by their lack of a midbrain/hindbrain boundary (Brand et al., 1996; Lun and Brand, 1998).
In situ hybridisation was performed as in Concordet et al. (1996). For double in situ hybridisation, embryos were treated with Signal Enhancer (Invitrogen) before antibody incubation. Mouse-Anti-Dig (1/5000, Jackson ImmunoResearch) and Rabbit-Anti-Flu (1/2500, Roche) were detected with Invitrogen Tyramide kits #12 (488) and #5 (594). For in situ hybridisation and immunohistochemistry double stainings, Rabbit-Anti-GFP (1/1000) from Molecular Probes or Rabbit-Anti-Pax2 (1/300) from BabCO was revealed with Alexa-Fluor Goat-Anti-Rabbit 488 (1/1000) from Molecular Probes. RNA was detected with Invitrogen Tyramide kit #5. Double immunohistochemistry experiments to simultaneously detect Pax2 and GFP or Eng1b and GFP utilised Mouse-Anti-GFP (1/300) from Molecular Probes (catalogue # A11121). Eng1b antibody was kindly provided by A. Joyner (Skirball Institute, New York). This antibody (anti-Enhb-1; rabbit polyclonal) was generated against the mouse En2 homeodomain and detects both En1 and En2 in mammals (Davis et al., 1991) but only Eng1b in zebrafish (Higashijima et al., 2004b).
To determine neurotransmitter phenotypes we used in situ hybridisation for genes that encode proteins that transport or synthesise specific neurotransmitters. glyt2a and glyt2b (glycinergic markers) encode for glycine transporters necessary for glycine reuptake and transport across the plasma membrane; gad65, gad67a and gad67b (GABAergic markers) encode for a glutamic acid decarboxylase, necessary for the synthesis of GABA from glutamate and vglut2.1, vglut 2.2a and vglut 2.2b (glutamatergic markers) encode proteins responsible for transporting glutamate to the synapse. In all of these cases, a mix of equal concentrations of the relevant probes was used (Higashijima et al., 2004a,c). Choline acetyltransferase (chat) encodes for an enzyme that catalyzes the synthesis of acetylcholine (Yokogawa et al., 2007).
Other probes were pax2a, pax2b, pax8, pax5 (Pfeffer et al., 1998), p53 (Robu et al., 2007) and eng1b (1.3 KB encompassing the ORF, a kind gift from Drs. Kikuchi and Westerfield at the University of Oregon).
Photographs were taken using a Zeiss Axio Imager M1 (DIC images) or a Leica TS SP2 confocal (fluorescent images) microscope and processed using Adobe Photoshop. All fluorescent images, with the exception of Figs. 2H–J and Figs. 4A'– E' and I' are projections of multiple optical sections performed in Image J (Abramoff et al., 2004).
Morpholino antisense oligonucleotides (MOs) were injected into 1–2 cell embryos from a cross of identified carriers heterozygous for noitu29a. Therefore, ~ 25% of injected embryos lacked Pax2a. All of these morpholinos have been used successfully in previous studies. The pax2b MO blocks translation and is GGTCTGCCTTACAGTGAATATCCAT (Bricaud and Collazo, 2006; Mackereth et al., 2005; Millimaki et al., 2007); the pax8 MOs block splicing and a combination of E5/15(TTTCTGCACTCACTGTCATCGTGTC) and E9/19(ACCGGCGGCAGCTCACCTGATACCA) (Hans et al., 2004) were used.
MOs were initially injected at concentrations of 1 mg/ml, 1.5 mg/ml and 2 mg/ml each. At 1.5 mg/ml and 2 mg/ml the phenotype was more severe than at 1 mg/ml (Supp. Data Figs. 4B and B'). However, at 2 mg/ml some embryos were morphologically disturbed (they had twisted axis and wavy notochords). Therefore, in all of the experiments presented in this paper, with the exception of Supplementary Data Fig. 1B, MOs were injected at 1.5 mg/ml each. At this concentration, triple-knock-down embryos (noi mutants injected with pax2b and pax8 MOs) completely lost expression of pax8 RNA and Pax2 protein (Figs. 1F and 3I), suggesting that Pax2a, Pax2b and Pax8 were fully knocked-down. To further confirm that the pax8 splice-blocking MOs were working we examined pax8 expression in embryos injected with a lower concentration of morpholinos (1 mg/ml). In this case, pax8 RNA was localised in the nucleus, indicating that splicing of pax8 RNA was blocked (Yan et al., 2002; cf.Supp. Data Figs. 1A and B).
