Accelerated Hemolysis and Neurotoxicity in Neuron-Glia-Blood Clot Co-cultures

Abstract

A growing body of experimental evidence suggests that an intracerebral hematoma is toxic to neighboring cells. However, injury mechanisms remain largely undefined, due in part to conflicting results from in vivo studies. In order to investigate blood toxicity in a more controlled environment, murine clots were co-cultured on porous membrane inserts with primary neurons and glia. Erythrocyte lysis was apparent within 48 hours, but was reduced by almost 80% in cultures lacking neurons, and by over 90% in the absence of both neurons and glial cells. By 72 hours, most released hemoglobin had oxidized to methemoglobin or its hemichrome degradation products. At this time point, approximately 50% of neurons were nonviable, as detected by propidium iodide staining; glia were not injured. Deferoxamine, Trolox and the NMDA receptor antagonist MK-801 prevented most neuronal death, but had no effect on hemolysis at neuroprotective concentrations. The 27-fold increase in culture malondialdehyde and 5.8-fold increase in heme oxygenase-1 expression were also attenuated by deferoxamine and Trolox, but not by MK-801. These results suggest that hemoglobin release from clotted blood is accelerated by adjacent neurons and glia. Subsequent neurotoxicity is mediated by both iron-dependent and excitotoxic injury pathways.

Introduction

Intracerebral hemorrhage (ICH) is the primary event in 10–15% of strokes worldwide (Qureshi et al. 2009). Patients surviving the initial ictus often sustain deterioration in their neurological status and progressive perihematomal edema over the following days (Xi et al. 2006). Although this secondary injury was originally attributed to ischemia produced by the mass effect of the hematoma, animal studies have indicated that perihematomal blood flow is usually not reduced to ischemic levels, at least within the first 4–5 hours of hemorrhage (Yang et al. 1994, Qureshi et al. 1999). Clinical studies have demonstrated that later blood flow reduction is associated with an oxygen extraction fraction and lactate:pyruvate ratio inconsistent with ischemia (Zazulia et al. 2001, Miller et al. 2007), but rather indicating a primary metabolic dysfunction. These key observations provide the rationale for the hypothesis that an intracerebral hematoma is toxic to neighboring cells by releasing one or more toxins.

Development of therapies targeting the toxicity of extravascular CNS blood may be facilitated by a better understanding of cellular injury mechanisms. However, studies in vivo aiming to define the effects of heme breakdown (Wang et al. 2006, Qu et al. 2007), iron release (Nakamura et al. 2004, Warkentin et al. 2010), and glutamate receptor activation (Kane et al. 1994, Ardizzone et al. 2004) have yielded conflicting results, which may be due to differences in the predominant injury mechanisms in the models used. Despite their inherent reductionism, cell culture studies can efficiently address mechanistic questions, and generate hypotheses that can then be tested in a more focused and detailed fashion in vivo. In vitro investigation of whole blood neurotoxicity is technically somewhat challenging, due in large part to the barrier effect of a clot, which prevents direct observation of cells and may inhibit gas and nutrient exchange. Studies to date have instead focused on the toxicity of purified hematoma components, in particular thrombin and hemoglobin (Wang et al. 2002, Regan & Rogers 2003, Liu et al. 2008), and have suggested mechanisms by which these proteins injure neurons. Direct investigation of hematoma toxicity may provide information that may be more directly relevant to ICH, and may lead to identification of erythrocyte and neuronal injury mechanisms that are amenable to therapeutic intervention. Toward that end, we have developed a model in which clotted blood is co-cultured on porous membrane inserts with cortical neurons and glia. In the present study, we characterized the time course of hemoglobin release and neural cell injury, and tested the effect of treatment with deferoxamine (DFO), antioxidants, and the NMDA receptor antagonist MK-801.

Materials and Methods

Cortical cell cultures

Mixed primary cortical cultures containing B6;129 strain murine neurons and glia (ratio ~ 1:8; >90% of glia GFAP+) were prepared on 24-well plates as previously described (Regan & Choi 1994, Rogers et al. 2003), using a plating density of 2.8–3 hemispheres/plate. Plating medium (400 μl/well) contained MEM (Gibco/Invitrogen, Carlsbad, CA, Cat #11430), supplemented with 5% fetal bovine serum (Hyclone, Logan, UT), 5% equine serum (Hyclone), 23 mM glucose, and 2 mM glutamine. Cultures were maintained at 37°C in a humidified 5% CO₂ atmosphere. Two-thirds of the culture medium was replaced every 3–4 days. Replacement medium was similar to plating medium but contained 10% equine serum and no fetal bovine serum.

