Neuron-microglia crosstalk up-regulates neuronal FGF-2 expression which mediates neuroprotection against excitotoxicity via JNK1/2


Address correspondence and reprint requests to Sukalyan Chatterjee, Centro Biologia Desenvolvimento, Instituto Gulbenkian de Ciência, Rua da Quinta Grande 6, Oeiras 2780-156, Portugal.


Glial cells and neurons are in constant reciprocal signalling both under physiological and neuropathological conditions. Microglial activation is often associated with neuronal death during inflammation of the CNS, although microglial cells are also known to exert a neuroprotective role. In this work, we investigated the interplay between cerebellar granule neurons (CGN) and microglia in the perspective of CGN survival to an excitotoxic stimulus, quinolinic acid (QA), a catabolite of the tryptophan degradation pathway. We observed that CGN succumb to QA challenge via extracellular signal regulated kinase 1 and 2 (ERK) activation. Our data with transgenic mice expressing the natural inhibitor of calpains, calpastatin, indicate that together with cathepsins they mediate QA-induced toxicity acting downstream of the mitogen-activated protein kinase kinase-ERK pathway. Microglial cells are not only resistant to QA but can rescue neurons from QA-mediated toxicity when they are mixed in culture with neurons or by using mixed culture-conditioned medium (MCCM). This effect is mediated via fibroblast growth factor-2 (FGF-2) present in MCCM. FGF-2 is transcriptionally up-regulated in neurons and secreted in the MCCM as a result of neuron-microglia crosstalk. The neuroprotection is associated with the retention of cathepsins in the lysosomes and with transactivation of inducible heat-shock protein 70 downstream of FGF-2. Furthermore, FGF-2 upon release by neurons activates c-jun N-terminal kinase 1 and 2 pathway which also contributes to neuronal survival. We suggest that FGF-2 plays a pivotal role in neuroprotection against QA as an outcome of neuron-microglia interaction.

Abbreviations used

cerebellar granule neurons




extracellular signal regulated kinase 1 and 2


fibroblast growth factor-2


green fluorescent protein


inducible heat-shock protein 70


c-jun N-terminal kinase 1 and 2


mitogen-activated protein kinase


mixed culture-conditioned medium

MEK 1/2

MAPK kinase 1 and 2


nerve growth factor


phosphate-buffered saline


propidium iodide


quinolinic acid

The steps mediating cell death and its consequences are better understood in development than in disease. Cellular homeostasis is crucial in normal physiological condition, the breach of which, can lead to pathology (reviewed in Morimoto 1998). This homeostasis is achieved by co-ordinated intra and intercellular crosstalk involving survival factors, such as growth factors and the family of heat-shock responsive proteins (Hsp) and/or the apoptotic machinery. Growth factors are recognized to have a neuroprotective role against metabolic insults (Mattson et al. 1989; Freese et al. 1992; Fernandez-Sanchez and Novelli 1993), neuronal injury, pathological conditions such as ischaemia (Boniece and Wagner 1993) and in neurodegenerative diseases (Mark et al. 1997; Wang et al. 2004).

Glial cells and neurons are in constant reciprocal signalling both under physiological and neuropathological conditions. Microglia, resident macrophages of the brain, are able to trigger inflammatory responses in the CNS, contributing to the neurotoxicity observed in some neurodegenerative diseases (Gonzalez-Scarano and Baltuch 1999). On the other hand, in response to neuronal injury, microglia secrete neurotrophic factors (Shimojo et al. 1991; Elin Lehrmann et al. 1998; Nakajima et al. 1998) that support the remaining healthy neurons.

Quinolinic acid (QA) is catabolite of the tryptophan–kynurenine metabolism and a potent endogenous NMDA receptor agonist, which can cause neurotoxicity both in vivo and in vitro (Schwarcz et al. 1983; Beal et al. 1986; Kerr et al. 1995). It has been shown that microglia when activated secrete this neurotoxin (Espey et al. 1997). Although details of the physiological role and the molecular characterization of QA remain to be elucidated, it has been reported that this metabolite can generate reactive oxygen species, induce lipid peroxidation (Stone et al. 2003) and cause mitochondrial dysfunction (Stone et al. 2003). Elevated concentrations of this metabolite have been proposed to cause cell death and proposed to be an underlying cause for Huntington’s disease (Beal et al. 1986; Stone 1993), human immunodeficiency virus-associated dementia (Heyes et al. 1992; Stone 1993), Alzheimer’s disease (Guillemin and Brew 2002), epilepsy (Tavares et al. 2005) and cerebral malaria (Dobbie et al. 2000).

In this study, we investigated the effect of QA on cerebellar granule neurons (CGN) independently and in the perspective of a mixed culture with microglia as these two cell types interact in the CNS both under physiological and neuropathological conditions. We observed that QA-induced cell death occurs in neurons through an extracellular signal regulated kinase (ERK)-dependent signalling mechanism involving lysosomal protease activity. Our findings show the neuroprotection conferred by mixed culture-conditioned medium (MCCM) against QA-induced cell death is associated with the increased expression of neuronal fibroblast growth factor-2 (FGF-2) and inducible heat-shock protein (Hsp70i) as a result of neuron-microglia crosstalk. We propose that microglial cells up-regulate neuronal basic FGF-2 which protects them from QA via c-jun N-terminal kinase 1 and 2 (JNK 1/2) activation and Hsp70i induction.

Material and methods


C57/Bl6 and Balb/C mice were bred and kept in sterile conditions in our animal house facilities. Transgenic mice over-expressing human calpastatin (hCAST Tg) were created in Takaoni C. Saido’s laboratory in Riken Brain Science Institute, Tokyo, Japan (Takano et al. 2005).