To rule out the possibility that the phenotype observed in morpholino injected embryos was due to either specific or non-specific toxicity or cell death we examined p53 expression in our injected embryos at 36 h. Non-specific cell death due to MO toxicity usually causes an upregulation of p53 expression (Robu et al., 2007). However, there was no expression of p53 in our injected embryos (36 h; n = 60; Supp. Data Fig. 1H). In addition, we demonstrated that pax2/8-expressing cells still form in morpholino injected embryos using an in situ hybridisation for pax2a and pax2b (Figs. 1D and K). We saw no significant changes in the number of cells expressing these genes in triple knock-down embryos compared to wild-type embryos. Therefore, the cells still form and the reduced number of glycinergic and GABAergic cells in triple knock-down embryos is not due to a reduction in the number of pax2/8-expressing cells. Finally, we also showed that CiAs, which are a specific subset of pax2/8-expressing cells, not only form in normal numbers in triple knock-down embryos, but their gross morphology (cell soma size, shape and axon trajectory) is also unchanged even though the vast majority of these cells are no longer glycinergic or GABAergic (Figs. 4J–M and Supp. Data Table 3). Taken together, these observations suggest that the phenotype from knocking-down Pax2/8 is very specific and is not due to toxicity or to the cells dying.
To further confirm that our results were not caused by non-specific effects of the morpholino injections we also injected two different morpholinos at the same total concentration (4.5 mg/ml) as our Pax2/8 triple knock-down experiments. These morpholinos were an evx1 translation blocking morpholino: CCTTTCCGTGCTTCGGCGAGCCCAT and a lis1 control morpholino: CTgGTaGCCTCTGTGACAGgACgAT (small case letters show mismatches) which was a kind gift from Richard Adams. We injected these morpholinos into both wild-type embryos and into noi (pax2a) mutant embryos. In all of these experiments there was no change in the number of glycinergic and GABAergic spinal neurons (Supp. Data Fig. 7).
In all of our experiments, we observed a continuum of results, presumably due to slight variations in the amount of knock-down achieved from the MOs in different embryos and in individual cells within injected embryos. Very occasional morphologically disturbed embryos were excluded from further analysis. Each experiment was repeated 3 times (~ 60 embryos injected each time). Results are shown for the 4 most severely effected embryos from each experiment (12 embryos in total).
In all cases, cell counts are for both sides of a 5-somite length of spinal cord adjacent to somites 6–10. Cell row numbers are assigned ventral to dorsal (e.g. cells directly above the notochord are in row 1, see Supp. Data Fig. 1F). The Pax2-expressing spinal cord domain is defined as rows 4–7 based on results shown in Fig. 1. Results were analysed using the students' T test. Statistically significant results where p < 0.05 are indicated with a star in the figures and individual p values are provided in Supp. Data Table 2. Error bars in figures indicate the standard deviation. Results in the text are shown as the mean ± standard deviation. For all of the knock-down experiments (morpholino injections and analysis of noi mutants) both the experimental and control results are an average of 12 embryos. For analyses that only include wild-types (e.g. determining the neurotransmitter phenotypes of Pax2-expressing cells in wild-type embryos) results are an average of 5 different embryos.
In 24 h zebrafish embryos, pax2a, pax2b and pax8 all have very similar expression patterns, suggesting that they are co-expressed by several distinct interneuron populations that form initially in the intermediate region of the spinal cord (Figs. 1A–C, E, I and J; Pfeffer et al., 1998). Consistent with this, similar numbers of cells are labelled by in situ hybridisation in wild-type embryos when pax2a, pax2b or pax8 probes are hybridised either singly or together (Figs. 1A–C, E and I) and 2-colour double in situ hybridisation for pax2a and pax2b shows that these two genes are indeed expressed by the same spinal cord cells (Fig. 1J). Pax2 and Pax8 are part of a subfamily of Pax transcription factors that also includes Pax5 (Bouchard et al., 2000, 2002; Hans et al., 2004; Holland et al., 2007; Pfeffer et al., 1998; Wada et al., 1998; Walther et al., 1991). However, pax5 is not expressed in the zebrafish spinal cord (Fig. 1H; http://www.zfin.org; Pfeffer et al., 1998).
