Brain-specific BNIP-2-homology protein Caytaxin relocalises glutaminase to neurite terminals and reduces glutamate levels

Cayman cerebellar ataxia is a recessive congenital disorder associated with hypotonia, variable psychomotor retardation, cerebellar dysfunction like truncal ataxia and intention tremor, scoliosis and ocular abnormalities (Nystuen et al., 1996). Mutations in the ATCAY gene on human chromosome 19 are linked to Cayman ataxia, but its function and mechanism of regulation remains unknown (Bomar et al., 2003). The gene is exclusively expressed in adult and embryonic neural tissues (Bomar et al., 2003) and the cDNA, first isolated in our laboratory (GenBank accession number AY220297), encodes the protein BNIP-H or Caytaxin. BNIP-H (for BNIP-2 homology) shares 52% amino acid sequence identity (69% similarity) with BNIP-2, originally identified as a partner for the anti-apoptotic proteins Bcl-2 and the adenovirus E1B 19-kDa protein (Boyd et al., 1994). We subsequently showed that it contains a novel homophilic and heterophilic protein-protein interaction domain known as the BNIP-2 and Cdc42GAP homology (BCH) domain that also targets the small GTPase, Cdc42 (Low et al., 2000a; Low et al., 2000b). Transient expression of BNIP-2 led to Cdc42-dependent cell dynamics such as cell elongation and membrane protrusions (Zhou et al., 2005a) whereas another homologue, BNIP-Sα induces cell rounding and apoptosis through its intact BCH domain (Zhou et al., 2002) by displacing p50RhoGAP and facilitating RhoA activation by directly binding to them by unique motifs (Zhou et al., 2005b). The importance of the BCH domain was further exemplified by BPGAP1, a novel Rho GTPase-activating protein, that uses its BCH domain in concert with the adjacent proline-rich region to promote cell migration (Lua and Low, 2004) and Ras-MAPK signalling (Lua and Low, 2005). Interestingly, the two types of BNIP-H mutation found to be associated with Cayman ataxia were predicted to cause an amino acid change at the highly conserved position 301 (Ser to Arg) of the C-terminal BCH domain or a truncation of the BCH domain, respectively (Bomar et al., 2003). More recently, BNIP-H deficiency in rats was also found to cause generalised dystonia (Xiao and LeDoux, 2005), a neurological movement disorder characterised by sustained muscle contractions producing abnormal movements or postures. The deficiency was found to be caused by a 3′ long terminal repeat portion of an intracisternal A particle element inserted into exon 1 of Atcay (Xiao and LeDoux, 2005). In another study, BNIP-H was found to be polyubiquitylated by the ubiquitin E3 ligase CHIP (C-terminus of Hsc70-interacting protein) in vitro, suggesting that BNIP-H degradation could be triggered by CHIP (Grelle et al., 2005). These discoveries and its exclusive expression in neural tissues imply that BNIP-H is likely to play important role(s) for the correct function and/or development of the human central nervous system.

Glutamate is the most abundant excitatory neurotransmitter in the human brain. At the same time, it is the precursor for the inhibitory neurotransmitter γ-aminobutyric acid (GABA) (Petroff, 2002). The synthesis of the neurotransmitter glutamate can be catalysed by phosphate-activated glutaminase (PAG), a mitochondrial enzyme converting glutamine to stoichiometric amounts of glutamate and ammonia (Kvamme et al., 2000). Four isoforms of PAG have been described: Kidney-type glutaminase (KGA) is highly expressed in the heart, kidney, intestine, foetal liver and brain, whereas its splicing isoforms, glutaminase C (GAC) and M (GAM), are expressed in tissues other than the brain (Curthoys and Watford, 1995; Elgadi et al., 1999). Liver-type glutaminase (LGA) is synthesised from a different gene (Aledo et al., 2000), and is expressed in postnatal liver, pancreas and brain (Aledo et al., 2000; Gomez-Fabre et al., 2000; Smith and Watford, 1990). Interestingly, brain LGA is localised to the nucleus (Olalla et al., 2002). Active KGA is a heterotetramer consisting of three 66 kDa subunits and one 68 kDa subunit. Both subunits are generated from a single 74 kDa precursor via a 72 kDa intermediate (Perera et al., 1990; Shapiro et al., 1991). PAG plays an important role in the glutamine-glutamate cycle between neurons and glia (Hassel et al., 1997; Hertz et al., 1999; Magistretti et al., 1999): During synaptic transmission, glutamate is released into the synaptic cleft. To avoid continuous activation of glutamate receptors, it is removed from the extracellular space and transported into glia, where glutamate is converted into glutamine. Glutamine is then released and transported back into neurons, where it is a substrate for PAG to replenish the released glutamate stores. Besides mediating synaptic transmission, high concentrations of extracellular glutamate could cause neuronal cell death by prolonged activation of glutamate receptors, a process referred to as excitotoxicity (Coyle and Puttfarcken, 1993; Atlante et al., 2001). Excitotoxicity contributes to neuronal damage in conditions such as cardiac arrest, stroke and seizures (Rothman and Olney, 1987; Choi and Rothman, 1990; Meldrum and Garthwaite, 1990; Muir and Lees, 1995). Recently, KGA was shown to be involved in an increased production of extracellular glutamate following neuronal death (Newcomb et al., 1997). In two other studies, elevated PAG expression and activity were linked to schizophrenia, supporting the glutamate hypothesis that dysregulation of glutamatergic neurotransmission contributes to schizophrenia (Gluck et al., 2002; Bruneau et al., 2005). Furthermore, Choudary et al. (Choudary et al., 2005) had suggested that changes in the cortical glutamatergic and GABAergic signal transmission could play an important role in depression.

