Tlx3 and Tlx1 are post-mitotic selector genes ... - Semantic Scholar

© 2004 Nature Publishing Group http://www.nature.com/natureneuroscience

ARTICLES

Tlx3 and Tlx1 are post-mitotic selector genes determining glutamatergic over GABAergic cell fates Leping Cheng1,2,9, Akiko Arata3,9, Rumiko Mizuguchi4,9, Ying Qian1,2, Asanka Karunaratne4, Paul A Gray1,2, Satoru Arata5, Senji Shirasawa6, Maxime Bouchard7, Ping Luo1,2, Chih-Li Chen1,2, Meinrad Busslinger7, Martyn Goulding4, Hiroshi Onimaru8 & Qiufu Ma1,2 Glutamatergic and GABAergic neurons mediate much of the excitatory and inhibitory neurotransmission, respectively, in the vertebrate nervous system. The process by which developing neurons select between these two cell fates is poorly understood. Here we show that the homeobox genes Tlx3 and Tlx1 determine excitatory over inhibitory cell fates in the mouse dorsal spinal cord. First, we found that Tlx3 was required for specification of, and expressed in, glutamatergic neurons. Both generic and region-specific glutamatergic markers, including VGLUT2 and the AMPA receptor Gria2, were absent in Tlx mutant dorsal horn. Second, spinal GABAergic markers were derepressed in Tlx mutants, including Pax2 that is necessary for GABAergic differentiation, Gad1/2 and Viaat that regulate GABA synthesis and transport, and the kainate receptors Grik2/3. Third, ectopic expression of Tlx3 was sufficient to suppress GABAergic differentiation and induce formation of glutamatergic neurons. Finally, excess GABA-mediated inhibition caused dysfunction of central respiratory circuits in Tlx3 mutant mice.

Glutamate and GABA (gamma-aminobutyrate) are the predominant neurotransmitters for excitatory and inhibitory neurons, respectively, in vertebrate brain1. These two neurotransmitter phenotypes are typically expressed in a mutually exclusive manner2,3, thereby defining the major functional subdivision in neuronal cell type. In the dorsal horn of the spinal cord, the major relay center for processing somatosensory information, most ascending projection neurons and a subset of local circuit interneurons are excitatory and use glutamate as their transmitter4–6 (Fig. 1a). These neurons are modulated by local inhibitory neurons, many of which are GABAergic (Fig. 1a)6–10. Very little is known about how neurons make the choice between excitatory and inhibitory cell fates. In the developing forebrain, two mechanisms account for the mutually exclusive acquisition of excitatory or inhibitory cell fate: (i) glutamatergic and GABAergic neurons develop from distinct pools of neural precursors and (ii) there is a genetic program that suppresses GABAergic differentiation in cortical glutamatergic cells11–13. To date, the transcription factors that control glutamatergic differentiation have not been identified. Nor is it known if there are any transcription factors that function as selectorlike genes14 to concomitantly promote excitatory and suppress inhibitory cell fates, or vice versa. In the present study, we investigated the development of glutamatergic and GABAergic neurons in the dorsal horn of the embryonic spinal cord, where neurogenesis has been extensively character-

ized15–17. During the early phase of neurogenesis in the mouse spinal cord (embryonic day (E)10.5 to E11.5), six classes of interneurons are generated along the dorsoventral axis, and these cells predominantly settle in the deep laminae of the dorsal horn15–17. From E11.5 to E13.5, two late-born populations of dorsal-horn neurons populate the superficial laminae, and they are distinguished by their complementary expression of the homeobox genes Lmx1b and Pax2 (refs. 15–17). The Tlx-class homeobox genes Tlx3 (previously known as Rnx and Hox11L2) and Tlx1 (previously known as Hox11) are also expressed in subpopulations of dorsal-horn neurons18, and they are required for proper formation of several brainstem nuclei18,19. However, the relationship between the expression of transcription factors and the development of GABAergic and glutamatergic transmitter phenotypes in the dorsal horn is not known. Here we provide evidence that Tlx1 and Tlx3 serve as post-mitotic selector genes that determine glutamatergic versus GABAergic cell fates in the embryonic dorsal spinal cord. Furthermore, our findings indicate that the respiratory dysfunction in Tlx3-deficient mice20 is caused by excess GABA-mediated inhibition. RESULTS Neurotransmitter identities of Pax2+ and Tlx3+ neurons Pax2 and Tlx3 are both expressed in subsets of differentiating dorsal horn neurons during the early18,21 and the late phases of dorsal neu-

1The

Dana-Farber Cancer Institute and 2Department of Neurobiology, Harvard Medical School, 1 Jimmy Fund Way, Boston, Massachusetts 02115, USA. 3Laboratory for Memory and Learning, Brain Science Institute, RIKEN, Wako, Saitama 351-0198, Japan. 4Molecular Neurobiology Laboratory, The Salk Institute for Biological Studies, 10010 North Torrey Pines Road, La Jolla, California 92037, USA. 5Center for Biotechnology, Showa University, Tokyo 142-8555, Japan. 6Department of Pathology, International Medical Center of Japan, 1-21-1 Toyama, Shinjuku-ku, Tokyo 162-8655, Japan. 7Research Institute of Molecular Pathology, Dr Bohr-Gasse 7, A-1030 Vienna, Austria. 8Department of Physiology, Showa University School of Medicine, Tokyo 142-8555, Japan. 9These authors contributed equally to this work. Correspondence should be addressed to Q.M. ([email protected]). Published online 4 April 2004; doi:10.1038/nn1221