To determine the neurotransmitter phenotypes of zebrafish pax2-expressing interneurons we combined immunohistochemistry for Pax2 and in situ hybridisation for markers specific for glycinergic, GABAergic and glutamatergic cells (see Materials and methods). At 24 h, about 90% of Pax2-expressing spinal cells are either glycinergic and/or GABAergic, with ~ 60% being glycinergic and a similar percentage GABAergic (Figs. 2A, B, D and E; 89.2% ± 4.7 express either GABAergic or glycinergic markers, 57.6% ± 6.5 are glycinergic and 57.1% ± 3.6 are GABAergic; see also Supp. Data. Fig. 2). This suggests that about a third of Pax2-expressing spinal cells express both GABAergic and glycinergic markers at this stage. Intriguingly a small number of Pax2-expressing spinal cells express glutamatergic markers (Figs. 2C and E; 16.4% ± 1.6). These cells are predominantly in the most dorsal row of Pax2 expression where they appear to correspond to the larger cells that are the first cells in the spinal cord to express Pax2 (Mikkola et al., 1992).
Pax2-expressing interneurons account for ~ 80% of the glycinergic and ~ 34% of the GABAergic interneurons in the zebrafish spinal cord (Figs. 2A, B and F; 80.4% ± 6.6 and 33.8% ± 2.9 respectively). However, most of the GABAergic cells that do not express Pax2 are found in the ventral spinal cord outside the Pax2-expression domain (Figs. 2B and cf 2F and 2G). Within the intermediate spinal cord region where Pax2 is expressed (rows 4–7), ~ 75% of GABAergic cells express Pax2 (Fig. 2G; 74.8% ± 11.7). This shows that not only are the vast majority of Pax2-expressing spinal cells glycinergic and/or GABAergic, but in the spinal cord region where Pax2 is expressed, most glycinergic and GABAergic neurons express Pax2.
To determine whether Pax2 is required for glycinergic and/or GABAergic fates in the spinal cord, we first examined zebrafish embryos that lack Pax2a function (no isthmus (noi) mutants) at 24 h (Brand et al., 1996; Lun and Brand, 1998). In all cases, loss of Pax2a had no effect on the number of cells with particular neurotransmitter phenotypes (Figs. 3M and N; Supp. Data Fig. 3 and Supp. Data Tables 1 and 2). To test whether this is due to functional redundancy between Pax2a and Pax2b, we knocked-down Pax2b function in noi mutants using a pax2b morpholino (MO). At 24 h, we observed a statistically significant reduction in the number of glycinergic cells (Fig. 3M and Supp. Data Tables 1 and 2) showing that Pax2 function is required for the glycinergic fates of some spinal cord cells. However, there was no significant change in the number of GABAergic cells (Fig. 3N and Supp. Data Tables 1 and 2) and the reduction in the number of glycinergic cells was far less than the number of glycinergic interneurons that express Pax2 (cf. Figs. 3M and 2G).
Therefore, we tested whether there was functional redundancy between Pax2 and Pax8 by injecting MOs against pax2b and pax8 into noi mutants. These triple knock-down embryos lost expression of pax8 RNA (Fig. 1F) and Pax2 protein (Fig. 3I), suggesting that Pax2a, Pax2b and Pax8 were all fully knocked-down. We also confirmed that pax5 expression was not activated in the spinal cord of these triple knock-down embryos (Fig. 1L). Interestingly, our results show that pax8 spinal cord expression is regulated by Pax2 and may also be regulated by Pax8 (pax8 expression is dramatically downregulated in the absence of either Pax2 or Pax8 function and it is completely lost in triple knock-down embryos; see Fig. 1F, Supp. Data Figs. 1C–E and discussion in Supp. Data Fig. 1 legend). However, in contrast pax2 expression does not require Pax2/8 function (Figs. 1D and K).
In these 24 h triple knock-down embryos, we observed a dramatic decrease in the number of glycinergic and GABAergic spinal cord cells in the intermediate region of the dorso-ventral axis where Pax2 is normally expressed (Figs. 3D, E and M–P). In contrast, wild-type embryos injected with pax8 MOs had only minor reductions in the number of glycinergic and GABAergic cells (Supp. Data Fig. 3 and Supp. Data Tables 1 and 2), supporting the hypothesis that pax2a, pax2b and pax8 function redundantly in specifying glycinergic and GABAergic fates.
We confirmed that our morpholino injections were not causing either specific or non-specific cell death by demonstrating that the same number of cells express pax2 RNA in wild-type and triple knock-down embryos (Figs. 1C, D and K), that there was no upregulation of p53 (Robu et al., 2007) in triple knock-down embryos (Supp. Data Fig. 1H) and that injection of two different control morpholinos into wild-type or noi (pax2a) mutants had no effect on glycinergic or GABAergic spinal neurons (Supp. Data Fig. 7; see also the longer discussion of morpholino control experiments in Materials and methods). In addition, we demonstrated that the reduction in the number of glycinergic and GABAergic cells is not caused by cells changing to a glutamatergic or acetylcholinergic fate as wild-type and triple knock-down embryos have the same number of glutamatergic and acetylcholinergic spinal cord cells (Figs. 3C, F, J–L and Q).