Primary microglial cell cultures

All reagents used in the preparation of the primary cultures except those specifically mentioned were purchased from Invitrogen (Carlsbad, CA, USA). Neonatal microglial cultures were prepared from newborn Balb/C or C57/Bl6 mice, as previously described (Santambrogio et al. 2001), with some modifications. Briefly, after removal of the meninges, brains were mechanically disrupted. Cells were maintained at 37°C in 5% CO2 atmosphere and seeded in Dulbecco’s Modified Eagle Medium with 4500 mg/mL glucose and 1%l-glutamine containing 10% heat-inactivated fetal bovine serum, 5 μg/mL insulin (Sigma, Sintra, Portugal), 2.0 mg/mL l-glucose (Sigma) and 0.25 ng/mL granulocyte-macrophage colony-stimulating factor (Peprotech, London, UK) for 14–16 days. Microglial cells were maintained and were harvested from the mixed primary glial cultures by 3 h shaking and then collected by centrifugation. Microglia obtained in the suspension were plated at a density of 4 × 105 cells in 24-well plate or 2 × 106 cells in 6-well plate for 24 h.

Primary cerebellar granule neurons cultures

Cerebellar granule neurons were prepared from 6-day-old Balb/C mice as previously described (Cohen and Wilkin 1995). Briefly, cerebella were dissected and the cells were dissociated and plated at a density of 1 × 105 neurons/cm2 on poly-l-lysine-coated plates (0.5 mg/mL, Sigma). Cells were maintained in Basal Medium Eagle supplemented with 10% (v/v) fetal bovine serum, 25 mM KCl (Sigma), 1% penicillin–streptomycin, 1% glutamax and 6 mg/mL glucose. After 24 h, the medium was replaced to remove dead cells by the above culture medium supplemented with 1% (v/v) N-2 (Invitrogen) and 50 μM cytosine arabinoside (Sigma) to prevent proliferation of non-neuronal cells, respectively. Cultures were kept in an incubator at 37°C in 5% CO2 and were used after 8 days.

Cerebellar granule neurons were incubated with the indicated drug for the period mentioned in the text. The final concentration of each drug was as follow: quinolinic acid, 3mM QA (Sigma); specific mitogen-activated protein kinase (MAPK) kinase 1 and 2 (MEK1/2) blocker, 30 μM PD98059 (Tocris Bioscience, Bristol, UK); pan-calpain inhibitor, 48 and 5 μM calpeptin (Tocris); cathepsin B inhibitor, 1 μM Ca-074 Me (Calbiochem, San Diego, CA, USA); cathepsin D inhibitor, 20 μM pepstatin A (Sigma); basic FGF, 50 ng/mL FGF-2 (Peprotech); 10 μM ionomycin (Calbiochem); FGF receptor 1 inhibitor, 50 μM SU5402, (Calbiochem) and JNK inhibitor, 0.7 μM SP600125 (Sigma).

Organotypical cultures of the cerebellum

Organotypic cultures of cerebellum slices were prepared, as previously described (Stoppini et al. 1991), with some modifications. Briefly, cerebellum slices (350 μm) were cut from 6-day-old mice brains and incubated in Basal Medium Eagle medium supplemented with 6.5 g/L glucose, 1% (v/v) glutamine, 5 μg insulin and 1% (v/v) penicillin–streptomycin. To examine cell death induced by QA in organotypic cultures, slices were first incubated with 300 μM PD98054 and then exposed to 20 mM QA. The extent of neuronal damage was monitored based on propidium iodide (PI, 4 μg/mL applied for 30 min; Sigma) incorporation into the cells after 24 h of treatment with QA. The cultures were examined using a Leica Sp2 Confocal microscope (Leica Microsystems, Wetzlar, Germany). The fluorescence intensity of PI in the cerebellum was used as a percentage of cell death.

Mixed cultures and treatment with conditioned medium

To prepare mixed cultures, freshly detached microglia (1 × 105 cells/mL) were directly added over CGN at 8 days in a 1 : 2 ratio. After 24 h, the MCCM was collected, centrifuged for 20 min at 3000 g to remove debris and stored at −20°C until used. CGN were then incubated for an additional 24 h with conditioned media in a 1 : 1 ratio as previously described (Toku et al. 1998).

Assessment of cell survival

To quantify and assess cell survival, cells seeded in glass coverslips were incubated with 10 μg/mL diamidino-2-phenylindole (DAPI) for 10 min at 37°C and 4 μg/mL PI for 2 min also at 37°C. PI-positive cells (c. 500) were counted in five fields per coverslip and cell survival was expressed as the percentage (%) of PI-negative cells from the total cells (DAPI-positive cells). For the quantification of PI-positive neurons in the mixed culture, we first used microglial cultures from green fluorescent protein (GFP) mice to distinguish it from neurons. However, neuronal nuclei are clearly distinct from microglia nuclei and based on this criteria the number of PI-positive cells obtained with GFP or non-GFP microglia was similar. The magnification used for cell counting was 630×. Figure S3 shows a representative image of the mixed cultures.

Immunofluorescence analysis

Cells were fixed with 4% formaldeyde (Sigma) for 20 min at 21°C followed by permeabilization with 0.1% Triton X-100 (Sigma) in phosphate-buffered saline (PBS) for 10 min and blocked with 3% bovine serum albumin (Sigma) in PBS for 1 h. Cells were incubated with the primary antibody at 4°C overnight against mouse monoclonal NF-L (1 : 50), rat monoclonal Lamp-1 (1 : 20, BD Pharmigen, San Jose, CA, USA) and rabbit polyclonal cathepsin B (1 : 300, kindly gifted by Dr John S. Mort, Shriners Hospitals for Children, Montreal, Canada) in 3% bovine serum albumin in PBS. After the PBS + 0.01% Tween wash, cells were subsequently incubated with secondary antibody-conjugated Alexa-488 and -594 (1 : 600, Molecular Probes Inc., Eugene, OR, USA) for 1 h at 21°C. Cells were then counter-stained with 0.1 μg/mL DAPI at 21°C for 10 min and mounted in vectashield mounting medium (Vector Laboratories, Burlingame, CA, USA).