510

VOLUME 7 | NUMBER 5 | MAY 2004 NATURE NEUROSCIENCE


ARTICLES Figure 1 Relationship between Tlx3 or Pax2 expression and the glutamatergic or GABAergic neurotransmitter phenotype. (a) Drawing shows excitatory and inhibitory neurons in the dorsal spinal cord. The sensory information carried by the primary sensory afferents is relayed through the dorsal horn of the spinal cord. The ascending relay sensory neurons and some local interneurons are excitatory (+) and glutamatergic, and their synaptic transmission is modulated by local inhibitory interneurons (–), many of which are GABAergic. (b–g) Transverse sections through E13.5 (b–d,f,g) and E12.5 (e) cervical/thoracic dorsal spinal cord. (b,c) In situ hybridization with Tlx3 and Pax2 as the probes, respectively. (d,f) Pseudocolor double staining of nuclear Pax2 protein (green) and cytoplasmic Tlx3 mRNA (d, red) or VGLUT2 mRNA (f, red). (e) Bright-field double staining of Pax2 protein (brown) and Gad1 mRNA (purple). (g) Double staining of Tlx3 protein (brown) and VGLUT2 mRNA (purple). (h) The percentages of Tlx3+ cells coexpressing Pax2 (971 Tlx3+ cell examined, E13.5) or VGLUT2 (953 Tlx3+ cells examined, E13.5) and the percentage of Pax2+ cells coexpressing VGLUT2 (720 Pax2+ cells examined, E13.5) and Gad1 (775 Pax2+ cells examined, E12.5) in the dorsolateral area of the dorsal horn (boxed area in b). (i) Numbers and percentages of neurons expressing Tlx3 and Pax2 in the boxed area of b. Total number of neurons was determined by counting cells expressing the pan-neural marker Stmn2. Data given as mean ± s.e.m.

rogenesis (Fig. 1b,c). Initially, Tlx3 and Pax2 are expressed in discrete longitudinal columns along the dorsoventral axis that correspond to dI3 (previously named D2) and dI5 (also called D4) neurons (Tlx3+), and to dI4 and dI6 neurons (Pax2+)18,21,22. Tlx3 expression in dI3 and dI5 neurons is transient18 and by E13.5, these cells have settled in the dorsal horn deep laminae (data not shown). At later times (E11.5 to E13.5), Tlx3+ and Pax2+ neurons are generated and migrate from a single large dorsal progenitor domain, from which they form the superficial laminae (probably laminae I–III/IV) of the dorsal horn18,21. To examine Tlx expression in late-born dorsal neurons, we focused on the most dorsolateral area of the E13.5 spinal cord where Tlx3 is expressed extensively (Fig. 1b). Double-staining of Tlx3 mRNA in the cytoplasm and Pax2 protein in the nuclei showed that fewer than 4% of Tlx3+ cells examined in E13.5 dorsal horn coexpressed Pax2 (Fig. 1d,h). We then examined the expression of the pan-neural marker Stmn2 (also known as SCG10)23 to determine the total neuron number in the dorsolateral dorsal horn. By counting the numbers of Stmn2+, Tlx3+ and Pax2+ neurons on adjacent sections, we estimated that 60% and 40% of neurons in the dorsal lateral spinal cord (Fig. 1b, boxed area) express Tlx3 and Pax2, respectively (Fig. 1i).

These data provide evidence that at E13.5, Tlx3+ and Pax2+ cells represent two intermingled populations of dorsal-horn neurons. Pax2 is expressed in developing GABAergic neurons in the cerebellum24. To determine whether Pax2 also marks GABAergic cells in the dorsal spinal cord, we examined the neurotransmitter phenotype of Pax2+ cells in the dorsal horn. GABAergic neurons specifically express two enzymes that regulate GABA synthesis, glutamic acid decarboxylases GAD67 (encoded by the gene Gad1) and GAD65 (encoded by the gene Gad2)25, as well as the inhibitory amino acid transporter that packages GABA or glycine into synaptic vesicles (encoded by the gene Viaat)26. Double-staining of Gad1 mRNA and Pax2 protein in the E12.5 dorsal horn showed that 98% of the Pax2+ cells coexpress Gad1 (Fig. 1e,h). At E15.5 and at birth (P0), Pax2+ cells in the superficial laminae continued to express Gad1; however, many Pax2+ cells in the deeper laminae no longer expressed Gad1 (data not shown). The complementary expression of Tlx3 and Pax2 is reminiscent of the mutually exclusive development of GABAergic and glutamatergic neurons2,3. We therefore asked whether Tlx3 is selectively expressed in glutamatergic neurons. Glutamatergic neurons are defined by the expression of the vesicular glutamate transporters VGLUT1–3 (also called Slc17 family proteins)3,27. In situ hybridization analyses showed

Figure 2 Compromised development of GABAergic neurons in the Pax2 mutant dorsal horn. (a–d) Transverse sections through E14 (a,b) and E13 (c,d) cervical/thoracic dorsal spinal cord. In situ hybridization was done with indicated probes. Note a loss of Gad1 expression in the dorsal horn (b). Expression of the glutamatergic marker VGLUT2 was, however, not affected (c versus d).

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 5 | MAY 2004

511


ARTICLES that only VGLUT2 (Slc17a6)3,27 was expressed in embryonic spinal cord from E11.5 to P0 (Figs. 1 and 3 and data not shown). The presence of a few cells expressing VGLUT1 (Slc17a7) in the adult spinal cord28 seems to reflect the late onset of VGLUT1 expression in these neurons. Significantly, double-staining of Pax2 protein and VGLUT2 mRNA revealed that over 99% of Pax2+ cells in the E13.5 dorsal horn did not express VGLUT2 (Fig. 1f,h). In contrast, double staining of Tlx3 protein and VGLUT2 mRNA showed that >96% of Tlx3+ cells in the dorsolateral area (Fig. 1b, boxed area) coexpressed VGLUT2 (Fig. 1g,h), consistent with our observation that both Tlx3 and VGLUT2 are expressed predominantly in Pax2– cells. However, newly formed Tlx3+ neurons in the region close to the ventricular zone did not yet express VGLUT2 (data not shown). At P0, Tlx3+ cells are enriched in the superficial laminae of the dorsal horn18 and they retain the coexpression with VGLUT2 (data not shown), although our studies did not rule out that Tlx3 might be expressed transiently in a subset of deep dorsal-horn neurons (in addition to the early-born dI3 and dI5 interneurons18). In summary, at least within the dorsal horn superficial laminae, Pax2 and Tlx3 are primarily expressed in differentiating GABAergic and glutamatergic neurons, respectively. Pax2 controls GABAergic differentiation Examination of the Pax2-null spinal cords29 revealed a nearly complete elimination of Gad1 expression in the dorsal horn of E13 (data not shown) and E14 mutant mice (Fig. 2). The expression of two other GABAergic markers, Gad2 and Viaat, was reduced but not completely eliminated (data not shown). In contrast, GABAergic neuron differentiation in the most ventral spinal cord was largely unaffected in Pax2 mutants, consistent with a lack of Pax2 expression in many ventralhorn GABAergic neurons (data not shown). No change in VGLUT2 expression was observed in the E13 Pax2 mutant dorsal horn (Fig. 2), indicating Pax2 is required specifically for GABAergic differentiation. Regulation of dorsal glutamatergic differentiation We then asked whether Tlx3 and its related family member Tlx1 are required for glutamatergic differentiation. Because Tlx1 is expressed in a subset of Tlx3+ cells in the dorsal spinal cord, where it partially compensates for the loss of Tlx3 function18, we analyzed both Tlx3 and Tlx1 single mutants as well as Tlx1/3 compound mutants (hereafter referred to as Tlx mutants). At E14.5, VGLUT2 expression in the superficial dorsal horn was unchanged in Tlx1 mutants (data not shown), reduced in Tlx3-deficient mice (Fig. 3a,b), and absent in Tlx compound mutants (Fig. 3c,d). The loss of VGLUT2 expression in the Tlx mutant dorsal horn was first observed at E12.5 (data not shown) and persisted at P0 (Fig. 3f), indicating that Tlx1 and Tlx3 are required for glutamatergic differentiation in these neurons. In contrast, glutamatergic neuron development in the deep laminae of the dorsal horn, where dI1 and dI2 neurons settle, was not affected at either E14.5 or P0 (Fig. 3d,f), which is consistent with a lack of Tlx gene expression in dI1 and dI2 interneurons15–17. Given that there is no increase in cell death in Tlx mutant spinal cord18, we conclude that Tlx3 and Tlx1 are necessary for the development of the glutamatergic neurotransmitter phenotype in the superficial laminae of the dorsal horn. Tlx genes suppress GABAergic differentiation We then analyzed the expression of GABAergic markers in Tlx mutants to determine whether the Tlx genes function as ‘selector’ genes14 that regulate the alternative transmitter fate in the dorsal horn. In E16.5 Tlx mutants, the expression of Viaat, Gad1 and Gad2 was markedly expanded throughout the dorsal horn, rather than showing a mosaic pattern as in wild-type embryos (Fig. 4 and data