Here, we describe the identification of KGA as a binding partner for BNIP-H in the brain. We show that BNIP-H regulates the spatial distribution and activity of KGA. To our knowledge, this is the first report on the physical interaction and regulation of a neurotransmitter-producing enzyme by a brain-specific protein, thus providing a possible molecular basis for the aetiology of Cayman ataxia and generalised dystonia, and possibly other related neurological disorders. The significance of this is discussed.

To gain insights into the physiological function of BNIP-H, we used GST pull-down assays and matrix-assisted laser desorption/ionisation-time of flight (MALDI-TOF) mass spectrometry to identify its cellular binding partners. Whole-cell lysates were prepared from brains of adult female rats or cultured mammalian cells (Neuro2A and PC12) and incubated with immobilised GST-BNIP-H that was expressed and purified from E. coli. Proteins bound to BNIP-H were eluted and resolved by SDS-PAGE. Visualisation by silver staining revealed a protein band around 70 kDa from the brain tissue lysate (Fig. 1A, lane 3, labelled with asterisk) as well as from the cell culture lysates (Fig. 1B,C, labelled with asterisks). All bands were identified as kidney-type glutaminase (KGA; Fig. 1D). The identification of KGA was further confirmed by western blotting of the eluate with a KGA polyclonal antiserum (Fig. 1A, lanes 6) although no signal was readily observed with the amount of whole-cell lysate loaded. This result suggests a strong enrichment of KGA by BNIP-H (lane 7, compare with lanes 4 and 5). To confirm that KGA is a bona fide interacting partner in vivo under physiological conditions, we raised polyclonal antibodies against BNIP-H (see Materials and Methods) and used them to immunoprecipitate endogenous BNIP-H from the brain lysate. Western blot analysis revealed the presence of BNIP-H and KGA in the precipitate (Fig. 2A, compare lane 1 with lane 4) supporting the fact that both proteins form a physiological complex in the brain. This interaction is also deemed specific as no binding was observed for an unrelated protein (EEN/endophilin II, compare lanes 1 and 4). To determine the main region of BNIP-H that is important for such interaction, epitope-tagged KGA and different fragments of BNIP-H (Fig. 2C) were expressed in 293T cells and subjected to co-immunoprecipitation. First, HA-tagged BNIP-H was coimmunoprecipitated with FLAG-tagged KGA (Fig. 2B, lane 3), and in the reciprocal immunoprecipitation, HA-tagged KGA could be precipitated with FLAG-BNIP-H (Fig. 2B, lane 4), confirming the in vivo interaction between both proteins. Next, we showed that the N-terminal half of the protein (aa 1-190) did not participate in the binding whereas the C-terminal half that carried the BCH domain (aa 190-371) did (Fig. 2C). To map the potential binding site(s) within this domain, a series of truncation mutants were generated at the C-terminus. Our results showed that the fragment encompassing residues 1-206 (that still retained aa 191-206 of the most proximal part of the BCH domain) exhibited reduced binding (Fig. 2C, lane 6) whereas the next fragment in size (aa 1-235) interacted as strongly as the full-length BNIP-H. Further extension of the C-terminus did not improve the binding further. These results indicate that residues 191-235 on BNIP-H serve as a binding region for KGA. To further investigate whether this region would serve as the only binding site (Site I), a larger region of 189-287 was internally deleted and similarly tested for KGA binding in co-immunoprecipitation studies. Intriguingly, fragment without aa 189-287 still retained the binding capability (Fig. 2C, lane 5). Since the similar binding data were obtained from fragments 150-371 and 150-332 (indicating that residues 333-371 were not involved in the binding), we concluded that the region of 288-331 harbours another site (Site II) for binding to KGA. Therefore, it could be seen that the deletion of a single binding region of BNIP-H while retaining the other, either devoid of Site I (i.e. in fragment without aa 189-287) or Site II (represented by fragment 1-287 or 1-235), did not exhibit any major loss of interaction - unless both regions were absent (e.g. in fragment 1-190). On the other hand, the presence of both sites did not enhance the total binding capacity. These results therefore imply the collaborative and yet mutually regulatory nature of two primary KGA-binding sites on BNIP-H.