Western blot analysis

Equal amounts of total protein extracts were resolved on 8% or 15% sodium dodecyl sulphate–polyacrylamide gel. After electroblotting to nitrocellulose (Schleicher & Schuell Bioscience, Dassel, Germany) membranes, proteins were visualized using appropriate primary antibodies. Blots were incubated for 1 h at 21°C with anti-phospho-p44/42 MAPK mouse monoclonal antibody (1 : 1000), anti-MAPK 1/2 rabbit polyclonal antibody (1 : 1000, Cell Signaling Technology, Beverly, MA, USA), anti-stress-activated protein kinase (JNK1)/JNK (JNK2) rabbit polyclonal antibody (1 : 500, Cell Signaling Technology), anti-phospho-stress-activated protein kinase (JNK1)/JNK (JNK2) rabbit polyclonal antibody (1 : 1000, Cell Signaling Technology), anti-phospho-p38 rabbit polyclonal antibody (1 : 1000, Cell Signaling Technology), anti-lactate dehydrogenase goat polyclonal antibody (1 : 300, Abcam, Cambridge, UK), anti-histone 1 rabbit polyclonal (1 : 200, Santa Cruz Biotechnology, Heidelberg, Germany), anti-FGF-2 rabbit polyclonal (1 : 200, Santa Cruz Biotechnology) and anti-γ-tubulin mouse monoclonal antibody for the loading control (1 : 1000). After repeated washes in 0.1% Tween in PBS, the membranes were incubated with horseradish peroxidase-conjugated secondary antibody (1 : 1000, Zymed Laboratories Inc, South San Francisco, CA, USA) for 1 h and the immunoreactivity was detected with enhanced chemiluminescence (ECL; Amersham Pharmacia Biotech, Piscataway, NJ, USA).

Reverse transcription-PCR analysis

Total RNA isolated from CGN was converted in cDNA using 200 U of M-MuLV reverse transcriptase (Fermentas, York, UK). The template produced from the RT reaction was amplified using primers for Hsp70i (sense: 5′-CATCAAGCGCAACTCCACC; antisense: 5′-TGGTACAGCCCACTGATGAT-3′), transforming growth factor (Derynck et al. 1986), FGF-2 (Einstein et al. 2006) and hypoxanhine-guanine phosphoribosyltransferase (sense: 5′-GTAATGATCGTCAACGGGGGAGGAC-3′; antisense: 5′-CCAGCAAGCTTGCAACCTTAACCTTAACCA-3′). PCR products were resolved on 1.5% ethidium bromide-stained agarose gel.

Subcellular fractionation

Cytosolic fractions were prepared as previously described (Dignam et al. 1983). Briefly, CGN pellets were resuspended in buffer A (10 mM HEPES pH 7.9, 1.5 mM MgCl2, 10 mM KCl and 0.5 mM dithiothreitol). After 10 min on ice, neurons were lysed with a glass dounce homogenizer followed by centrifugation at 420 g for 10 min at 4°C. Supernatants were then mixed with 0.11 volumes of buffer B (0.3 M HEPES pH 7.9, 0.03 M MgCl2 and 1.4 M KCl) and then centrifuged at 100 000 g for 60 min at 4°C. The supernatant was designated as the cytosolic fraction.

Immunodepletion of MCCM

Fibroblast growth factor-2 was immunodepleted from MCCM according to the adapted protocol of Zhao and Eghbali-Webb (2001). Briefly, 600 μL samples of either control (CGN culture medium) or MCCM were incubated with 1 μg of FGF-2 antibody and 30 μL protein G-sepharose beads (Amersham Pharmacia Biotech) at 4°C for 16 h with constant rotation. The immunocomplex was removed by centrifugation at 300 g for 1 min at 4°C and the supernant was collected and stored at −20°C before usage.

Statistical analysis

Data are expressed as the mean ± SEM and were analysed for significance using Student’s t-test or one-way anova for comparison between two groups or more than two, respectively. The probability values are as follows: ***< 0.005, **< 0.01 and *< 0.05. All experiments were performed in triplicates and repeated at least three times.


QA-mediated neurotoxicity is dependent on ERK1/2 activation

Quinolinic acid, one of the metabolites of the kynurenine pathway with relevance in degenerative diseases of the CNS (Smith et al. 2001) has been reported to be secreted by activated microglia upon indoleamine 2,3 dioxygenase induction. Hence, microglia may be the primary endogenous cell type responsible for QA secretion in the CNS. To elucidate the mechanism and pathway(s) of the deleterious effects of QA, primary cultures of CGN were used as a neuronal model as these cells constitute the largest homogeneous neuronal population of mammalian brain cells and are commonly used as a model for the study of cell death (Contestabile 2002). Primary cell cultures of CGN were exposed for 24 h to an increasing range of QA concentrations as previously described (Kim and Choi 1987; Chiarugi et al. 2001; Kumar 2004) followed by cell survival evaluation using PI staining (Fig. 1a). The cytotoxicity analysis of QA shows a significant dose-dependent decrease in the cell viability of CGN starting with 3 mM up to 9 mM. The minimal concentration of QA causing a significant decrease of cell survival in neurons was 3 mM when incubated for 24 or 48 h (Fig. 1b). CGN treated with QA show disrupted neurites and disintegrated nuclei, as visualized by neurofilament or nuclear stainings, respectively (Fig. 1c).