Finally, to confirm that glycinergic and GABAergic fates are not just delayed in triple knock-down embryos, we also examined wild-type and triple knock-down embryos at 36 h and 48 h. In both cases, we still observed a significant reduction in the number of glycinergic and GABAergic neurons in triple knock-down embryos compared to stage-matched wild-type controls (Supp. Data Figs. 4A and A' and Supp. Data Fig. 5).
As our results are consistent with those previously reported by other researchers in the dorsal horn of the mouse Pax2 knock-out (Cheng et al., 2004; Pillai et al., 2007), it seems very unlikely that the phenotypes we observe in triple knock-down embryos are due to off-target effects from the morpholinos (i.e. to one of the morpholinos knocking-down a different gene, that is required for glycinergic or GABAergic fates). However, to further confirm this, we examined all of the different combinations of Pax2/8 knock-down (Supp. Data Fig. 3 and Supp. Data Table 1 and Table 2). Two observations argue strongly that loss of glycinergic and GABAergic spinal fates is specifically caused by knocking-down Pax2/8 function. Firstly, we see a continuum of phenotypes depending on how many Pax2/8 proteins we knock-down, regardless of whether we use the pax2b MO, the pax8 MOs or the noi (pax2a) mutants (Supp. Data Fig. 3 and Supp. Data Table 1 and Table 2). Secondly, we observe a statistically significant difference between wild-type embryos injected with pax8 and pax2b morpholinos and noi mutants injected with pax8 and pax2b morpholinos (the triple knock-downs are always more severely affected than the double knock-downs, p = 0.005 in the case of glycinergic cells and p = 0.0045 in the case of GABAergic cells; see also Supp. Data Fig. 3 and Supp. Data Table 1 and 2). If our results were due to off-target effects from the morpholinos then this would not be the case. We would instead expect these two results to be very similar to each other.
Taken together, our results demonstrate that Pax2/8 are redundantly required for the glycinergic and/or GABAergic fates of many zebrafish spinal cord interneurons, but that the absence of Pax2/8 function (and of a glycinergic and/or GABAergic fate) is not sufficient for cells to become glutamatergic or cholinergic.
In mouse, Pax2 is required for the neurotransmitter fates of dorsal horn GABAergic neurons but it is not required for the neurotransmitter fates of more ventral Pax2-expressing interneurons (Cheng et al., 2004; Pillai et al., 2007). In contrast, our results suggest that Pax2/8 are required for GABAergic and glutamatergic fates throughout the dorsal–ventral extent of the Pax2-expression domain. To further confirm this, we specifically examined the most ventral population of Pax2-expressing spinal cells. In amniotes, these are V1 cells. In zebrafish, Circumferential Ascending interneurons (CiAs) (Bernhardt et al., 1990) are thought to be homologous to V1 cells (Higashijima et al., 2004b; Sapir et al., 2004). Both V1 cells and CiAs are the only spinal cells to express the transcription factor Eng1b (in the case of zebrafish) or En1 (in the case of amniotes) (Higashijima et al., 2004b; Sapir et al., 2004). In addition, both of these cell types share morphological and functional characteristics. For example, both CiAs and V1 cells are inhibitory, they have ipsilateral ascending axons and they are involved in regulating fast locomotion movements (Gosgnach et al., 2006; Higashijima et al., 2004b; Li et al., 2004). However, before this study, it had not been determined whether CiAs also, like V1 cells, express Pax2.
Using in situ hybridisation for eng1b and immunohistochemistry for Pax2 we demonstrated that at 24 h CiAs express Pax2 and that they are indeed the most ventral spinal cord cells to do so (Fig. 4A; see also Supp. Data Fig. 6A). Consistent with this, we also showed that 24 h embryos from a transgenic line where GFP is regulated by a partial pax2 promoter (Picker et al., 2002), express GFP in a subset of spinal cord Pax2-expressing cells, the majority of which are CiAs (Figs. 4F and I–M).
Determining that CiAs express Pax2 provided us with the opportunity not only to investigate the effects of Pax2/8 knock-down on the most ventral population of Pax2/8-expressing spinal neurons, but also to examine the morphology and neurotransmitter phenotypes of a single identified class of neurons. Our more global analysis of triple knock-down embryos identified a dramatic decrease in the number of glycinergic and GABAergic cells in the intermediate region of the spinal cord where Pax2/8 are normally expressed (Figs. 3D, E and M–P). However, some glycinergic and GABAergic cells still remain in these triple knock-down embryos and it was unclear whether this was due to incomplete penetrance of the phenotype or due to Pax2/8 only being required for the glycinergic and/or GABAergic fates of specific subsets of Pax2/8-expressing neurons. Determining the phenotype of CiAs in triple knock-down embryos should enable us to distinguish between these two possibilities. In the former case, we would expect some CiAs to maintain their glycinergic and/or GABAergic fates in triple knock-down embryos, whereas in the latter case we would expect all CiAs to have the same phenotype (either loss of glycinergic and GABAergic fates or no effect).