To delineate the reciprocal binding site(s) on KGA, various deletion constructs of FLAG-tagged KGA were co-expressed with HA-BNIP-H and tested for their binding inside the cells (Fig. 2D). Unlike the enzymatic domain of KGA (Region B), the flanking regions A and C were not involved in the binding to BNIP-H at all. To pinpoint more precise interaction sites, region B was further sub-divided and expressed as discrete subregions of B1, B2, B3 or their composites (B12 or B23). Our results clearly showed that Region B2 (aa 269-408) and Region B3 (aa 402-547) of the enzymatic domain of KGA were crucial for its interaction with BNIP-H (Fig. 2D). Finally, to determine whether BNIP-H-KGA interaction was primarily mediated through their direct binding and not by third party protein(s), FLAG-tagged KGA that harboured the BNIP-H binding region was produced by in vitro transcription and translation (Fig. 2E, lane 5), and then subjected to precipitation assays with purified GST-BNIP-H, GST-BNIP-2 or GST-BNIP-S fusion proteins. Western blot analysis showed that only the GST-BNIP-H bound to FLAG-tagged KGA (Fig. 2E, lane 3), but no binding was observed for GST-BNIP-2 or GST-BNIP-S (Fig. 2E, lane 1 and 2, respectively). A control western blot did not show nonspecific binding of FLAG antibody to the GST fusion proteins (data not shown). Taken together, these results confirm the identification of KGA as a novel and specific binding partner of BNIP-H, the brain-specific member of the BNIP-2 family proteins, thus highlighting their possible functional relevance in brain development or neurophysiology.

To shed light on the possible physiological roles of BNIP-H, we next determined the expression profiles of BNIP-H protein in mouse tissues and different regions of the brain. The BNIP-H antibody recognised a specific protein at around 60 kDa in the brain lysate but not in other tissues (Fig. 3A). This result confirmed the restricted neural distribution of BNIP-H mRNA (Bomar et al., 2003; Xiao and LeDoux, 2005). More significantly, this antibody allowed us to detect endogenous BNIP-H expression in different domains of the brain as performed by immunohistochemistry (IHC) with mouse brain sections (Fig. 3). BNIP-H expression was observed in major parts of the brain, including the cortex (Fig. 3B) but was more enriched in the hippocampus (Fig. 3B), cerebellar cortex (Fig. 3C), deep cerebellar nuclei (Fig. 3D) and pontine nuclei (Fig. 3E). The results are in agreement with previous studies where these brain structures (except for the deep cerebellar nuclei) were intensively labelled either in ISH experiments targeting the glutaminase mRNA (as for the hippocampus) (Najlerahim et al., 1990) or with a monoclonal antibody against glutaminase (as for the cerebellar cortex and the pontine nuclei) (Kaneko et al., 1987). Interestingly, the granule cell layer of the cerebellar cortex showed a staining pattern of small clusters of grains suggesting a possible BNIP-H localisation at the axon terminals (Fig. 3C). The staining pattern also shows striking similarity to that of glutaminase in rat brain slices (Kaneko et al., 1987). The CA3 of the hippocampus presented dot-like staining, indicating that BNIP-H is localised in the neuropil and synapses. No cell bodies were recognisable (Fig. 3B). Besides a fine punctate staining pattern in the deep cerebellar nuclei and pontine nuclei, cell bodies within these regions were intensively labelled with the BNIP-H antibody (Fig. 3D,E). To further investigate the possible role of BNIP-H in neuronal function, we used the embryonic carcinoma cell line P19 to further show that BNIP-H was expressed in neurons. These P19 cells when treated with retinoic acid could differentiate into neurons and glial cells (MacPherson and McBurney, 1995). BNIP-H was not detected in undifferentiated P19 cells but was expressed after nine days of differentiation as shown by western blotting (Fig. 4A). By contrast, glutaminase immunoreactivity was detected in cell lysates of both undifferentiated as well as differentiated P19 cells (Fig. 4A). Several bands were detected with a polyclonal antiserum, suggesting the presence of constitutively expressed multiple isoforms. The KGA form that migrated at the expected size is indicated by the arrow (Fig. 4A). Immunofluorescence studies further showed the colocalisation of BNIP-H and glutaminase with the neuron-specific marker neurofilament-160 within the cell bodies and neurites (Fig. 4B,C). These results further support the importance of BNIP-H in carrying out specific function(s) in differentiating neurons. Based on the labelling pattern and intensity, our results strongly suggest that BNIP-H could play an important role in the hippocampus and the cerebellum, possibly in the process of neurotransmission at synapses.

To correlate the protein expression profiles of BNIP-H with its gene expression, we conducted parallel in situ hybridisation (ISH) histochemistry against its mRNA in rat brain slices. Our results showed a similar staining pattern to that observed with IHC (Fig. 5A,B). Furthermore, the expression of BNIP-H in the spinal cord was detected (Fig. 5C). Interestingly, some differences were noted: CA1, CA2 and the dentate gyrus of the hippocampus were intensely labelled by ISH histochemistry but were not stained by IHC. The Purkinje cell layer was also strongly labelled by ISH histochemistry (Fig. 5B inset), but the labelling was relatively weak with BNIP-H antibody. This might be due to the fact that BNIP-H expression could be under control at both transcriptional and translational levels in specific regions of the brain. Parallel experiments with a probe against juxtanodin mRNA revealed a distinct staining pattern (Zhang et al., 2005), indicating stringent hybridisation conditions.