Figure 1.

 Quinolinic acid (QA)-mediated neurotoxicity is dependent on extracellular signal regulated kinase 1 and 2 (ERK1/2) activation. (a) Primary murine cultures of cerebellar granule neurons (CGN) were challenged with several concentrations of QA for 24 h. Values are expressed as the mean ± SEM of triplicate determinations (***< 0.005 control vs. QA by anova). (b) CGN were treated with 3 mM of QA for the indicated times (***< 0.005 control vs. QA by anova). (c) Representative fields of untreated and treated CGN were photographed under phase contrast (200×) and immunofluorescence microscopy. Scale bar, 10 μm. (d) Cell extracts were analysed by immunoblotting with antibodies against phospho-ERK1/2 (pERK1/2 or p44/42) and ERK1/2. CGN were treated for 2 h with 30 μM PD98059 (PD), specific mitogen-activated protein kinase kinase 1 and 2 inhibitor, and then exposed to 3 mM QA for 24 h. Protein molecular weights markers are indicated to the left of each western blot. (e) Neuronal survival was measured by propidium iodide (PI) staining as in (a) (***< 0.005 QA 24 h vs. PD + QA). (f) Confocal microscopy images of organotypic cultures of cerebellum (i, brightfield image) which were exposed to medium alone (ii), 300 μM PD for 24 h (iii), 3 mM QA for 24 h (iv) and 2 h PD followed by 20 mM QA for 24 h (v). (g) The fluorescence intensity of PI in the cerebellum was used as the percentage of cell death (**< 0.01 QA vs. PD + QA).

We observed that QA-induced cell death is NMDA receptor-dependent accompanied with extracellular Ca2+ mobilization (Fig. S1c). To elucidate the pathway of neuronal death from NMDA receptor to the machinery of destruction, the activation profile of three signalling cascades reported to be downstream of NMDA receptor, namely, ERK1/2, stress-related p38 MAPK pathway and JNK1/2 (Choe and McGinty 2000) were analysed. We observed that only ERK1/2 is activated in CGN after QA treatment (Fig. 1d) and this activation is blocked in the presence of NMDA receptor antagonist, MK-801 (Fig. S1d). To address the role of MEK-ERK pathway in QA-induced cell death, we used PD98059, a specific inhibitor of the MEK1/2, the upstream kinase and activator of ERK1/2. Inhibition of MEK1/2 for 24 h completely abolished the levels of activated ERK1/2 (Fig. 1d) and significantly rescued neuronal death. Neuronal survival which was 40.0 ± 2.4% in QA-treated cells increased to 89.0 ± 2.8% when cells were pre-treated with PD98059 (Fig. 1e). Increased survival by PD98059 also holds true in organotypic cultures of cerebellum thus supporting our hypothesis that ERK1/2 plays a crucial role in QA-induced neuronal death (Fig. 1f and g).

Pharmacological inhibition of calpains and cathepsins abrogates QA neurotoxicity

Quinolinic acid-induced CGN death is terminal deoxynucleotidyl transferase dUPT nick end labeling (TUNEL)-negative and treatment with pan-caspase inhibitor, z-VAD-fmk, did not avoid cell death unlike CGN in low K+ concentration which were TUNEL-positive and rescued by caspase inhibition (D’Mello et al. 1993) (data not shown). Furthermore, protein synthesis inhibitor (cycloheximide) or transcription inhibitor (actinomycin D) failed to protect neurons (data not shown) contrary to what has been observed in certain models of cell death (Subramaniam et al. 2004).

Often caspase-independent cell death involves proteases other than caspases such as calpains, lysosomal and cytoplasmic proteases (reviewed in Syntichaki and Tavernarakis 2003). Activation of calpains compromises lysosomal membrane integrity releasing cathepsins into the cytosol (Yamashima 2004). The possible role of calpains and cathepsins as executioners of cell death was examined by adding a general calpain inhibitor, calpeptin (Tsujinaka et al. 1988) to CGN prior to QA treatment (Fig. 2a). Neuronal survival significantly increased to 85.1 ± 4.5% (< 0.005) in neurons exposed to 5 μM calpeptin + QA (Fig. 2a). As a positive control for calpain-dependent neuronal death, survival was measured on treatment with 48 μM calpeptin before ionomycin challenge showing 81.5 ± 1.1% of cell survival with calpeptin + ionomycin compared to 51.5 ± 1.1% with ionomycin alone (Lee et al. 2000). Furthermore, our finding is corroborated with data obtained from neuronal cultures prepared from calpastatin transgenic mice (hCAST Tg) (Fig. 2b). Neurons obtained from these mice were significantly resistant to QA induced-toxicity (78.6 ± 5.1% compared with 48.7 ± 1.0 of cell survival in neurons from wild-type mice; < 0.005). Western blot analysis confirmed hCAST expression in CGN (Fig. S2), as its expression has only been analysed in forebrain regions (Higuchi et al. 2005).

Figure 2.