At 24 h, ~ 90% of CiAs are glycinergic and just over 40% are GABAergic, suggesting that several CiAs express both of these neurotransmitters at this stage (Figs. 4B, D and N; 89.52% ± 6.87 of CiAs are glycinergic and 42.59% ± 5.04 are GABAergic; see also Supp. Data Fig. 6 and Higashijima et al., 2004b). Consistent with our analyses of whole spinal cords, in triple knock-down embryos the number of CiAs (eng1b-expressing cells) is not altered, but the number of glycinergic and GABAergic CiAs is reduced (Figs. 4C, E and N; in triple knock-down embryos only 23.52% ± 5.43 of CiAs are glycinergic and 27.69% ± 6.86 are GABAergic; see also Supp. Data Fig. 6). However, no CiAs are glutamatergic, in either wild-type or triple knock-down embryos (Figs. 4G and H) and the general size and shape of CiA somata and CiA axon lengths and trajectories are indistinguishable in wild-type and triple knock-down embryos (Figs. 4J–M and Supp. Data Table 3). This suggests that the loss of glycinergic and GABAergic fates is a very specific phenotype and that in other respects these cells develop normally (at least at these early stages).
These results also suggest that the incomplete penetrance of the Pax2/8 knock-down phenotype is not due to specific Pax2/8-expressing populations being resistant to loss of Pax2/8 function. CiAs are thought to constitute a single class of neurons (Higashijima et al., 2004b), but in triple knock-down embryos some CiAs still maintain their glycinergic and GABAergic fates.
In this paper, we provide the first systematic analysis of the neurotransmitter fates of all Pax2-expressing spinal interneurons. We show that in zebrafish embryos, the vast majority of Pax2-expressing interneurons are glycinergic or GABAergic and these cells account for ~ 60% of all glycinergic and GABAergic spinal interneurons and ~ 86% of glycinergic and GABAergic interneurons within the spinal cord regions where Pax2 is expressed. Studies of specific subsets of Pax2-expressing spinal cells in mouse have demonstrated that many of these neurons are also GABAergic and/or glycinergic (Cheng et al., 2004; Lewis, 2006 and references therein; Pillai et al., 2007; Sapir et al., 2004), suggesting that this correlation of Pax2 expression and glycinergic and GABAergic fates is highly conserved among vertebrates.
We also show that Pax2a, Pax2b and Pax8 act in a functionally redundant manner to specify the glycinergic and GABAergic fates of many Pax2/8-expressing spinal interneurons. When we knock-down Pax2a, Pax2b and Pax8 many interneurons lose their glycinergic and GABAergic fates, including the majority of CiAs, which are the most ventral population of Pax2/8-expressing spinal cells. We also establish that this function of Pax2/8 in spinal interneurons is very specific: loss of Pax2/8 function does not cause cells to change their neurotransmitter phenotype and become excitatory (glutamatergic or cholinergic); they are just no longer glycinergic or GABAergic. In addition, at least in the case of CiAs, their soma shapes and sizes and axon trajectories are unchanged. It is still formally possible that these neurotransmitter phenotypes are an indirect effect caused by a lack of synapse formation and/or synaptic activity in these neurons. However, given that we observe a dramatic neurotransmitter phenotype as early as 24 h we consider that this is unlikely.
Our results suggest that the lack of a neurotransmitter phenotype in ventral Pax2-expressing cells in the mouse Pax2 knock-out, is at least partly due to redundancy between Pax2 and Pax8. Consistent with this, in Pax2 mutant mice Pax8 continues to be expressed in the ventral spinal cord, but it is lost from the Pax2-expressing cells that migrate into the dorsal horn (Pillai et al., 2007). This suggests that Pax2/8 have a major and crucial function in specifying glycinergic and GABAergic fates of multiple spinal cord interneurons in both the simple anamniote and the more complex mammalian spinal cord.