512

Figure 3 Reduction or loss of the glutamatergic marker in Tlx single- or double-null dorsal spinal cord. (a,b) Transverse sections through the dorsal horn of the cervical/thoracic spinal cord of E14.5 wild type (a) and Tlx3 single mutants (b). (c–f) Transverse section through E14.5 (c,d) and P0 (e,f) wildtype or Tlx1/3 double mutant embryos. In situ hybridization was done with VGLUT2 as the probe. VGLUT2 expression in the dorsal horn was reduced in Tlx3 single (b) or eliminated in Tlx compound mutants (d, f), but its expression in the deep laminae was not affected (d, arrowheads).

not shown). The expansion of these GABAergic markers in the Tlx mutant mice was first seen at E11.5, and persisted at P0 (data not shown). An expansion of Pax2 expression in the mutant dorsal horn was also observed from E11.5 to P0 (Fig. 4f and data not shown). Two possible mechanisms could underlie the increase of GABAergic neuron cell number in the Tlx mutant dorsal horn: (i) an expansion in the endogenous GABAergic neuron population and/or (ii) a transformation of glutamatergic into GABAergic cells. The first possibility seems unlikely, as the Tlx genes are expressed only in postmitotic neurons19,30. More importantly, even though ∼60% of the neurons in the dorsal horn express Tlx3 (Fig. 1), Tlx3 mutant dorsal horn does not show increased cell death18. To assess whether prospective glutamatergic neurons differentiate as GABAergic neurons do, we tested whether GABAergic markers such as Pax2 are upregulated in these cells. To do this, we made use of the observation that 95% of the cells in the dorsal horn that express the LIM-class homeobox gene Lmx1b coexpress Tlx3 at E13.5 (Fig. 4g,i), although at late stages many Lmx1b+ cells in the deep laminae no longer express Tlx3 (data not shown). Moreover, double staining of Lmx1b protein and VGLUT2 mRNA showed that at E13.5, 98% of Lmx1b+ cells express VGLUT2 in the dorsolateral area of the dorsal horn (Fig. 4h,i). Pax2, however, is not expressed in these cells21,22 (Fig. 4j). Because Lmx1b is



ARTICLES Figure 4 Derepression of GABAergic markers in Tlx-null dorsal spinal cord. (a–f) Transverse sections through the dorsal horn of the cervical/thoracic spinal cord of E16.5 wild-type and mutant dorsal horn. In situ hybridization was done with indicated probes. (g) Double-staining of Tlx3 protein (brown) and Lmx1b mRNA (purple). (h) Double-staining of Lmx1b protein (brown) and VGLUT2 mRNA (purple). (i) Percentages of Lmx1b+ cells expressing Tlx3 (980 Lmx1b+ cells examined) or VGLUT2 (970 Lmx1b+ cells examined). (j,k) Pseudocolor double-staining of Pax2 protein in the nuclei (green) and Lmx1b mRNA (red) in E12.5 wildtype (j) and Tlx mutants (k). (l) The percentage of Lmx1b+ cells that coexpress Pax2 in E12.5 wildtype and Tlx-null mutant embryos. Only the dorsolateral area of the dorsal horn was counted.

still present in the E12.5 and E13.5 Tlx mutant dorsal horn (Fig. 4k), we used it as a surrogate marker for dorsal Tlx3+ glutamatergic neurons. Whereas fewer than 2% of Lmx1b+ cells coexpressed Pax2 in E12.5 wild-type embryos (Fig. 4j,l), in E12.5 (Fig. 4k,l) and E13.5 (data not shown) Tlx mutants, we observed coexpression of Pax2 in more than 90% of Lmx1b+ cells in the dorsolateral region (equivalent to the boxed area in Fig. 1b). However, newly born Lmx1b+ cells adjacent to the ventricular zone had not yet shown a derepression of Pax2 expression (data not shown). This finding suggests that prospective glutamatergic neurons are transformed into Pax2+ GABAergic cells in the Tlx mutant spinal cord. Fate switch by ectopic Tlx3 expression We then tested whether the Tlx genes are sufficient to suppress GABAergic cell fate by misexpressing mouse Tlx3 in the developing chick spinal cord (Fig. 5a,d). Stages 11–12 (E2) brachial neural tubes were electroporated with a Myc-tagged mouse Tlx3 expression construct. When electroporated neural tubes were analyzed three days later, Pax2 protein expression was markedly repressed on the Tlx3electroporated side of the neural tube (Fig. 5a,b). GABA immuno-