 Pharmacological inhibition of calpains and cathepsins abrogates quinolinic acid (QA) neurotoxicity. (a) Cerebellar granule neurons (CGN) were treated with calpeptin (Calp) for 30 min and then exposed to 3 mM QA for 24 h or 10 μM ionomycin for 5 h as a positive control (**< 0.01; QA 24 h vs. Calp + QA; *< 0.05 ionomycin vs. Calp + ionomycin). Data are the mean ± SEM of at least three experiments. (b) Neuronal cultures of wild type (WT) and transgenic mice over-expressing human calpastatin (hCAST Tg) were exposed to 3 mM QA for 24 h (**< 0.01; WT + QA vs. hCAST + QA). (c) CGN were treated with 1 μM Ca-074 Me and 20 μM pepstatin A (Pepst A) for 30 min and then exposed to 3 mM QA for 24 h (***< 0.005; QA 24 h vs. Ca-074 Me + QA or PepstA + QA). (d) Cytosolic extracts of CGN with QA for the indicated times were resolved by sodium dodecyl sulphate–polyacrylamide gel and analysed by immunobloting with an antibody against cathepsin B and lactate dehydrogenase (LDH) as cytosolic fraction control. (e) Immunofluorescence images of lysosomes (Lamp-1 in red) and cathepsin B (green). CGN were exposed to medium alone (Control) or to 3 mM QA for 24 h (QA). Scale bar, 10 μm. (f) Protein extracts of CGN from WT and hCAST Tg mice were subjected to western blotting analysis using phospho-extracellular signal regulated kinase 1 and 2 (pERK1/2) and ERK1/2 antibodies.

In addition, inhibitors of cathepsin B and cathepsin D (Chan et al. 1999), Ca-074 Me and pepstatin A, respectively (Umezawa 1976; Marciniszyn et al. 1977), when added prior to QA treatment (Fig. 2c) led to a significant increase in neuronal survival (87.7 ± 7.2% and 91.5 ± 7.3%, respectively; < 0.01) compared with neurons challenged with QA alone (44.7 ± 13.6%). As shown by western blot (Fig. 2d), the active form of cathepsin B appears in the cytosol of QA-treated neurons, starting after a 6 h treatment as reported in other in vitro models of cell death (Foghsgaard et al. 2001; Erdal et al. 2005). To analyse lysosomal membrane integrity, we performed immunofluorescence by staining against the lysosomal membrane protein, Lamp-1 (Lewis et al. 1985) and cathepsin B (Fig. 2e). We observed that cathepsin B exhibits a punctate staining in untreated cells consistent with its lysosomal localization. After QA treatment, cathepsin B staining assumed a diffuse distribution in the cytoplasm of neurons suggesting that they are no more contained within lysosomes.

Western blot analysis shows that ERK1/2 is still activated in neurons when calpains are blocked by over-expression of its natural inhibitor (hCAST) (Fig. 2f). These data indicate that activation of calpains, lysosomal disruption and cathepsin release are responsible for cell death and occur downstream of MEK-ERK pathway.

Mixed culture of CGN and microglia abrogates QA-induced neurotoxicity

In vivo neurons and microglia are in close contact with each other and the outcome of survival or death may be dependent on intercellular interactions directly or via secreted factors. To investigate the functional consequences of the interaction between the two cell types in an excitotoxicity model such as QA, we established a mixed culture system by plating microglia over a neuronal culture in a 1 : 2 ratio, as reported before (Minghetti et al. 2005; Eljaschewitsch et al. 2006) (Fig. S3). We previously observed that neurons were sensitive to QA (Fig. 1a and b) but we also found that microglia were resistant to it probably because of the absence of the NMDA receptor 2B subunit (NR2B subunit) (data not shown), which is necessary for QA-binding and triggering excitotoxicity. In fact, neither Ca2+ mobilization nor activation of ERK1/2 was observed in microglia (data not shown). Interestingly, when QA was added to the mixed culture, neuronal survival significantly increased (97.2 ± 9.0%, < 0.005) in comparison to neurons treated with QA alone (41.1 ± 4.9%, Fig. 3a). The same protective effect was observed in the mixed culture challenged with glutamate (data not shown). Our results clearly indicate that when the two cell populations are cultured together, neurons are protected from excitotoxicity.

Figure 3.

 Mixed culture or mixed culture-conditioned medium (MCCM) confer neuroprotection. (a) Microglia were added to the cerebellar granule neurons (CGN) culture in a 1 : 2 ratio. After 1 h, 3 mM quinolinic acid (QA) was added to the mixed culture. Neuronal survival was assessed as previously described after 24 h of treatment with QA (***< 0.005; QA 24 h vs. mixed culture + QA). Data are the mean ± SEM of at least three experiments. (b) MCCM obtained from a neuron-microglia mixed culture was added to fresh neurons in culture in a 1 : 1 ratio 1 h before QA challenge (***< 0.005; QA 24 h vs. MCCM + QA).

Neuroprotection against QA conferred by microglial presence in the neuronal culture can also occur when the two cell populations are cultured in the same well but physically separated by a membrane (0.22 μm). We observed the percentage of neuronal survival was 96.4 ± 1.8% when microglial cells were plated on cell culture inserts (0.22 μm) compared to 94.5 ± 7.5% when in direct contact. This result indicates the involvement of soluble survival factor(s) present in the mixed culture.

If neuroprotection indeed can be mediated by soluble factors, MCCM should rescue neurons. When MCCM was added to a fresh neuronal culture in a 1 : 1 ratio of total volume of media before QA treatment, there was a significant increase in neuronal survival, 76.2 ± 2.0%, compared to cell survival in QA only treated cells, 40.0 ± 3.2% (< 0.005) (Fig. 3b). However, the percentage of cell survival was slightly less than that observed in the mixed culture (97.2 ± 9.0%) (Fig. 3a) which may reflect the dilution factor involved in mixing the two culture media in a 1 : 1 ratio of total volume. So far, our results reveal that neurons can survive QA challenge in the presence of microglia or soluble factors in the MCCM.