Blood exposure

All experiments were begun when cultures were 10–13 days in vitro. Culture medium was replaced with MEM (Gibco/Invitrogen Cat #51200) containing 15 mM glucose and 1 mM glutamine (MEM15). This MEM is similar to that used in plating and replacement medium but lacked phenol red as an indicator dye, which interferes with spectrophotometric determination of hemoglobin concentrations. A 12mm diameter hanging cell culture insert (Millicell, Millipore, Billerica, MA, Cat. #PIHT12R48, membrane pore size 0.4 μm) was carefully placed into each well, taking care to avoid formation of bubbles. This resulted in moistening of the submerged membrane, but MEM15 did not immediately pool in the insert, and so added blood was not diluted prior to clotting. After donor mice were deeply anesthetized with isoflurane, blood was removed via cardiac puncture using a 1 ml syringe attached to a BD Vacutainer Blood Collection Set. No anticoagulants or other additives were used. Blood (50 μl/well) was then rapidly placed on the membrane surface of each insert; sham cultures were treated with an equal volume of MEM15 supplemented with 28 μl equine serum to approximate the serum content of 50 μl murine whole blood (Russell et al. 1951). Culture plates were then returned to the incubator. One hour later and every 24 hours until analysis, 50 μl MEM15 was added to inserts to keep clots moist and to provide supplemental glucose. No erythrocytes were observed outside of the inserts throughout the course of all experiments, with the exception of an occasional defective insert that leaked immediately after blood was added; these wells were excluded from further analysis. MK-801 (Merck, Whitehouse Station, NJ), DFO (Sigma-Aldrich, St. Louis, MO), ascorbate (Sigma-Aldrich) and Trolox (Sigma-Aldrich and Acros Organics, Geel, Belgium) were diluted in MEM15 and added to cultures immediately prior to addition of insert and blood. Direct comparisons within each experiment were made using blood obtained from the same donor mouse.

Culture medium analysis

1) Hemoglobin: At indicated time points, 75 μl of culture medium was removed from each well and was diluted with 675 μl phenol red-free MEM. After mixing, absorbance at 560 nm, 577 nm, 630 nm, and 700 nm was immediately determined. Concentrations of oxyhemoglobin (oxyHb), oxidized hemoglobin (methemoglobin, metHb), and low spin degradation products collectively called hemichromes (HC) were then calculated using the method of Winterbourn (Winterbourn 1990). Hemoglobin was expressed as the concentration of heme. 2) Glucose: 1.5 μl medium was removed under sterile conditions and tested using the Accu-Chek Aviva system (Roche Diagnostics, Indianapolis, IN). 3) Malondialdehyde: After protein precipitation with 4.5% trichloroacetic acid, malondialdehyde was assayed as previously described in detail (Regan et al., 1998).

Cell viability assay

Cell death was quantified by measuring fluorescence intensity of cultures stained with propidium iodide (PI), which stains the nuclei of cells with disrupted membranes, as previously described (Vanderveldt & Regan 2004). After removal of inserts, cultures were washed with MEM and then were incubated with 13 μg/ml PI at 37°C for five minutes. Cultures were visualized using a Nikon Diaphot 300 fluorescent microscope at 100X magnification. Fields in the center of each well were viewed and captured, and fluorescence intensity was assessed using Scanalytics IPLab software. Baseline fluorescence was determined by analyzing sister cultures from the same plating incubated with inserts containing culture medium only and then stained with PI, and was subtracted from all values to quantify the fluorescence signal produced by blood toxicity. Since the blood exposure used in these experiments injured neurons but spared glial cells, fluorescence intensity was scaled to that in sister cultures treated with 300 μM NMDA, which selectively kills all neurons in this culture system.

In experiments assessing injury in cultures containing only glial cells, quantitative fluorescence methods were unfeasible since treatments that killed all glial cells resulted in detachment of the monolayer from the culture plate. In these experiments, percentage cell death was determined with cell counts after staining with both PI and membrane-permeable Hoechst 33258 (13 μg/ml of each).

Immunoblotting

Inserts containing residual clot were removed, and medium was exchanged with fresh MEM. After aspiration, 100 μl ice-cold lysis buffer (210 mM mannitol, 70 mM sucrose, 5 mM HEPES, 1mM EDTA, 0.1% sodium dodecyl sulfate) was added to each well. Cells were then scraped with a pipette tip and collected. After sonication and centrifugation, a sample was removed for protein assay (Pierce BCA assay, Rockford, IL). Sample preparation was previously described in detail (Li et al. 2009). After protein separation on 12% polyacrylamide gels and transfer, membranes were incubated overnight at 4°C with primary antibodies (rabbit anti-HO-1, Assay Designs, Ann Arbor, MI, Product No. SPA-895; rabbit anti-actin, Sigma-Aldrich Product No. A2066, 1:400), followed by goat anti-rabbit IgG-HRP (Pierce, Product # 1858415, 1:3000) for one hour at room temperature. Immunoreactivity was visualized using Super Signal West Femto Reagent (Pierce) and Kodak Gel Logic 2200, and was quantified with Kodak 1D software.