staining and Gad1 expression were also reduced (Fig. 5c,e), indicating Tlx3 was sufficient to suppress endogenous GABAergic differentiation. To ascertain whether ectopic Tlx3 expression (Fig. 5d) led to a concomitant induction of glutamatergic neurons, we examined chick VGLUT2 expression in adjacent sections. At E5, VGLUT2 was normally expressed in alternating dorsoventrally restricted stripes in wild-type E5 chick neural tube (Fig. 5f). These stripes appeared to be complementary to Gad1 expression (Fig. 5e,f). Misexpression of Tlx3 resulted in an expansion of VGLUT2 expression in the domains where GABAergic neurons were normally located, resulting in a continuous band of VGLUT2-expressing cells (Fig. 5f). These findings suggested that ectopic Tlx gene expression was able to switch differentiating neurons from a GABAergic to a glutamatergic cell fate. Interestingly, ectopic expression of Pax2 in chick neural tube did not induce GABAergic differentiation (data not shown), suggesting that Pax2 is necessary but not sufficient for GABAergic differentiation in the dorsal neural tube. Reciprocal regulation of AMPA and kainate receptors Tlx1 and Tlx3 are not expressed in the ventral spinal cord or in the forebrain, suggesting that in addition to a central role in regulating the neurotransmitter phenotypes, they may control region-specific neuronal identities. We therefore examined the expression of additional markers that are differentially expressed in either glutamatergic

Figure 5 Cell fate switch by ectopic expression of mouse Tlx3. (a–f) Transverse sections through two E5 chick embryos (a–c and d–f, respectively), with the left side of the neural tubes electroporated with mouse Tlx3 expression construct (RCAS-Tlx3). (a) Expression of the MycTlx3 fusion protein detected by immunostaining with a Myc antibody (arrow). (b,c) Immunostaining with Pax2 and GABA antibodies, respectively. Note a reduction of Pax2 expression and GABA staining on the electroporated sides (arrowheads vs. arrows). (d,f) In situ hybridization with Tlx3, Gad1 and VGLUT2 as the probes. Mouse Tlx3 was expressed throughout the entire spinal cord on the electroporated side (d, arrow). Whereas expression of dorsal Gad1 was repressed on the electroporated side (e, arrowheads vs. arrows), VGLUT2 expression was accordingly expanded (f, arrowheads vs. arrows).


513


ARTICLES Figure 6 Regulation of AMPA and kainate receptors by Tlx genes and a summary of Tlx gene functions. (a–f) Transverse sections through the dorsal horn of E14.5 (a,b,e,f) and E16.5 (c,d) mouse embryos. (a,b) Double-staining of Pax2 protein (brown) and Gria2 mRNA (a, purple) or Grik2 mRNA (b, purple) in wild-type embryos. Note most Grik2+ cells coexpressed Pax2 (b, arrow), although a small fraction of them did not (b, arrowhead). (c–f) loss of Gria2 (d) and an expansion of Grik2 (f) in Tlx mutants. (g) Tlx1 and Tlx3 likely function as selector genes that promote glutamatergic and suppress GABAergic cell fates in the dorsal horn of the spinal cord. Tlx target genes include transcription factors, molecules for neurotransmitter synthesis or transport, and several region-specific genes, including the AMPA receptor (Gria2) and the kainate receptors (Girk2/3). DRG11, a Tlx-dependent paired class homeobox gene18, is expressed in Pax2– cells, thus likely in Tlx3+ glutamatergic neurons also (data not shown). Molecules expressed in Pax2– cells, most of which are Tlx3+ glutamatergic, are eliminated in Tlx mutants, whereas molecules expressed in Pax2+ cells, most of which are GABAergic, are all derepressed in Tlx mutants. Loss of Tlx genes result in a transformation from a glutamatergic to a GABAergic cell fate, whereas ectopic Tlx3 expression seems to cause an opposite cell-fate switch.

or GABAergic spinal neurons. In the adult dorsal horn, strong expression of the AMPA-class glutamate receptor GluR2 (encoded by the gene Gria2) is selectively detected in glutamatergic neurons31. Consistent with this finding, strong Gria2 expression in the dorsal horn was first observed at E14.5 in presumably glutamatergic Pax2– neurons (Figs. 1 and 6a and data not shown). GABAergic neurons in neonatal dorsal horn express the kainate class glutamate receptor GluR6 (encoded by the gene Grik2)9,32. In the E14.5 dorsal horn, GluR6 and another kainate receptor subunit, GluR7 (encoded by Grik3), were preferentially expressed in presumably GABAergic Pax2+ cells (Figs. 1 and 6b). At E16.5, strong Gria2 expression observed in wild-type dorsal horn (Fig. 6c) was abolished in Tlx mutants (Fig. 6d). Meanwhile, Grik2 and Grik3, which are normally expressed in a mosaic pattern (Fig. 6e and data not shown), had an almost uniform pattern of expression in the E14.5 mutant dorsal horn (Fig. 6f and data not shown). These findings provide further evidence that the Tlx genes reciprocally control classes of genes that are selectively expressed in either glutamatergic or GABAergic dorsal horn neurons (summarized in Fig. 6g). Excess GABA inhibition causes respiratory failure The finding that Tlx3 selects glutamatergic over GABAergic cell fates led us to reexamine the neurological basis for the respiratory failure in mice lacking the Tlx3 gene20. Two interconnected oscillators in the ventrolateral medulla are thought to be involved in generating respiratory rhythms33,34. One consists of inspiratory (Insp) neurons in the preBötzinger Complex (preBötC), and the other consists of preinspiratory (Pre-I) neurons located rostral to the preBötC33,34. We used an in vitro brainstem-to-spinal cord preparation35 to record rhythmic respiratory-like activity from the C4 motor roots (Fig. 7). As previously reported20, the duration of the C4 inspiratory burst was significantly shorter in Tlx3 mutant mice (Fig. 7b,g). The burst duration of Insp and Pre-I neurons was also significantly shorter in Tlx3 mutants (6 Insp, 6 Pre-I) than in wild-type mice (7 Insp, 3 Pre-I) (Fig. 7g), with both neuronal populations showing arrhythmic firing patterns (Fig. 7b,e). Nevertheless, the resting membrane potentials and input resistances of