To gain further insight into the possible cellular role of BNIP-H interaction with KGA, we used rat pheochromocytoma PC12 cells that acquire a neuronal phenotype when stimulated with nerve growth factor (NGF; Fig. 6A). Cells were transfected with full-length BNIP-H or the fragment aa 150-371 of BNIP-H, in the presence or absence of KGA. Confocal immunofluorescence studies revealed that full-length BNIP-H or the fragment 150-371 that harbours the BCH domain (Fig. 2C), were distributed throughout the cytoplasm and they were enriched in neurite terminals (Fig. 6B,C). Expression of KGA alone however showed a perinuclear distribution, which could be co-stained with a mitochondria-specific dye (Fig. 6B). These patterns were observed in NGF-stimulated cells as well as in unstimulated cells (data not shown). Interestingly, cotransfection of both proteins did not change the BNIP-H-staining pattern. Instead, this led to a redistribution of KGA that colocalised with BNIP-H, especially at neurite terminals (Fig. 6C). Similar results were also obtained with unstimulated PC12 cells (data not shown). Such a redistribution pattern was also observed when we cotransfected the fragment 150-371 instead of the full length BNIP-H (Fig. 6C). This result is therefore in strong agreement with our binding studies showing that a certain region of the BCH domain is important for KGA binding (Fig. 2C). The staining pattern of BNIP-H was different from that observed for mitochondria (Fig. 6B) indicating that BNIP-H has no direct effect on mitochondria localisation. As a control and to show specificity, we used an N-terminal BNIP-H fragment (aa 1-190) that does not bind KGA (Fig. 2C). Overexpression of this fragment lacking the BCH domain did not alter the distribution of KGA (Fig. 6D). As another control we co-expressed human elongation factor 1A1 (EF1A1) with KGA in PC12 cells. Similar to that observed for the BNIP-H fragment 1-190, co-expression of EF1A1 did not alter the localisation of KGA (data not shown). Finally, triple labelling of co-expressed BNIP-H and KGA, with the mitochondria revealed a distinct staining pattern of KGA and mitochondria, which was especially apparent at the endings of the neurites (Fig. 6E, see arrows) indicating that KGA was specifically redistributed by BNIP-H independently of and away from mitochondria. Our results therefore strongly suggest that the redistribution of KGA is regulated by BNIP-H in a manner independent of NGF and mitochondria, but requiring certain motif(s) within the BCH domain of BNIP-H.

KGA is a phosphate-dependent glutaminase possibly responsible also for the production of the neurotransmitter glutamate. Our results suggest that spatial regulation of KGA by BNIP-H might influence the steady-state levels of glutamate. However, as a result of low transfection efficiency of PC12 and the inability to generate stable cells with singly or doubly transfected BNIP-H and/or KGA in these cells, we opted for 293T epithelial cells that were efficiently transfected for optimal biochemical assays. Aliquots of lysates from single- or double-transfected cells were analysed for KGA expression by western blotting. The rest of the lysate was deproteinated and the total amount of glutamate intracellularly as well as in the medium was determined enzymatically (see Materials and Methods). Values were corrected for the endogenous amount of glutamate in mock-transfected samples (Fig. 7A). Cells expressing KGA showed a marked increase in the total glutamate level, either intracellularly (42.0±2.3 nmol, mean ± s.d., P<0.0001) or as released into the medium (130.6±20.8 nmol, P<0.0001), when compared with the mock-transfected samples. This result showed that exogenously expressed KGA is active inside the cells. By contrast, expression of BNIP-H had no significant effect on the basal glutamate levels. However, when compared with those elicited by KGA alone, co-expression of KGA with BNIP-H led to a significant reduction in the glutamate levels either intracellularly (8.0±2.0 nmol, P<0.0001) or in the medium (71.6±45.9 nmol, P=0.02). By contrast and as a control, co-expressed EF1A1 that did not interact or redistribute KGA as observed before (Fig. 2C and data not shown) had no effect on the production of glutamate elicited by KGA (42.9±3.4 nmol, P=0.59; for the intracellular level and 126.8±9.8 nmol, P=0.84; for the medium). These results strongly suggest that BNIP-H either directly or indirectly downregulates the steady-state levels of glutamate. To show that BNIP-H inhibited glutaminase activity through its binding to KGA, we performed an enzyme assay by reconstituting the BNIP-H-KGA complex in vitro. We used bacterially expressed and purified GST-fusion of the full-length (FL) BNIP-H and incubated it with lysates from 293T cells that overexpressed FLAG-tagged KGA. As controls, GST-BNIP-H (aa 1-190), which does not bind KGA, or the GST alone, was used. After pre-incubation of lysate with equal amounts of these GST constructs, glutamine was added to the sample for the times indicated and the total amount of glutamate was determined as described in the Materials and Methods. Fig. 7B shows that the rate of glutamate production was greatly reduced in the presence of GST-BNIP-H FL but not with GST-BNIP-H (aa 1-190). Calculation of the close-linear rates at 5 minutes indicated that the rate of glutamate production in the presence of GST-BNIP-H FL was only 14±0.1% (P=0.005) of that in the sample with GST alone or with the GST-BNIP-H (aa 1-190). All these results indicate that BNIP-H downregulates the steady-state levels of glutamate in the cells by directly binding and inhibiting the glutaminase activity of KGA. This could in turn modulate the homeostasis of glutamate necessary for proper neuronal functions. The significance of this is discussed.