Immunostaining and tomato lectin staining

After fixation with ice-cold 4% paraformaldehyde for one hour, cultures were washed twice with 750μl cold TBS and were then serially treated with: 0.25% Triton X-100 for 10 min, 10% normal goat serum for 15 min, 1:100 dilutions of monoclonal anti-NeuN antibody (Alexa Fluor®488-conjugated, Millipore) or rabbit anti-glial fibrillary acidic protein (GFAP) polyclonal antibody (Invitrogen), or 1:500 dilution of rabbit anti-HO-1 (Assay Designs, Ann Arbor, MI), overnight at 4°C. Cultures treated with anti-GFAP or anti-HO-1 were then treated with biotinylated anti-rabbit IgG (1:200, Vector Laboratories, Burlingame, CA) for 30 min followed by NeutrAvidin Rhodamine Red-X conjugate (1:200, Invitrogen) for 30 min at room temperature.

Microglia were identified by incubation overnight with FITC-conjugated tomato lectin (Sigma-Aldrich, Cat # L0401, 1:300 dilution), according to the method of Saura et al. (2007).

Enhanced Perl’s Staining

After fixation as above, cultures were washed 3 times with 750 μl PBS (pH 7.4), and then were incubated for 30 minutes at room temperature in Perl’s solution (1:1, 2% HCl and 2% potassium ferrocyanide). After PBS wash, cultures were then treated with 0.05% 3,3′-diaminobenzidine (DAB, Sigma-Aldrich) in PBS for 10min, followed by 0.033% H₂O₂ in 0.05% DAB in PBS for 10min.

Statistical Analysis

Statistical analysis was performed using GraphPad Prism 3.0 and GraphPad Instat v3.05. Differences between treatment groups were assessed with one-way ANOVA followed by the Bonferroni multiple comparisons test for analysis of three or more groups, and with Student’s t-test for analysis of two groups.

Results

Rate of hemoglobin release varies with culture type

The mean hemoglobin concentration in the donor mouse population was 9.4± 0.5 mM (expressed as heme concentration). Hemoglobin was barely detectable in the medium of co-cultures containing neurons, glia and the blood clot during the first 24 hours of incubation, but subsequently increased rapidly (Fig. 1A). At 48 hours, total medium hemoglobin in these mixed cultures was 167±43 μM (expressed as heme concentration), and was predominantly in its reduced, oxyhemoglobin (oxyHb) form. By 72 hours, most hemoglobin had oxidized to ferric iron-containing methemoglobin (metHb). Further degradation to hemichromes was also apparent at this time point.

Fig. 1.

(A) Time course of hemoglobin (Hb) accumulation in medium of mixed neuron-glia cultures. Data points represent mean (± SEM, n = 8–13 cultures/condition) oxyhemoglobin (OxyHb), methemoglobin (MetHb), hemichrome (HC) and total Hb concentrations (expressed as heme content) at indicated time points after onset of incubation with 50μl blood clot. *P < 0.05, **P < 0.01 vs. corresponding MetHb concentration, Bonferroni multiple comparisons test. (B) Rate of hemoglobin release varies with culture type. Mean hemoglobin species concentrations (± SEM, n=12–13/condition) at 72 hours in cultures containing clot alone or with indicated cell types. ***P< 0.001 vs. glia-clot and clot only cultures. (C) Schematic diagram is side view of the apparatus (courtesy of Millipore Corp., revised with their permission). Photographs depict culture wells incubated for three days with hanging inserts containing clots, demonstrating effect of co-cultured cells on Hb release into medium. Mixed cultures contained both neurons and glia.

Hemoglobin accumulation in the culture medium was markedly reduced in co-cultures containing only blood and glial cells (> 90% GFAP+). At 72 hours, total hemoglobin was 517±39 μM in mixed neuron-glia-clot cultures, compared with 113±11 μM in glia-clot cultures (Fig. 1B). Levels of all hemoglobin species were significantly reduced in the absence of neurons, although the difference was less pronounced for oxyhemoglobin. Hemolysis was further attenuated when a clot was cultured in the absence of both neurons and glia, with total medium hemoglobin at 72 hours of only 41±11 μM.