Neuroprotection conferred by MCCM is mediated by up-regulated FGF-2 and Hsp70 expression

The two growth factors, FGF-2 and nerve growth factor (NGF) have been described to be neuroprotective against excitotoxicity caused by glutamate or QA (Mattson et al. 1989; Freese et al. 1992; Fernandez-Sanchez and Novelli 1993; Kume et al. 2000). We found that treatment with either FGF-2 or NGF indeed reverted the neurotoxic effect of QA (87.7 ± 12.5%; 89.9 ± 10.2%, respectively; < 0.01). In addition, the inhibition of the FGF receptor in neurons by a specific inhibitor, SU5402 (Mohammadi et al. 1997), completely abolished the protective effect of MCCM (40.1 ± 11.6% compared with 89.3 ± 8.2% of cell survival in MCCM + QA; < 0.005; Fig. 4a). The percentage of neuronal death was similar to that observed in neurons challenged with QA alone (41.7 ± 8.1%, Fig. 4a).

Figure 4.

 Heat-shock protein 70 (Hsp70) induction and fibroblast growth factor (FGF-2) up-regulation by neuron-microglia crosstalk. (a) Neuronal survival was evaluated by propidium iodide staining after 24 h quinolinic acid (QA) challenge (***p < 0.005; MCCM + QA vs. SU5402 + MCCM + QA). Data are the mean ± SEM of at least three experiments. (b) Immunodepleted mixed culture-conditioned medium (MCCM) samples were added to neurons for neuronal survival assay 1 h later, FGF-2 was added to the neuronal culture followed by QA (*< 0.05). (c) Total and FGF-2-depleted MCCM were concentrated by spin columns and analysed by western blot using FGF-2 antibody. (d) To isolate independent protein extracts from each cell type, microglia were laid on 0.22 μm membrane inserts over the neuronal culture plated in the wells. Protein extracts of cerebellar granule neurons (CGN) and microglia cultured alone or together in the mixed culture for the indicated times were subjected to western blot analysis using FGF-2 antibody. (e) RT-PCR analysis of FGF-2 expression in CGN or microglial cells. As previously mentioned, microglia were laid on 0.22 μm membrane inserts over the neuronal culture. After the indicated times of incubation mRNA was isolated from each cell type. The figure is a representative of three independent experiments. (f) Immunofluorescence images of cathepsin B (green) and nucleus (DAPI; blue). Untreated CGN (Control) and treated with 50 ng/mL FGF-2 alone or 2 h before QA addition (FGF-2 + QA). (g) Neuronal culture was incubated with 50 μM FGF receptor inhibitor 1 (SU5402) for 2 h before adding MCCM. QA was only added 1 h later to the neuronal culture. RNA levels for heat-shock protein (Hsp70i) and HPRT were determined by RT-PCR 24 h later. DAPI, diamidino-2-phenylindole. [Correction added on 26 August 2008, after first online publication: original Fig. 4 was replaced with an updated version.]

Our data until now indicate that MCCM confers neuroprotection against QA via soluble factors and FGF-2 may be a factor likely to be released when neurons and microglia are cultured together. We analysed the role of neuronal FGF-2 in the neuroprotection against QA by removing it from MCCM through immunoprecipitation (Fig. 4b). The exposure of the neuronal cultures to the FGF-2-depleted MCCM followed by QA stimulus resulted in a significant reduction of neuronal survival (49.5 ± 11.0%; < 0.05) when compared with the MCCM (MCCM + QA, 93.2 ± 1.3%). We could also reverse the effect of FGF-2 depletion by adding back FGF-2 together with the immunodepleted MCCM in presence of QA (89.9 ± 5.9%). The successful removal of FGF-2 was confirmed by western blot (Fig. 4c). Furthermore, immunofluorescence analysis shown in Fig. 4f demonstrates that lysosomal membrane integrity is maintained in rescued neurons when pre-treated with MCCM. In this case cathepsin B exhibited a typical punctate staining suggesting the presence of cathepsins in the lysosomes, whereas upon FGF receptor inhibition it appeared diffused indicating its localization in the cytosol similar to neurons stimulated with only QA.

The enhanced function of FGF-2 in presence of microglia was addressed by culturing neurons and microglia in the presence of an insert 0.22 μm such that the medium bathed both cell types without any physical contact. Neuronal cultures at the bottom of the well and the microglial cultures on the insert were used to make independent protein extracts and analysed by western blot to assay the expression of FGF-2 (Fig. 4d). It is clear from Fig. 4d that FGF-2 expression was up-regulated in neurons after 16 h in the presence of microglia. Microglial extracts showed no detectable expression of FGF-2. This was further substantiated by transcriptional expression of FGF-2 by RT-PCR showing that neurons express higher levels of FGF-2 mRNA after 2 h incubation with microglia (Fig. 4e). Based on the western blots, we could not exclude the possibility that there are basal levels of FGF-2 in the neurons (Fig. 4d). The amount of γ-tubulin in the lane for neurons that are not in culture with microglia is less than that of the bands visible at 3 and 16 h (Fig. 4d), but the difference in the FGF-2 bands between 3 and 16 h is clear and together with the results of the RT-PCR are convincing about the effect on expression (Fig. 4d and e). Our data suggest that as a consequence of neuron-microglia crosstalk, FGF-2 is transcriptionally up-regulated and released by neurons mediating neuroprotection.