We have identified kidney-type glutaminase as an in vivo binding partner for BNIP-H in the brain. BNIP-H expression was detected in all parts of the brain with high expression in the hippocampus and cerebellum. Furthermore, we detected BNIP-H in differentiating neurons. Overexpression of BNIP-H in PC12 cells relocalised KGA from the cell body to neurite terminals. It also reduced the steady-state levels of glutamate by inhibiting the glutaminase activity. Thomas et al. (Thomas et al., 1989) showed that the upregulation of glutaminase expression is parallel to the formation of neurites and synapses of cerebellar granule cells between days three and ten in vitro. Several groups have shown the localisation of KGA in axon terminals (Aoki et al., 1991; Laake et al., 1999) where its function is believed to be the synthesis of the neurotransmitter glutamate. Therefore, the absence of KGA or lowering of its enzymatic activity in synapses could lead to reduced levels of glutamate in the axon terminals. In Drosophila, increased or reduced levels of glutamate by genetic manipulations have recently been shown to inversely affect the postsynaptic receptor field size (Featherstone et al., 2002). Furthermore, pharmacological interference blocking the degradation of γ-aminobutyric acid (GABA) caused an increase in the synaptic efficacy (Engel et al., 2001). It is therefore possible that in the absence or malfunction of BNIP-H, the misdisposition of KGA would result in a reduction in local concentrations of glutamate in the axon terminals. Likewise, this could result in an elevated glutaminase activity that leads to an overall rise in glutamate levels, which can affect the efficacy of neurotransmission.

It had previously been shown that experimentally enhanced extracellular glutamate levels led to cell death of mature cortical neurons in cell culture (Choi et al., 1987). Furthermore, it was shown that knocking down glial glutamate transporters important for glutamate clearance could also result in elevated extracellular glutamate levels and neurodegeneration characteristic of neurotoxicity (Rothstein et al., 1996). Consistently, changes in the gene expression of a key glutamate-metabolising enzyme, the L-glutamate-ammonia ligase and glutamate transporters could also modulate the levels of extracellular glutamate. This could in turn lead to the neurotoxicity associated with depression (Choudary et al., 2005). In patients suffering from Cayman ataxia an apparent cerebellar hypoplasia was reported (Bomar et al., 2003), further implicating that the absence of functional BNIP-H, presumably leading to an accumulation of glutamate in the cell bodies of neurons, could lead to abnormal neuronal growth and/or neurotoxicity induced by glutamate in the extracellular space.

In addition to human Cayman ataxia, various forms of mutation in the BNIP-H gene have now been identified in three mice mutants (jittery, hesitant and sidewinder) and in one rat mutant model for dystonia (Bomar et al., 2003; Xiao and LeDoux, 2005). However, unlike human Cayman ataxia, the mutations in the dystonic rat model and those in the jittery and sidewinder mouse mutants are generally lethal as they probably cause more drastic changes in BNIP-H structure and quantity than the mutations observed in humans (Xiao and LeDoux, 2005). Interestingly, cerebellectomy could partially rescue the phenotype and prolong the lifespan of the dystonic rat (LeDoux et al., 1993), supporting the notion that BNIP-H does play an important role in the correct functioning of the cerebellum for movement control. Indeed, induction of dystonia upon injection of low doses of kainic acid into cerebellar vermis of mice depends on glutamatergic activation (Pizoli et al., 2002). Furthermore, the authors suggested that the cerebellar cortex plays a crucial role in this model of dystonia (Pizoli et al., 2002). These findings are consistent with our data of strong BNIP-H expression in the cerebellar cortex as detected by in situ hybridisation and immunohistochemistry (Figs 3 and 5). All these results, related to either neurotransmission or neurotoxicity, seem to point to a plausible circuitry network of control between the glutaminase, glutamate and BNIP-H. Dysregulation of such checkpoint, such as the loss of BNIP-H, could underlie the molecular basis and the etiology of cerebellar ataxia, dystonia, and possibly other related neurological disorders.

Finding the precise mode of physical interaction between BNIP-H and glutaminase is important as it allows better understanding on their functional control. We showed that such interaction is mediated by their direct binding in a constitutive manner. Our current work also shows that it involves at least two regions within the BCH domain of BNIP-H - Site I (aa 191-235) and Site II (aa 288-331) - targeting the more distal parts of the enzymatic moiety of KGA (aa 269-547). Interestingly, Site II of the BCH domain encompasses the homozygous Ser301Arg mutation associated with Cayman ataxia (Bomar et al., 2003). However, as expected from the nature of dual-site interaction, this single-point mutation alone does not affect the binding of BNIP-H to KGA (our unpublished data). On the other hand, the splicing mutation in intron 9 of the BNIP-H gene is predicted to cause a truncation of the conserved BCH domain (Bomar et al., 2003). It remains unclear if indeed this truncated form is ever expressed in patients suffering from Cayman ataxia. Further deletion studies within these regions failed to conveniently pinpoint any specific or subtle motifs, indicating that the nature of their interaction is likely to be complex. The answer to this should await high-resolution structural determination. Nonetheless, our current findings on the absolute importance of a functional BCH domain in binding and regulating KGA activity, hence glutamate homeostasis, seem to be a plausible model for further investigation on the molecular basis for this and other related neurological disorders.