514

the Insp and Pre-I neurons did not differ significantly between Tlx3 mutant and wild-type mice (Fig. 7g). Moreover, the C4 inspiratory activity was in-phase with Insp neuron activity (Fig. 7a,b), suggesting that the neuronal connections from Insp neurons to motor neurons are not affected in Tlx3 mutants. Finally, the suppression of Pre-I neuron firing during the inspiratory phase was observed in both wide-type (Fig. 7d, arrow) and Tlx3-null (Fig. 7e, arrow) mice, indicating that the putative neuronal connections from Insp to Pre-I neurons were likely to be intact in the Tlx3 mutant hindbrain. Our observation that glutamatergic neurons are transformed into GABAergic cells in Tlx mutant spinal cord led us to test whether the short burst duration in Tlx3 mutant mice is due to excess GABAmediated inhibitory inputs to the respiratory pattern generator. To do this, we examined the effects of the GABAA antagonist, bicuculline, on the burst pattern and respiratory activity of Tlx3 mutant animals. In wild-type mice, application of 2 µM bicuculline did not cause a significant change in either the burst duration or burst rate of C4 motor neurons (data not shown)36. In contrast, the duration of C4 motor activity in Tlx3-null mice was lengthened, and this was accompanied by a decrease in burst rate to wild-type levels (Fig. 7c,f,g). Both Insp and Pre-I neurons returned to normal rhythmic activity in Tlx3 mutant mice treated with 2 µM bicuculline (Fig. 7c,f,g). Similar changes were induced with 10 µM picrotoxin (n = 5), a chloride channel blocker, or with a low Cl– solution (n = 5), where Cl– concentration was reduced to 30% of control (data not shown). To test whether respiratory neurons directly receive excess GABAergic synaptic input in Tlx3 mutants, we examined the effect of Cl– loading into respiratory neurons on the bursting activity. Following the rupture of the patch membrane with a high-Cl– electrode, depolarization of the burst phase appeared within several minutes, owing to a rapid increase in the intracellular Cl– concentration. The development of depolarization during the inspiratory phase after Cl– loading to inspiratory neurons was significantly larger in Tlx3 mutant mice (252%) than in wild-type mice (112%, P < 0.05; data not shown), suggesting that inspiratory neurons directly receive excess GABA-mediated inhibition.



ARTICLES Figure 7 Inspiratory and pre-inspiratory neurons in wild-type (Tlx3+/+) and Tlx3 mutant (Tlx3–/–) mice, and transformation of the burst pattern of Tlx3–/– respiratory neuron by bicuculline. (a–c) Inspiratory neurons (Insp) and (d–f) pre-inspiratory neurons (Pre-I). (a–f) The upper trace is the membrane potential trajectories (MP), and the lower trace is C4 inspiratory activity (C4). (a) An inspiratory neuron in a Tlx+/+ mouse. (b) An inspiratory neuron in a Tlx3–/– mouse. Note the short burst duration and irregular rhythms in Tlx3–/– mice. (c) Activity of the same neuron as in b after 10 min application of 2 µM bicuculline. Note a complete recovery of rhythmic activity. (d) A pre-inspiratory neuron in a Tlx+/+ mouse. (e) A pre-inspiratory neuron in a Tlx–/– mouse. Note short burst duration. (f) Activity of the same neuron as in e after 10 min application of 2 µM bicuculline. Note a complete recovery of rhythmic activity. Like in wild-type mice (d, arrow), synaptic inhibition of Pre-I neurons during inspiratory phases (e, arrow) was weakened after bicuculline treatment (f). All neurons were recorded in the ventrolateral medulla. (g) Quantitative changes of burst duration and rates between Tlx3+/+ and Tlx3–/– mice and a rescue of mutant phenotypes by the GABA antagonist bicuculline (Bic). Values are presented as mean ± s.d. *P < 0.01; **P < 0.001 (by Student’s t-test).

DISCUSSION Selector genes belong to a class of genes that control the fate of groups of cells during development14,37. Selector genes typically control the choice between two alternative fates, and they do so by activating a large number of genes that are expressed in a particular cell type, tissue or organ, while at the same time suppressing the genes associated with alternative fates37. Our findings suggest that Tlx3 and its family member Tlx1 function as selector genes to promote glutamatergic over GABAergic differentiation in the dorsal embryonic spinal cord. Support for this conclusion comes from several observations. First, Tlx3 is expressed in glutamatergic neurons in the developing dorsal horn of the spinal cord, and both generic and region-specific glutamatergic markers are absent in Tlx-null dorsal horn (Fig. 6g). Second, Tlx genes are able to repress the GABAergic cell fate, with the loss of Tlx genes causing a concomitant derepression of generic and regionspecific GABAergic markers in prospective glutamatergic cells (Figs. 4 and 6g). Third, ectopic Tlx3 expression is sufficient to repress endogenous GABAergic differentiation and to induce formation of glutamatergic cells in developing chick spinal cord (Fig. 5). Taken together, these findings suggest that the ‘on’ and ‘off ’ of Tlx gene expression determines whether neurons in the superficial laminae of embryonic dorsal horn develop as glutamatergic or GABAergic cell types. Moreover, since the Tlx genes are expressed in post-mitotic neurons19,30, the observation that glutamatergic neurons are transformed into GABAergic neurons in Tlx-null embryos suggests in the developing dorsal spinal cord, newly born post-mitotic neurons retain the plasticity to become either excitatory or inhibitory. Our results indicate that Pax2 is critically involved in the initial acquisition of the GABAergic cell phenotype in the dorsal spinal cord.


However, unlike the neurotransmitter switch in Tlx mutants, the loss of GABAergic markers in the Pax2 mutant spinal is not accompanied by an expansion of glutamatergic cells (Fig. 2). Thus, Pax2 seems to control terminal aspects of the GABAergic differentiation program. Whereas most Pax2+ cells at E12.5 are GABAergic (Fig. 1), many Pax2+ cells in the deep laminae of dorsal horn no longer express Gad1 at P0 (data not shown). This latter finding is consistent with the progressive reduction in the number of GABAergic neurons in dorsal horn deep laminae at prenatal and postnatal stages38,39. An endogenous neurotransmitter switch has been described in the sympathetic nervous system40. However, it is presently unclear whether the apparent loss of GABAergic markers in a subset of Pax2+ neurons reflects a change of neurotransmitter phenotypes, such as a switch from GABAergic to glycinergic transmission41. In the forebrain, the proneural genes Mash1 and Ngn1/2 are expressed in ventral and dorsal progenitors that give rise to GABAergic and glutamatergic neurons, respectively12. Interestingly, mutation of Ngn1/2 leads to ectopic development of GABAergic neurons from the dorsal precursors11, indicating that in both the dorsal spinal cord and the forebrain, there are genetic programs suppressing GABAergic differentiation in prospective glutamatergic precursors or neurons. It should be noted that Tlx1 and Tlx3 are not expressed in the forebrain19,30 (data not shown). Furthermore, the Dlx-class homeobox genes that have been implicated in controlling GABAergic differentiation in the forebrain are not expressed in the developing spinal cord42. Therefore, regulation of glutamatergic and GABAergic transmitter phenotypes in different areas of the nervous system is controlled by distinct sets of transcriptional regulators. These region-specific genes may function to coordinate the neuro-