It has been reported that FGF-2 induces the up-regulation of the Hsp90 and Hsp70 in MCF-7 a human breast cancer cell line (Vercoutter-Edouart et al. 2001). A recent report showed that over-expression of Hsp70 can promote cell survival by inhibiting lysosomal membrane permeabilization and therefore prevent cathepsin release (Nylandsted et al. 2004). Hsp70i may play a role in the CNS as in post-neuronal injury it is expressed in vivo (Uney et al. 1988; Brown et al. 1989; Vass et al. 1989; Ferriero et al. 1990; Hamos et al. 1991). All these findings prompted us to enquire whether Hsp70i induction is downstream of FGF-2. Hsp70i mRNA levels were increased by RT-PCR in neurons treated with QA and MCCM (Fig. 4g) and the induction was lost on inhibition of the FGF receptor. FGF and NGF added exogenously in presence of QA also caused transcriptional induction of Hsp70i (Fig. S4a) unlike other neuroprotective molecules such as transforming growth factor (Fig. S4a). Addition of FGF-2 in presence of QA led to the typical punctate staining of cathepsins (Fig. S4b) suggestive of intact lysosomes as obtained with MCCM (Fig. 4f). In conclusion, our data reveal that in the presence of microglia, FGF-2 is induced in neurons acting in an autocrine manner to protect neurons against QA excitoxicity through Hsp70 induction, which may block lysosomal rupture.

MCCM activates JNK1/2 and pharmacological inhibition of which results in loss of MCCM-mediated neuroprotection

We have shown in Fig. 1d and e that QA-induced neuronal cell death involves chronic activation of ERK1/2. However, ERK1/2 remained activated in neurons rescued from QA toxicity by calpain inhibition (Fig. 2f), FGF-2 (data not shown) or MCCM treatments (Fig. 5a). To explain the rescue of neuronal death by all these treatments without inactivating ERK1/2, we investigated whether other signalling cascades downstream of FGF-2 could be activated by MCCM to confer neuroprotection against QA-induced death. Hence, phosphatidylinositol 3-kinase and protein kinase B (PI3-kinase/AKT), p38 and JNK1/2 pathways (Tokuda et al. 2003; Im et al. 2007; Rose et al. 2007) were analysed by western blot. As shown in Fig. 5a, treatment with MCCM or MCCM + QA clearly induced the phosphorylation of JNK1/2 (p54/p46) and ERK1/2 continued to remain active. This chronic activation phenotype of ERK1/2 has been reported under oxidative stress where activated ERK1/2 translocates to the nucleus when cells are under death dealing stress and shuttles back to the cytoplasm when being tolerant to the death stimuli (Stanciu et al. 2000; Stanciu and DeFranco 2002). Analysis of subcellular localization (Fig. 5b) by immunofluorescence of neurons treated with MCCM + QA for 24 h and simultaneously treating HT22 cells with glutamate for 6 h as a positive control (Stanciu and DeFranco 2002), shows activated ERK1/2 located in the nucleus following QA or glutamate treatments as early as 2 h (data not shown). In contrast, it localizes in the cytosol after the treatment with either MCCM or MCCM + QA (Fig. 5b) but retaining its phosphorylated active state. The significance of this relocalization is still unknown to us and others but may be a consequence of rescue from QA-induced neuronal death even though these results strongly suggest that neuroprotection conferred by FGF-2 occurs without affecting ERK1/2 activation. In summary, the inhibition of ERK1/2 activation does protect neurons from QA, hence we conclude that activated ERK1/2 plays a critical role in causing death but its inactivation may not be necessary for survival downstream of FGF-2.

Figure 5.

 Mixed culture-conditioned medium (MCCM) activates c-jun N-terminal kinase 1 and 2 (JNK1/2) and pharmacological inhibition of which results in loss of MCCM-mediated neuroprotection. (a) Cerebellar granule neurons (CGN) protein extracts were analysed by immunoblotting with antibodies against phospho-extracellular regulated kinase 1 and 2 (pERK1/2), phospho-JNK1/2 (pJNK1/2), pAKT, p-p38 and β-actin when neurons were pre-treated with SU5402 and then maintained in the presence or absence of MCCM and quinolinic acid (QA). (b) Immunofluorescence images of phospho-ERK1/2 (green) and nucleus (DAPI; blue) of an untreated (control) neuronal culture (CGN) and when the MCCM was added before QA challenge. HT22 cells were used in the assay as a positive control for nuclear localization of activated ERK after 6 h of glutamate treatment. a, b, c and d represent enlarged areas of the white boxes. Representative cells are shown. Scale bar, 10 μm. (c) Neuronal cultures were incubated with 0.7 μM JNK inhibitor (SP600125) for 1 h before adding MCCM. QA was added 1 h later to the CGN culture. Neuronal survival was evaluated by propidium iodide staining after 24 h of QA challenge (***p < 0.005 MCCM + QA vs. SP600125 + MCCM + QA). Data are the mean ± SEM of at least three experiments. DAPI, diamidino-2-phenylindole.

Accordingly to our results, the neuroprotective effect of MCCM via FGF-2 may function through JNK1/2 (Fig. 5a). Western blot analysis on pharmacological inhibition of the FGF receptor with SU5402 shows reduced levels of phosphorylated JNK1/2 in presence of MCCM + QA as compared with uninhibited FGF-2 receptor (Fig. 5a). JNK1/2 shows slight activation even in untreated neurons probably because CGN requires a basal activity of this kinase for their survival as shown before for cortical neurons (Jara et al. 2007). To confirm the putative role of JNK1/2 in MCCM-induced neuroprotection, neuronal survival was evaluated on neurons treated with a specific JNK inhibitor, SP600125 (Bennett et al. 2001), prior to MCCM and QA challenge (Fig. 5c). Data reveal that JNK inhibitor prevents MCCM-induced neuroprotection against QA toxicity (43.1 ± 4.0% compared with 91.8 ± 1.2% of cell survival in MCCM + QA; < 0.005). In conclusion, our data indicate JNK1/2 activated by MCCM/FGF-2 plays a role in neuroprotection.