In addition, our binding studies have further exemplified the repertoire of BCH domain as a dynamic protein-protein interaction device. It is now evident that different members use specific motifs to mediate their homophilic or heterophilic interactions (Low et al., 1999; Low et al., 2000b; Zhou et al., 2002; Shang et al., 2003), and also engaging Rho subfamily small GTPases to elicit cell protrusions (e.g. BNIP-2 and BPGAP1) (Zhou et al., 2005a; Shang et al., 2003) or cell rounding during apoptosis (e.g. BNIP-Sα) (Zhou et al., 2002; Zhou et al., 2005b). It has been postulated that part of the BCH domain of BNIP-H that weakly resembles the CRAL/TRIO domain (a domain that binds small lipophilic molecules) might be important for targeting specific ligand for its normal function (Bomar et al., 2003). However, this hypothesis has never been tested. Our data on the other hand revealed the unexpected role of the BCH domain in promoting intracellular trafficking and regulating the activity of a key enzyme that is important for neuronal function. Whether or not this property of trafficking is linked to its ability of targeting specific small GTPases, its homophilic or heterophilic binding or lipid binding remains to be seen. Currently, our data show that BNIP-H, unlike BNIP-2 and BNIP-S, does not interact with members of the Rho subfamily GTPases such as Cdc42, RhoA and Rac1 (our unpublished data), raising the possibility that other types of GTPases might be uniquely involved in the BNIP-H signalling pathway. A more detailed investigation is currently underway in our laboratory to probe the mechanistic control of such trafficking, in particularly the possible involvement of cytoskeletal components.

At present, we also show that retinoic-acid-induced differentiation of P19 cells was accompanied by the activation of BNIP-H gene expression in developing and differentiating neurons, suggesting that BNIP-H expression is also regulated at cellular level during the early stages of brain development. This is consistent with the BNIP-H mRNA being developmentally regulated in the hippocampus and cerebellum (Xiao and LeDoux, 2005). Interestingly, the Purkinje cell layer known for its GABAergic neurons appeared to express BNIP-H although glutaminase is known to be absent in Purkinje cells of the adult rodent brain. This implies that BNIP-H could have functions other than just the relocalisation of glutaminase in certain cellular systems or developmental stages.

Taken together, we have identified and confirmed KGA as a novel target of BNIP-H in the brain where their interaction could play an important role in reducing the steady state levels of glutamate in the cells. Based on the biochemistry of glutamate and the pathophysiological outcomes arising from the loss of BNIP-H in human and mice or rat models of ataxia and dystonia as discussed above, our findings could present a novel paradigm for regulating the homeostasis of glutamate synthesis important for proper neurotransmission and/or neuronal cell growth. The immediate challenge now is to determine both the in vivo levels of glutaminase activity and glutamate in the neurons of the mouse or rat models with defective BNIP-H function, as well as their effects on neurotransmission and neurotoxicity. Following this, they should be subjected to genetic or biochemical means of lowering the activity of glutaminase (e.g. by conditional knockout of KGA or inhibitory peptide from BNIP-H) or to `dampen' the excessive glutamate from uncontrolled glutamatergic activation or neuroexcitotoxicity.

Brains from adult female rats or cultured mammalian cells (Neuro2A and PC12) were homogenised in lysis buffer (50 mM HEPES pH 7.4, 100 mM NaCl, 10 mM MgCl₂, 5 mM EDTA, 10% glycerol, 1% Triton X-100, 5 mM sodium orthovanadate, 5 mM glycerol 2-phosphate and a mixture of protease inhibitors) using a Dounce tissue grinder (Wheaton) with Pestle A and B. The homogenate was centrifuged and the supernatant was carefully collected. Glutathione-Sepharose beads (Amersham Biosciences) coated with 100 μg of bacterially expressed GST or GST-BNIP-H were mixed with the lysate and incubated overnight at 4°C with gentle shaking. As a control, glutathione-Sepharose beads coated with 100 μg GST-BNIP-H were mixed with lysis buffer and processed the same way. The beads were then extensively washed with lysis buffer. The bound proteins were eluted with lysis buffer containing 4 M urea, resolved by SDS-PAGE and visualised by silver staining (Bio-Rad). Protein bands excised from SDS-polyacrylamide gels were reduced, alkylated, and then in-gel digested with trypsin (Shevchenko et al., 1996). The extracted peptides were analysed on a matrix-assisted laser desorption/ionisation mass spectrometer equipped with a time-of-flight analyser (MALDI-TOF; Voyager STR BioSpectrometry workstation, Applied System). The obtained mass fingerprints were then used to perform a MS-Fit search on NCBI databases to identify the protein.

The FLAG and HA expression vectors were from Ed Manser (Institute for Molecular and Cell Biology, Singapore). The KGA cDNA was a generous gift from Norman P. Curthoys (Colorado State University, Fort Collins; GenBank accession number AF327434) (Holcomb et al., 2000). KGA full-length cDNA or various fragments were generated through polymerase chain reaction (PCR) with appropriate primers containing either a BamHI or XhoI restriction site for cloning into the FLAG- and HA-pXJ40 expression vectors. Full-length BNIP-H or its fragments were generated in the same way using BamHI and XhoI restriction sites and cloned into FLAG- and HA-pXJ40 expression vectors. Constructs were verified through sequencing and propagated in E. coli strains XL1-blue and DH5α.