515


ARTICLES transmitter phenotype of particular neuronal populations with other region-specific aspects of neuronal identity. For example, the Tlx genes reciprocally control the expression of an AMPA receptor gene Gria2 and the two kainate receptor genes Grik2/3 in the dorsal horn neurons (Fig. 6g). Tlx3-deficient mice die from a defect in the central control of respiration20. Our studies suggest that the basic respiratory neuronal network in the ventral medulla probably forms normally in Tlx3 mutant mice. However, rhythmic respiratory neurons in Tlx3null mice receive excess GABAergic synaptic inputs and remarkably, once GABA-mediated inhibition is reduced, a normal rhythmic firing pattern can be generated (Fig. 7). The source of excess GABA-mediated inhibition in Tlx3-null mice remains to be determined. Tlx3 is expressed in four groups of neurons in the hindbrain: the nucleus of the solitary tract (nTS), the (nor)adrenergic (NA) centers, the trigeminal nuclei and D4 interneurons18,19. The nTS and NA neurons provide stimulating and depressing inputs to the ventral respiratory center, respectively43, and their development is dependent on Tlx3 expression19. It is possible that nTS and/or NA neurons might be converted into GABAergic neurons, thus providing excess GABAergic inhibitory inputs to the respiratory center in Tlx3 mutants. Confirmation of this hypothesis will require the development of a marking system to identify NTS and NA neurons so that neurotransmitter phenotypes can be analyzed in a mutant background. In summary, our results suggest that Tlx1 and Tlx3 serve as postmitotic selector genes that determine a glutamatergic over GABAergic neuron cell fate in the embryonic dorsal spinal cord. Our studies also show that mutation of a region-specific selector gene can create an imbalance of excitation and inhibition that has a profound behavioral consequence. METHODS Animals. The generation of Tlx1 and Tlx3 mutant mice and Pax2 mutants has been described previously20,29,44. The morning that vaginal plugs were observed was considered to be E0.5. Genotyping was done as described previously18,29. In situ hybridization, immunostaining and cell counting. Our in situ hybridization methods were as described previously45. The following mouse in situ probes, including Gad1 (1 kb), Gad2 (0.79 kb), VGLUT1 (0.71 kb), VGLUT2 (0.72 kb), Pax2 (0.7 kb), Gria2 (0.8kb), Grik2 (0.8 kb) and Grik3 (0.75 kb) were amplified with gene-specific sets of PCR primers and cDNA templates prepared from E13.5 or P0 mouse brains. Chick Gad1 (0.74 kb) was amplified from cDNA from E7 chick spinal cord. A fragment of the chick VGLUT2 gene (0.83 kb), which encodes an amino acid sequence showing 97% identity to the mouse VGLUT2 protein, was amplified by degenerate oligonucleotides 5′-TT(T/C) AA(T/C) TGG GA(T/C) CCI GA(A/G) AC-3′ and 5′-CAT ICC (A/G)AA ICC ICC (A/G)CA (A/G)TT CAT-3′ that are derived from the conserved regions within the VGLUT family members, FNWDPQT and MNCGGFGM, respectively. Other probes include Tlx118, Tlx318, Lmx1b46 and Islet118. For double staining, we used an in situ hybridization procedure, followed by immunostaining with Pax2-specific antibody (Zymed Laboratories Inc.), Lmx1b-specific antibody (from T. Jessell, Columbia University) or Tlx3specific antibody (M. Goulding, Salk Institute). The specificity of the antibody to Tlx3 was indicated by the perfect match between Tlx3 mRNA and Tlx3 immunostaining (data not shown). For robust antibodies, such as Pax2, in situ hybridization was performed with a pre-treatment of proteinase K (2 µg/ml; Roche) for 5 min, which enhanced the hybridization signal. Biotin-conjugated secondary antibody (Vector) was used for brightfield double staining. For in situ hybridization combined with fluorescent immunostaining (with Pax2 antibody), the in situ signals were photographed under trans-luminescent light and converted into pseudo-red fluorescent color, whereas Pax2 protein was detected with Alexa 488-conjugated secondary antibodies (Molecular Probes). Immunostaining on frozen