As mentioned earlier, Pax2 and Pax8 are part of a subfamily of Pax transcription factors that also includes Pax5 (Bouchard et al., 2000, 2002; Hans et al., 2004; Holland et al., 2007; Pfeffer et al., 1998; Wada et al., 1998; Walther et al., 1991). Unlike in zebrafish, in mouse Pax5 is expressed in the spinal cord (Pillai et al., 2007), raising the possibility that it may also function redundantly with Pax2 and Pax8 in mammalian spinal cord.
While knock-down of Pax2a, Pax2b and Pax8 in zebrafish embryos results in substantial and statistically significant reductions in the number of spinal interneurons with glycinergic and/or GABAergic fates, several glycinergic and/or GABAergic spinal interneurons remain in these triple knock-down embryos. Many of the remaining GABAergic neurons are located in the very ventral spinal cord (rows 1–3; Fig. 3E) outside the Pax2/8 expression domain (rows 4–7: Fig. 1I) and, hence, these GABAergic neurons must be specified by a different mechanism. However, in addition, a minority of Pax2/8-expressing cells retain their glycinergic and/or GABAergic fates in triple knock-down embryos. Interestingly, this is the case even for CiAs, which are thought to constitute a single class of neurons (Higashijima et al., 2004b). Furthermore, in triple knock-down embryos the number of glycinergic cells is reduced much more dramatically than the number of GABAergic cells. This might suggest that Pax2/8 have a more pronounced role in specifying glycinergic neurons than GABAergic neurons. Alternatively Pax2/8 might be required for maintenance of glycinergic and GABAergic fates rather than their initial specification, as CiAs (and potentially other zebrafish spinal neurons) change from a GABAergic to a glycinergic fate during their development (Higashijima et al., 2004b). However, in this case we would expect there to be a more severe reduction of glycinergic and/or GABAergic neurons in triple knock-down embryos at later stages, but this is not what we observe (Supp. Data Figs. 4A and A' and Supp. Data Fig. 5).
One possible explanation for the phenotype not being completely penetrant might be an incomplete knock-down of Pax2/8 function. However, in this case any remaining Pax2/8 expression must be very weak as it is not detected by Pax2 immunohistochemistry or pax8 in situ hybridisation (Figs. 1F and 3I). Therefore, we think that it is more likely that, while Pax2/8 are major players in specifying glycinergic and GABAergic spinal fates, there are other factors that can compensate for the loss of Pax2/8 in some cells. For example, it is possible that Pax2/8 may only be required for glycinergic and/or GABAergic neurotransmitter expression in as-yet-unidentified distinct subsets of CiAs and other Pax2/8-expressing cells. Alternatively, induction of glycinergic and GABAergic fates may be a strongly buffered mechanism where other transcription factors can sometimes, in a stochastic manner, substitute for loss of Pax2/8. If this is the case, then it is not yet clear what these additional, as-yet-unidentified, transcription factors might be. In the mouse dorsal horn, Lbx1, Ptf1a, Lhx1 and Lhx5 transcription factors are also required for correct specification of GABAergic fates. However, all of these proteins act upstream of Pax2 and control neurotransmitter fates by regulating pax2 expression (Cheng et al., 2004, 2005; Glasgow et al., 2005; Pillai et al., 2007) so they are unlikely to compensate for loss of Pax2/8.
Finally, another possible explanation for some Pax2/8-expressing glycinergic and GABAergic neurons maintaining their neurotransmitter fates is suggested by the observation that spontaneous neuronal activity can homeostatically bias the specification of excitatory versus inhibitory spinal fates. Experiments in frogs have shown that spinal cord neurons with particular neurotransmitter fates share specific patterns of calcium spiking during their early development. If these calcium spikes are blocked by genetic or pharamacological agents then the number of inhibitory neurons decreases and the number of excitatory neurons increases, whereas the opposite phenotype is observed if calcium activity levels are increased (Borodinsky et al., 2004; Spitzer et al., 2004). Interestingly, the genetic identities of these cells are unchanged, suggesting that activity levels can, in some instances, over-ride other developmental cues that specify neurotransmitter fates. Spontaneous intracellular calcium signals have also been observed during early development of zebrafish spinal cord neurons (Ashworth and Bolsover, 2002). Therefore, the incomplete loss of glycinergic and GABAergic fates in neurons that normally express Pax2/8 in triple knock-down embryos could be due to some glycinergic and GABAergic fates being maintained by this activity-based mechanism. However, it is not yet clear whether this mechanism acts upstream, downstream or in parallel to transcription factor specification of neurotransmitter fates. In addition, in these experiments neurons either switched their fates (from inhibitory to excitatory or vice versa) or, in a small number of cases, they co-expressed excitatory and inhibitory neurotransmitters (Borodinsky et al., 2004), whereas in Pax2/8 triple knock-down experiments many neurons lose their glycinergic and GABAergic fates, but they do not acquire glutamatergic or cholinergic fates.