Polyclonal antibodies against a fusion protein consisting of GST and full-length human BNIP-H were generated in New Zealand white rabbits by subcutaneous injections of approximately 1 mg of antigen. For the first injection the antigen was mixed with Freund's complete adjuvant, subsequently incomplete adjuvant (both from Sigma) was used. After three booster injections every 3 weeks, blood was collected and serum was prepared. Anti-BNIP-H antibodies were affinity purified with a fusion protein consisting of thioredoxin and full-length human BNIP-H expressed in E. coli BL21 and transferred onto a PVDF membrane. Anti-BNIP-H antibodies were eluted with 100 mM glycine (pH 2.0); the solution was neutralised with 1/10 volume 1.5 M Tris-HCl (pH 8.0).

Human 293T cells were grown in RPMI 1640 medium (Hyclone) supplemented with 10% (v/v) fetal bovine serum, 10 mM HEPES, 2 mM L-glutamine, 100 U/ml penicillin, and 100 μg/ml streptomycin (all from Hyclone). Cells at 60-80% confluence in six-well plates were transfected with 1-1.75 μg plasmid using Fugene 6 cationic lipid (Roche), according to the manufacturer's instructions. PC12 cells were cultured in Dulbecco's Modified Eagle's Medium (DMEM) supplemented with 4500 mg glucose, 10 mM HEPES, 5% fetal bovine serum (all from Hyclone) and 10% horse serum (Gibco). PC12 cells on poly-D-lysine (Sigma) coated surfaces were transfected with Lipofectamine 2000 Reagent (Invitrogen), according to the manufacturer's instructions. Differentiation was induced with 40 ng/ml nerve growth factor (NGF) in the presence of 0.5% serum. P19 cells were cultured in alpha-modified Minimal Essential Medium supplemented with 7.5% bovine serum (both from Gibco) and 2.5% fetal bovine serum (Hyclone) and differentiated essentially as described (Jones-Villeneuve et al., 1982; MacPherson and McBurney, 1995). Briefly, neuronal differentiation was induced with 0.5 μM retinoic acid (RA) in serum-supplemented medium for 4-5 days in bacterial grade cell culture dishes. Cell aggregates were than plated into cell culture dishes containing normal growth medium without RA and further differentiated for 5-6 days. Cells were treated with cytosine arabinoside (5 μg/ml) for several days to inhibit the growth of non-neuronal cells. For confocal immunofluorescence studies, P19 cells were grown on gelatin-coated coverslips.

Transfected cells were lysed in 200 μl lysis buffer (50 mM Tris-HCl pH 7.3, 150 mM NaCl, 0.75 mM EDTA, 1% sodium deoxycholate, 1% Triton-X-100, 0.2% sodium fluoride, 25 mM glycerol 2-phosphate, 5 mM sodium orthovanadate and a mixture of protease inhibitors) per well. Aliquots were either directly analysed by western blotting or were used for protein-binding studies. For use in CoIP, lysates were incubated with anti-FLAG antibody conjugated to agarose beads (Sigma) at 4°C overnight. The beads were extensively washed with lysis buffer and analysed by western blotting with monoclonal and polyclonal anti-FLAG antibodies (Sigma) and anti-HA antibody (Zymed). To show direct binding, purified GST fusion proteins (5 μg) of BNIP-H, BNIP-2 and BNIP-S immobilised on glutathione-Sepharose (Amersham Biosciences) were incubated with FLAG-tagged KGA (aa 134-669) produced by in vitro transcription and translation (Promega) at 4°C overnight. After washing with lysis buffer, samples were analysed by western blotting with an anti-FLAG antibody (Sigma). For CoIP of endogenous proteins, mouse brain was homogenised in lysis buffer with a Dounce tissue grinder (Wheaton). After centrifugation, the lysate was precleared with protein A/G-agarose (Santa Cruz Biotechnology) for 5 hours at 4°C. Anti-BNIP-H antibody or pre-immune serum was added to the lysate and samples were incubated at 4°C overnight. Immunocomplexes were captured by adding protein A/G-agarose for 3.5 hours. After washing with lysis buffer, samples were analysed by western blotting with various antibodies. The polyclonal glutaminase antiserum (Curthoys et al., 1976) was a generous gift from Norman P. Curthoys (Colorado State University, Fort Collins).

Adult rats were deeply anesthetised and transcardially perfused with saline followed by 3% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4). The brain and the cervical spinal cord were dissected, postfixed, and sectioned into cryostat sections. Digoxigenin-labelled riboprobes were synthesised by in vitro transcription with rat BNIP-H cDNA (nucleotides 334-1452, GenBank accession number XM_343154.1) inserted into pGEM-T Easy (Promega) as a template. ISH histochemistry (probe concentration of 0.2 μg/ml) followed a protocol previously described (Liang et al., 2000).