516

chick spinal sections was done with mouse anti-Myc (Hybridoma Bank) and guinea pig anti-GABA (Chemicon) antibodies. Counting of single- or double-positive cells was done as follows. Three to four adjacent sets of E12.5 or E13.5 spinal cord sections at the cervical levels were used, and only those cells containing nuclei and in the dorsal lateral areas were counted (boxed area in Fig. 1b). The dorsolateral area was selected for two reasons: first, the areas closer to the ventricular zone contain newly formed Tlx3+ cells that may have not yet turned on VGLUT2 expression, and second, early-born DI1 and DI2 interneurons do not express Tlx genes18 and they are also glutamatergic (data not shown). These cells normally settle in the ventral portion of the dorsal horn15–17. In-ovo electroporation. The cDNA fragment, encoding a fusion protein of Tlx3 and Myc epitope tag, was cloned to the RCASBP chick viral expression vector47, and the resulting construct is referred to as RCAS-Tlx3. The purified plasmid DNA was re-suspended at concentrations of 2–5 µg/µl. RCAS-Tlx3 plus a GFP (encoding the green fluorescence protein) control expression plasmid, pCAX-IRES-GFP21 were co-injected into neural tubes of stage 11–12 chick embryos. After electroporation, the embryos were allowed to grow at 37.5 °C for a further 48–72 h. Embryos with a high level of GFP fluorescence were fixed, and changes in the cell pattern in the spinal cord were analyzed. Electrophysiological recording. The medulla and spinal cord of mice of the first day after birth (38 preparations) were isolated under deep ether anesthesia, according to methods previously described in rats35,48. Briefly, the brainstem was rostrally decerebrated between the sixth cranial nerve roots and the lower border of the trapezoid body, so that most of the pons was removed. The preparation was continuously superfused at a rate of 2.5–3.0 ml/min in a 2-ml chamber with the following modified Krebs solution: NaCl, 124 mM; KCl, 5.0 mM; KH2PO4, 1.2 mM; CaCl2, 2.4 mM; MgSO4, 1.3 mM; NaHCO3, 26 mM; and glucose, 30 mM; equilibrated with 95% O2 and 5% CO2; at 25– 26 °C, pH 7.4. Low chloride solutions were prepared by mixing the standard solution and the chloride-free solution that was made by substituting sodium isethionate, potassium propionate and calcium propionate for the chloride salts of sodium, potassium and calcium of the standard solution, respectively. Drugs were obtained from Sigma. Respiratory-like activity corresponding to the inspiratory rhythm was monitored at the C4/C5 ventral root35 through a glass capillary suction electrode and a high-pass filter with a 0.3-s time constant. Membrane potentials of inspiratory neurons in the ventrolateral medulla were recorded using conventional whole-cell patch clamp methods49. The electrodes that had a tip inner diameter of 1.2–2.0 mm and resistance of 4–8 MΩ were filled with following pipette solution: K-gluconate, 130 mM; EGTA, 10 mM; HEPES, 10 mM; Na2-ATP, 2 mM; CaCl2, 1 mM; and MgCl2, 1 mM, with pH 7.2–7.3 adjusted by KOH. A high Cl– patch electrode solution contained 130 mM KCl instead of potassium gluconate. The membrane potentials were recorded with a single-electrode voltage-clamp amplifier (CEZ-3100, Nihon Kohden) after compensation of the series resistance (20– 50 MΩ) and capacitance. For comparison with previous studies49, the membrane potential values were not corrected for junction potentials (less than –10 mV). Neuronal activity and C4/C5 activity were stored on magnetic tape in a DAT recorder (RD-120TE, TEAC) for subsequent data analysis. Values are presented as mean ± standard deviation (s.d.). The differences in values were considered to be significant at P < 0.05 by Student’s t-test. The GenBank accession number for the chick VGLUT2 cDNA is AY559247. ACKNOWLEDGMENTS We thank S. Korsmeyer for providing Tlx3 and Tlx1 knockout mice. We are grateful to F. Guillemot, C. Schuuman, C. Stiles, M. Greenberg, G. Lemke and Z. He for critical comments or discussion. Q.M. is a Claudia Adams Barr Scholar and a Pew Scholar in Biomedical Sciences. P.A.G. is a Parker B. Francis Fellow in Pulmonary Medicine and C.C. is a Medical Foundation Fellow. This work was supported by grants from the National Institutes of Health to Q.M. and M.G., and in part by a Showa University Grant-in-Aid for Innovative Collaborative Research Projects from the Japanese Ministry of Education, Culture, Sports, Science and Technology to H.O. COMPETING INTERESTS STATEMENT The authors declare that they have no competing financial interests.


ARTICLES


Received 21 January; accepted 8 March 2004 Published online at http://www.nature.com/natureneuroscience/ 1. Bennett, M.R. & Balcar, V.J. Forty years of amino acid transmission in the brain. Neurochem. Int. 35, 269–280 (1999). 2. Bellocchio, E.E., Reimer, R.J., Fremeau, R.T.J. & Edwards, R.H. Uptake of glutamate into synaptic vesicles by an inorganic phosphate transporter. Science 289, 957–960 (2000). 3. Fremeau, R.T.J. et al. The expression of vesicular glutamate transporters defines two classes of excitatory synapse. Neuron 31, 247–260 (2001). 4. Polgar, E., Fowler, J.H., McGill, M.M. & Todd, A.J. The types of neuron which contain protein kinase C gamma in rat spinal cord. Brain Res. 833, 71–80 (1999). 5. Azkue, J.J. et al. Glutamate-like immunoreactivity in ascending spinofugal afferents to the rat periaqueductal grey. Brain Res. 790, 74–81 (1998). 6. Lu, Y. & Perl, E.R. A specific inhibitory pathway between substantia gelatinosa neurons receiving direct C-fiber input. J. Neurosci. 23, 8752–8758 (2003). 7. Melzack, R. & Wall, P.D. Pain mechanisms: a new theory. Science 150, 971–979 (1965). 8. Malcangio, M. & Bowery, N.G. GABA and its receptors in the spinal cord. Trends Pharmacol. Sci. 17, 457–462 (1996). 9. Kerchner, G.A., Wang, G.D., Qiu, C.S., Huettner, J.E. & Zhuo, M. Direct presynaptic regulation of GABA/glycine release by kainate receptors in the dorsal horn: an ionotropic mechanism. Neuron 32, 477–488 (2001). 10. Dickenson, A.H. Gate control theory of pain stands the test of time. Br. J. Anaesth. 88, 755–757 (2002). 11. Fode, C. et al. A role for neural determination genes in specifying the dorsoventral identity of telencephalic neurons. Genes Dev. 14, 67–80 (2000). 12. Anderson, S.A., Kaznowski, C.E., Horn, C., Rubenstein, J.L. & McConnell, S.K. Distinct origins of neocortical projection neurons and interneurons in vivo. Cereb. Cortex 12, 702–709 (2002). 13. Muzio, L. et al. Conversion of cerebral cortex into basal ganglia in Emx2–/– Pax6(Sey/Sey) double-mutant mice. Nat. Neurosci. 5, 737–745 (2002). 14. Garcia-Bellido, A. Genetic control of wing disc development in Drosophila. Ciba. Found. Symp. 0, 161–182 (1975). 15. Caspary, T. & Anderson, K.V. Patterning cell types in the dorsal spinal cord: what the mouse mutants say. Nat. Rev. Neurosci. 4, 289–297 (2003). 16. Goulding, M., Lanuza, G., Sapir, T. & Narayan, S. The formation of sensorimotor circuits. Curr. Opin. Neurobiol. 12, 508–515 (2002). 17. Helms, A.W. & Johnson, J.E. Specification of dorsal spinal cord interneurons. Curr. Opin. Neurobiol. 13, 42–49 (2003). 18. Qian, Y., Shirasawa, S., Chen, C.L., Cheng, L. & Ma, Q. Proper development of relay somatic sensory neurons and D2/D4 interneurons requires homeobox genes Rnx/Tlx3 and Tlx-1. Genes Dev. 16, 1220–1233 (2002). 19. Qian, Y. et al. Formation of brainstem (nor)adrenergic centers and first-order relay visceral sensory neurons is dependent on homeodomain protein Rnx/Tlx3. Genes Dev. 15, 2533–2545 (2001). 20. Shirasawa, S. et al. Rnx deficiency results in congenital central hypoventilation. Nat. Genet. 24, 287–290 (2000). 21. Gross, M.K., Dottori, M. & Goulding, M. Lbx1 specifies somatosensory association interneurons in the dorsal spinal cord. Neuron 34, 535–549 (2002). 22. Muller, T. et al. The homeodomain factor Lbx1 distinguishes two major programs of neuronal differentiation in the dorsal spinal cord. Neuron 34, 551–562 (2002). 23. Stein, R., Orit, S. & Anderson, D.J. The induction of a neural-specific gene, SCG10, by nerve growth factor in PC12 cells is transcriptional, protein synthesis dependent, and glucocorticoid inhibitable. Dev. Biol. 127, 316–325 (1988). 24. Maricich, S.M. & Herrup, K. Pax-2 expression defines a subset of GABAergic interneurons and their precursors in the developing murine cerebellum. J. Neurobiol. 41, 281–294 (1999). 25. Erlander, M.G., Tillakaratne, N.J., Feldblum, S., Patel, N. & Tobin, A.J. Two genes encode distinct glutamate decarboxylases. Neuron 7, 91–100 (1991).