Microglia potentially can exert both neurotoxic and neuroprotective effects but how this duality is negotiated remains to be detailed. In this study, we addressed this issue by focusing on CGN and microglia crosstalk in the perspective of QA challenge in both isolated and mixed cultures. Microglia are neither sensitive to QA-induced death nor activated by QA (data not shown), while neurons die as previously reported (Schwarcz et al. 1983; Beal et al. 1986; Kerr et al. 1995). We show that QA, an agonist of the NMDA receptor, increases Ca2+ influx, activates ERK1/2 and calpain leading to lysosomal release of cathepsins thus causing the neuronal demise (Figs S1 and S2). Pharmacological intervention at different stages of this trajectory rescues neurons from QA-mediated cell death (Figs 1 and 2).

We further show that neurons become resistant to QA-induced cell death when cultured with microglia or in the presence of MCCM (Fig. 3). The neuroprotective effect of MCCM is by transcriptional up-regulation of FGF-2 induced in neurons by crosstalk with microglial cells. This protection is lost when the FGF receptor was impaired or when FGF-2 is depleted from the conditioned medium of the mixed culture endorsing its role in neuroprotection (Fig. 4). Interestingly, the rescue of neuronal cell death mediated by MCCM, FGF-2 or transgenic expression of calpastatin does not abrogate the activation of ERK1/2. It is uncommon to find ERK1/2 playing a role in cell death but recent findings suggest that ERK1/2 can contribute to neuronal death (Jiang et al. 2000; Stanciu et al. 2000; Subramaniam et al. 2003, 2004). Emerging evidence also indicates that cell type, stimulus type, altered kinetics and subcellular localization of activated ERK1/2 can account for its dual effects on cell fate (Stanciu and DeFranco 2002). Indeed, we found ERK1/2 localizes to the cytosol from the nucleus during MCCM-induced neuroprotection (Fig. 5). Calpains implicated in our work for the excitotoxicity induced by QA have been described as possible ERK1/2 targets (Yoon and Seger 2006) supporting our finding of persistent ERK1/2 activation even when these proteases were inhibited with their natural inhibitor calpastatin.

We demonstrate that the medium we call MCCM is enriched in FGF-2 as result of neuron-microglia interaction. Our data suggest that FGF-2 activates the JNK pathway, prevents the release of cathepsins from the lysosome and induces Hsp70i (Fig. 4, Fig. 5 and Fig. S4). The Hsp70i has been demonstrated to have a neuroprotective role both in animal and cell culture models of neurotoxicity such as ischaemia (Ferriero et al. 1990; Xu et al. 2006), trauma (Brown et al. 1989), seizures (Uney et al. 1988; Vass et al. 1989) and Alzheimer’s disease (Hamos et al. 1991). Recently, a mechanism of protection against death stimuli has been proposed in which the inducible Hsp70i, by localizing in the lysosomal membrane, prevents the loss of membrane integrity and the subsequent release of cathepsins (Nylandsted et al. 2004). We propose that Hsp70i may have a similar role in the protection against QA by its transcriptional up-regulation in neurons mediated by FGF-2 present in MCCM.

In conclusion, as presented schematically in Fig. 6, we have mapped the trajectory of death under QA challenge and elucidated a pathway triggered by neuron-microglia interaction involving transactivation of FGF-2, upstream of JNK1/2 activation for neuroprotection against QA. The protective mechanism involves Hsp70i induction which is probably preventing the lysosomal disruption and cathepsin release in the cytosol (Fig. 6). It is likely that microglial effect on neurons is via secreted soluble factor(s) the identity of which is yet to be known.

Figure 6.

 Working model presenting pathway of death mediated by quinolinic acid (QA) in cerebellar granule neurons (CGN) and the mechanism of neuroprotection by microglia in mixed culture. (a) QA binds to the NMDA receptor through NR2B subunit causing a massive calcium influx. NMDA receptor overstimulation leads to extracellular regulated kinase 1 and 2 (ERK1/2) activation and its translocation into the nucleus and cathepsin release from the lysosomes. Consequently, these events lead to neuronal cell death. (b) Fibroblast growth factor-2 (FGF-2), a growth factor released by neurons when they are in the presence of microglia prevents neuronal cell death induced by QA via the FGF-1 receptor. This neuroprotection is mediated by c-jun N-terminal kinase 1 and 2 and it is associated with inducible heat-shock protein. Consequently, lysosomal membrane integrity is maintained and cathepsins are not released into the cytosol. Altogether, our data indicate that the neuron-microglia crosstalk is responsible for the neuronal protection against QA-mediated cell death. Hsp70i, inducible heat-shock protein. [Correction added on 26 August 2008, after first online publication: original Fig. 6 was replaced with an updated version.]


We thank Dr John S. Mort for the cathepsin B antibody. We also acknowledge Dr Nobuhisa Iwata for kindly giving the transgenic mice to our collaborator Dr Carlos Duarte (Centro de Neurociências de Coimbra, CNC). We thank Helena Cabaço for the help in preparing the organotypic cultures of cerebellum. The authors are grateful to Dr Carlos Duarte, Dr Ana L. Carvalho, Dr Helena Soares, Dr Rui Costa and Dr Gabriela Silva for critical reading and comments on the manuscript. The technical help of Nuno Moreno for two-photon microscope and the photographic reproduction are gratefully acknowledged. This work was funded by grants (SFRH/BD9093/2002, POCTI/BCI/42249/01, POCTI/CBO/47565/02 and POCI/SAU-NEU/56986/04) from Fundação para Ciência e Tecnologia (Portugal).