IHC was carried out with the Vectastain Elite ABC-Kit (Vector Laboratories) according to the manufacturer's instructions. Briefly, paraffin-embedded mouse brain sections were deparaffinised and hydrated through xylenes and a graded alcohol series, respectively. After rinsing with water, sections were cooked in 0.1 M citric acid (pH 6.1) for 10 minutes and allowed to cool down to room temperature. Sections were washed with PBS and placed in 0.3% H₂O₂ to quench endogenous peroxidase activity, and washed again. Sections were incubated with normal blocking serum for 1 hour and then with anti-BNIP-H antibody overnight. After washing, sections were incubated for 1 hour with biotinylated secondary antibody followed by incubation with a preformed complex of avidin and biotinylated peroxidase. Sections were incubated in peroxidase substrate solution (diaminobenzidine tetrahydrochloride, DAB) until desired stain intensity developed, rinsed with water, cleared and mounted. Pictures were taken with an AxioCam MRc5 camera fixed on an Axioskop 2 mot plus microscope (both from Zeiss) and Axiovision 4 software (Zeiss).

PC12 cells grown on sterilised poly-D-lysine-coated glass coverslips were washed twice with PBS and fixed with 3.7% formaldehyde for 15 minutes at 37°C. Fixed cells were washed twice with PBS and permeabilised with 0.2% Triton X-100 (BioRad) in PBS for 15 minutes at room temperature. Blocking was carried out with 2% bovine serum albumin and 7% fetal bovine serum in PBS for 30 to 60 minutes at room temperature. Cells were incubated with anti-FLAG (Sigma) and anti-HA (Zymed) antibodies in blocking solution. Samples were washed three times with 0.1% Triton X-100-containing PBS before incubation with Rhodamine-conjugated goat anti-mouse IgG (Chemicon), Fluorescein isothiocyanate (FITC)-conjugated goat anti-rabbit IgG (Jackson ImmunoResearch Laboratories), Alexa Fluor 488-conjugated donkey anti-mouse IgG, Alexa Fluor 594-conjugated goat anti-rabbit IgG or Pacific Blue-conjugated goat anti rabbit IgG (all Molecular Probes). Mitochondria were stained with MitoTracker Orange CMTMRos (Molecular Probes) according to manufacturer's instructions. P19 cells were processed the same way except that cells were postfixed with 3.7% formaldehyde at 4°C overnight. The neurofilament-160 antibody was from Sigma. Pictures were taken with a confocal microscope (Zeiss).

The total amount of glutamate in the cell lysate and cell culture medium was determined enzymatically essentially as described (Lund, 1986). Transfected cells were washed twice with PBS and harvested with cold lysis buffer (50 mM HEPES pH 7.4, 150 mM NaCl, 10 mM MgCl₂, 5 mM EDTA, 10% glycerol, 1% Triton X-100, 5 mM sodium orthovanadate, 5 mM glycerol 2-phosphate and a mixture of protease inhibitors). After centrifugation, an aliquot of the lysate was analysed by western blotting for equal expression of FLAG-KGA in double transfected cells and expression of HA-BNIP-H and FLAG-EF1A1, respectively. The rest of the lysate was deproteinated with 1/20 volume of 100% trichloroacetic acid (TCA) and centrifuged for 5 minutes. The supernatant was immediately neutralised with potassium hydroxide. 20 μl of the deproteinated lysate in a final volume of 200 μl were incubated with 80 mM Tris-acetate pH 9.4, 200 mM hydrazine, 0.25 mM ADP, 2 mM NAD and 2.2 U glutamate dehydrogenase for 40 minutes at room temperature. The absorbance at 340 nm was measured with an ELISA reader (SpectraMax 340, Molecular Devices) to evaluate the conversion of NAD⁺ to NADH, and thus the amount of oxidised glutamate. The measurement was repeated after 5 minutes to confirm completeness of enzymatic conversion. The glutamate amount in the medium was determined the same way but without trichloroacetic acid precipitation. Analysis of variance was carried out separately for each group of values (lysate, medium, total, n=3) with the Newman-Keuls multiple range test and was calculated with StatsDirect statistical software. For the glutaminase enzyme assay GST-tagged BNIP-H FL, GST-BNIP-H (aa 1-190) and GST were exogenously expressed in E. coli, purified and eluted from glutathione-Sepharose beads with 10 mM glutathione. Lysates from 293T cells containing overexpressed FLAG-tagged KGA were separately incubated with equal amounts (5 μg) of GST fusion proteins or GST for 1 hour at room temperature. After incubation 20 mM glutamine was added to the sample for 0, 2, 5 and 10 minutes. Samples were than TCA precipitated and the total amount of glutamate was determined as described above. The results presented are from three independent experiments. Statistical significant difference was analyzed with Student's t-test (paired, two-tailed).

We thank Shashikant Joshi and Wang Xian Hui at the Protein and Proteomics Centre for their technical advice and help on MALDI-TOF analyses, and the National University Medical Institutes for use of confocal microscopy units. We also thank Dr Norman P. Curthoys (Colorado State University, Fort Collins) for his generous gifts of glutaminase cDNA and antibody. This work was supported by the Young Investigator Award to B.C.L. from the Biomedical Research Council of Singapore.