26. McIntire, S.L., Reimer, R.J., Schuske, K., Edwards, R.H. & Jorgensen, E.M. Identification and characterization of the vesicular GABA transporter. Nature 389, 870–876 (1997). 27. Kaneko, T. & Fujiyama, F. Complementary distribution of vesicular glutamate transporters in the central nervous system. Neurosci. Res. 42, 243–250 (2002). 28. Kullander, K. et al. Role of EphA4 and EphrinB3 in local neuronal circuits that control walking. Science 299, 1889–1892 (2003). 29. Bouchard, M., Pfeffer, P. & Busslinger, M. Functional equivalence of the transcription factors Pax2 and Pax5 in mouse development. Development 127, 3703–3713 (2000). 30. Logan, C., Wingate, R.J.T., McKay, I.J. & Lumsden, A. Tlx-1 and Tlx-3 homeobox gene expression in cranial sensory ganglia and hindbrain of the chick embryo: markers of patterned connectivity. J. Neurosci. 18, 5389–5402 (1998). 31. Kerr, R.C., Maxwell, D.J. & Todd, A.J. GluR1 and GluR2/3 subunits of the AMPA-type glutamate receptor are associated with particular types of neuron in laminae I–III of the spinal dorsal horn of the rat. Eur. J. Neurosci. 10, 324–333 (1998). 32. Kerchner, G.A., Wilding, T.J., Huettner, J.E. & Zhuo, M. Kainate receptor subunits underlying presynaptic regulation of transmitter release in the dorsal horn. J. Neurosci. 22, 8010–8017 (2002). 33. Onimaru, H., Arata, A. & Homma, I. Neuronal mechanisms of respiratory rhythm generation: an approach using in vitro preparation. Jpn. J. Physiol. 47, 385–403 (1997). 34. Feldman, J.L., Mitchell, G.S. & Nattie, E.E. Breathing: rhythmicity, plasticity, chemosensitivity. Annu. Rev. Neurosci. 26, 239–266 (2003). 35. Suzue, T. Respiratory rhythm generation in the in vitro brain stem-spinal cord preparation of the neonatal rat. J. Physiol. 354, 173–183 (1984). 36. Ritter, B. & Zhang, W. Early postnatal maturation of GABAA-mediated inhibition in the brainstem respiratory rhythm-generating network of the mouse. Eur. J. Neurosci. 12, 2975–2984 (2000). 37. Mann, R.S. & Carroll, S.B. Molecular mechanisms of selector gene function and evolution. Curr. Opin. Genet. Dev. 12, 592–600 (2002). 38. Ma, W., Behar, T. & Barker, J.L. Transient expression of GABA immunoreactivity in the developing rat spinal cord. J. Comp. Neurol. 325, 271–290 (1992). 39. Schaffner, A.E., Behar, T., Nadi, S., Smallwood, V. & Barker, J.L. Quantitative analysis of transient GABA expression in embryonic and early postnatal rat spinal cord neurons. Brain Res. Dev. Brain Res. 72, 265–276 (1993). 40. Landis, S.C. Target regulation of neurotransmitter phenotype. Trends Neurosci. 13, 344–350 (1990). 41. Nabekura, J. et al. Developmental switch from GABA to glycine release in single central synaptic terminals. Nat. Neurosci. 7, 17–23 (2004). 42. Panganiban, G. & Rubenstein, J.L. Developmental functions of the Distal-less/Dlx homeobox genes. Development 129, 4371–4386 (2002). 43. Blessing, W.W. The Lower Brainstem and Body Homeostasis (Oxford Univ. Press, New York, 1997). 44. Roberts, C.W., Shutter, J.R. & Korsmeyer, S.J. Hox11 controls the genesis of the spleen. Nature 368, 747–749 (1994). 45. Birren, S.J., Lo, L.C. & Anderson, D.J. Sympathetic neurons undergo a developmental switch in trophic dependence. Development 119, 597–610 (1993). 46. Chen, Z.F. et al. The paired homeodomain protein DRG11 is required for the projection of cutaneous sensory afferent fibers to the dorsal spinal cord. Neuron 31, 59–73 (2001). 47. Morgan, B.A. & Fekete, D.M. Manipulating gene expression with replication-competent retroviruses. in Methods in Avian Embryology Vol. 51 (ed. Bronner-Fraser, M.E.) 185–218 (Academic, San Diego, 1996). 48. Onimaru, H., Arata, A. & Homma, I. Primary respiratory rhythm generator in the medulla of brainstem-spinal cord preparation from newborn rat. Brain Res. 455, 314–324 (1988). 49. Onimaru, H. & Homma, I. Whole cell recordings from respiratory neurons in the medulla of brainstem-spinal cord preparations isolated from newborn rats. Pflugers Arch. 420, 399–406 (1992).

517