While about 90% of Pax2/8-expressing spinal interneurons are either glycinergic or GABAergic at 24 h, only ~ 60% are glycinergic and ~ 60% are GABAergic. In addition, a minority of Pax2/8-expressing cells in both zebrafish and mouse dorsal spinal cords express glutamatergic markers (Figs. 2C and E; Cheng et al., 2004). This suggests that while Pax2/8 are required for glycinergic and GABAergic fates in many cells, expression of these transcription factors is not sufficient to specify one or other of these neurotransmitter fates. This is consistent with data reported in chick, where ectopic expression of Pax2 in the neural tube did not induce GABA expression (Cheng et al., 2004).
It is not clear from our results why some Pax2/8-expressing cells are glycinergic, some are GABAergic and some express both of these neurotransmitters. It is possible that at least some of the GABAergic cells or the cells expressing both GABA and glycine will later become glycinergic as has been suggested for CiAs (Higashijima et al., 2004b). Therefore, at least some of these differences in neurotransmitter fates may reflect a temporal progression within particular neurons from (GABAergic) to (GABAergic and glycinergic) to (glycinergic). Consistent with this, the number of glycinergic neurons more than doubles between 24 h and 48 h whereas the number of GABAergic neurons stays pretty much constant (Supp. Data. Figs. 4A and A'). However, it is also possible that in at least some cases, these different neurotransmitter fates reflect more profound differences between cells. In this case, these differences may result from differential expression of other transcription factors or different developmental histories.
The simplest explanation for the Pax2/8-expressing glutamatergic cells would be that there is another transcription factor expressed in these cells that inhibits the function of Pax2/8 and instead specifies a glutamatergic phenotype. There is a precedence for this type of mechanism, in that the Lbx1 transcription factor normally specifies GABAergic interneurons in the dorsal spinal cord, but in a subset of Lbx1-expressing cells, Tlx3 inhibits the function of Lbx1 and induces a glutamatergic phenotype (Cheng et al., 2005).
One of the potentially surprising aspects of our results is the specificity of the phenotype that we observe in the absence of Pax2/8 function. While we cannot rule out that additional phenotypes develop at later stages in the neurons that normally express Pax2/8 in triple knock-down embryos (and it would be hard to test this using morpholinos as their efficacy decreases with increasing age of the embryos and dilution of the morpholinos), the only phenotype that we have identified so far in these neurons is a loss of glycinergic and GABAergic fates. This might initially be surprising, given that, as discussed above, neuronal activity can homeostatically adjust neurotransmitter fates in the spinal cord (Borodinsky et al., 2004) and given a recent report that knocking-down glycine receptors reduces the number of neurons (including Pax2-expressing neurons) in the zebrafish spinal cord (McDearmid et al., 2006). However, with respect to the former study, the specification of neurotransmitter fates by calcium activity is thought to be cell autonomous (so defects in a subset of spinal neurons shouldn't affect other neurons) and in the case of the latter study, decreases in neuronal numbers were only observed at later stages of development. While neuronal morphology was not examined in the mouse Pax2 mutant dorsal horn, these studies also observed that the cells that lost GABAergic fates did not become glutamatergic, suggesting that the specificity of the Pax2/8 phenotype may be highly conserved between different vertebrates.
In zebrafish, the only Pax2/8-expressing interneurons that have been identified morphologically or functionally are CiAs (this report). These are genetically and functionally homologous to mammalian V1 cells (Alvarez et al., 2005; Higashijima et al., 2004b; Sapir et al., 2004; Saueressig et al., 1999). However, comparisons with amniotes suggest that more dorsal Pax2/8-expressing interneurons in the zebrafish spinal cord are likely to be functionally equivalent to amniote V0, dI6, dI4 and/or DILA interneurons (Lewis, 2006 and references therein). Of these neurons, DILA neurons migrate into the superficial lamina of the dorsal horn where they are probably involved in nociception, dI4 cells are thought to migrate into deeper layers of the dorsal horn and V1, V0 and dI6 interneurons form part of the locomotion central pattern generator in the ventral spinal cord. At least a subset of V0 interneurons are essential in mice for correct walking movements as they control the alternating left–right activity of the motor neurons that innervate hindlimb muscles (Lanuza et al., 2004). In contrast, V1 cells are required for fast locomotion in mouse: if these cells are genetically ablated or temporarily inactivated, mice are unable to walk at fast speeds but they can still walk at slower speeds (Gosgnach et al., 2006). This is also consistent with data from tadpoles, which shows that aINs (which are genetically homologous to both CiAs and V1 cells) provide early cycle inhibition to central pattern generator neurons during swimming, particularly at faster swimming frequencies (Li et al., 2004). Based on these results in other vertebrates, we would predict that altering the neurotransmitter phenotypes of Pax2/8-expressing cells might have functional consequences both for sensory processing and correct locomotion.
Unfortunately, this is not easy to assess in our experiments, as we can't distinguish between behavioural defects due to the spinal cord phenotype that we describe in this paper and behavioural defects due to previously described brain phenotypes (for example even noi single mutant embryos lack a midbrain–hindbrain boundary";; Brand et al., 1996; Lun and Brand, 1998). However, we do observe considerable locomotion defects in triple knock-down embryos. For example, triple knock-down embryos only move when they are poked (even at stages where wild-type embryos undergo spontaneous fast swimming movements), a significant percentage of triple knock-down embryos don't move at all (5/20 at 24 h, 3/20 at 48 h) and an even larger percentage swim in circles (8/20 at 48 h), or vibrate on the spot (6/20 at 48 h) when poked. All of this is at least consistent with the idea that locomotion control is perturbed in these embryos.
Taken together, all of these results suggest that Pax2/8 transcription factors have crucial but redundant functions in specifying the glycinergic and GABAergic fates of multiple spinal interneurons in both the simple anamniote and the more complex mammalian spinal cord. Comparisons between zebrafish, Xenopus and mouse have suggested that mechanisms of spinal cord patterning and resulting neuronal circuitry might be highly conserved in vertebrates, with distinct functional classes of interneurons being specified by different combinations of post-mitotically expressed transcription factors (Goulding and Pfaff, 2005; Lewis, 2006). However, to our knowledge, our study is the first that has demonstrated that some of these post-mitotically expressed transcription factors, do indeed have similar functions in anamniote and mammalian spinal cord interneurons.
This work was supported by a Wellcome Trust project grant awarded to KEL (Ref 079971), Portuguese Foundation for Science and Technology PhD funding to MFB and a Royal Society University Research Fellowship to KEL. MFB is a student of the Gulbenkian PhD Program in Biomedicine, Portugal. We are grateful to Rob Shaw for help with some initial analysis of the Tg(pax2a:GFP) line, Claus Schule for some of the Evx1 control morpholino injections, Alex Joyner for her kind gift of anti-Enhb-1 antibody, Stephan Hans and Monte Westerfield for their kind gift of pax2b and pax8 morpholinos and to Bill Harris, Roger Keynes, Murray Hargrave, Giuseppe Lupo, David Rivers and Naomi Stevens for their helpful comments on previous drafts of this manuscript.
Appendix ASupplementary data associated with this article can be found, in the online version, at doi:10.1016/j.ydbio.2008.08.009.
Number of cells expressing particular neurotransmitters in the Pax2/8 expression domain (rows 4–7 of the spinal cord) in all of the different Pax2 and Pax8 knock-down experiments. In all cases, cell counts are for a 5 somite length of the spinal cord adjacent to somites 6–10. All values are an average from 12 different embryos and are shown as the mean + standard deviation. ND = not done.
P values for all of the pair wise comparisons between the numbers of glycinergic or GABAergic cells in the different Pax2 and Pax8 knock-down experiments, calculated using the student T test. Results from experiments examining the number of glycinergic cells are in the left hand corner. Statistically significant results are shown in green (p < 0.05) and non-significant results are shown in blue. Results from experiments examining the number of GABAergic cells are in the right hand corner. Statistically significant results are shown in purple (p < 0.05) and non-significant results are shown in red. For example, the first box in the left hand column gives the p value for the hypothesis that noi mutants have the same number of glycinergic cells as wild-type embryos. As the p value < 0.05, the difference between the number of glycinergic cells in noi mutants and wild-type embryos is not statistically significant. In contrast, the bottom box in the left hand column gives the p value for the hypothesis that triple knock-down embryos have the same number of glycinergic cells as wild-type embryos. In this case, the p value for the comparison (and hence the difference between the number of glycinergic cells in triple knock-down embryos and wild-type embryos) is statistically significant.
CiA axon length and soma size in wild-type and triple knock-down embryos at 24 h, measured using Zeiss Axiovision software. All of the measured CiAs were at the same rostral–caudal level (adjacent to somites 7 and 8). Values shown are averages + standard deviation. In the case of the WT results, the values are an average of 22 cells and in the case of the TKD results the values are an average of 19 cells. p values are for the student T test. Neither of the differences between WT and TKD CiAs is statistically significant.