Clot glucose consumption

The minimal hemolysis observed through 72 hours in cultures containing a blood clot only was consistent with continued erythrocyte viability under these culture conditions in the absence of neural cells. In order to determine if the clot remained metabolically active, glucose consumption was quantified daily. In the first 24 hours, mean glucose use by the clot was 1.62±0.15 μmoles/50 μl blood (Fig. 2). It remained stable at approximately this rate for the subsequent 48 hours, resulting in a linear plot of cumulative glucose use over time.

Fig. 2.

Clot glucose consumption remains stable over time in the absence of other cell types. Data points represent mean cumulative glucose consumption (± SEM, n = 7–8/condition) at indicated time points in clot-only cultures.

Neurons are selectively vulnerable

In mixed neuron-glia cultures, neurons are easily distinguished from glia by their phase-bright cell bodies, which tend to congregate in small groups and send out processes to nearby neurons (Fig. 3A). In control cultures subjected to medium change and then incubated with an insert containing culture medium only, most cells remained viable at 72 hours, as assessed by exclusion of propidium iodide (PI, Fig. 3B). In cultures containing 50 μl blood in the insert, neuronal injury was apparent on inspection with phase contrast microscopy at this time point (Fig. 3C); the background glial monolayer remained intact. PI staining was largely restricted to areas containing groups of degenerating neurons (Fig. 3D). Time course experiments demonstrated that PI staining tended to be slightly increased compared with sham control cultures 24 hours after addition of blood, but by 72 hours had significantly increased to approximately half the value of that observed in cultures treated for 24h with 300 μM NMDA, which kills virtually all neurons in this culture system without injuring glia.

Fig. 3.

Phase contrast (A,C) and fluorescence photomicrographs (B,D) of propidium iodide (PI)-stained neuron-glia cultures after incubation for 72 hours with inserts containing: A,B) Culture medium only. Phase-bright neurons form small groups with visible processes against a glial background; most cells remain viable, as indicated by PI exclusion. C,D) 50μl blood clot. Degeneration of phase-bright cells and increased PI staining are apparent. Scale bar = 100 μm. E) Bars represent mean PI fluorescence intensity (± SEM, n=8–9/condition) at indicated time points, scaled to that in control sister cultures treated with 300 μM NMDA for 24 hours (=100), which kills all culture neurons without injuring glial cells. Background fluorescence in sham-treated sister cultures incubated with inserts containing culture medium only was subtracted from all values to quantify the fluorescence signal specific for blood clot toxicity. **P< 0.01, ***P < 0.001 vs. cell death at 24 h, Bonferroni multiple comparisons test.

The selective vulnerability of neurons in this model was confirmed by immunostaining with antibodies raised against the neuronal marker NeuN and the astrocyte marker GFAP (Fig. 4). In cultures containing only glial cells and clot, 1.6±0.3% of glia stained with PI at 72 hours, compared with 2.8±1.1% of control cultures incubated with an insert containing culture medium only (P > 0.05, Fig. 5). However, the microglial subpopulation, as detected by tomato lectin staining (Saura 2007), increased from 2.7±0.5% to 9.1±1.0% of glial cells with clot co-culture (P < 0.001).

Fig. 4.

Neurons are selectively vulnerable to blood clot toxicity. Phase contrast (A, D) and fluorescence photomicrographs of mixed neuron-glia cultures immunostained with anti-NeuN (B, E) or anti-GFAP (C, F) after incubation for 72 hours with: A,B,C) Insert containing culture medium only. Phase bright neuronal cell bodies (A) stain with anti-NeuN (B), against a largely GFAP (+) glial background (C). D, E, F) Insert containing blood clot. Phase-bright cells have degenerated to debris; NeuN immunoreactivity is diminished (E), while GFAP immunoreactivity persists (F). Scale bar = 100μm.

Fig. 5.

Blood clot exposure is not toxic to glial cells. Merged images of glial cultures, stained with Hoechst 33258 to identify all nuclei and propidium iodide (PI) to identify cells with disrupted membranes, 72 hours after incubation with inserts containing: A) culture medium only; B) 50μl blood clot. Scale bar= 100 μm. C) Bars represent mean % PI positive cells (± SEM, n=13/condition, P > 0.05, medium vs. clot).

Effect of deferoxamine, MK-801 and antioxidants

Deferoxamine (DFO) is a hydrophilic metal chelator that detoxifies iron by binding to all six of its ligand coordination sites (Graf et al. 1984). It reduces edema and improves behavioral outcome in a rat blood injection model of ICH (Nakamura et al. 2004, Okauchi et al. 2009), but has recently been reported to be ineffective in a collagenase model (Warkentin et al.). MK-801 is a noncompetitive NMDA receptor antagonist that has also produced disparate results in vivo (Kane et al. 1994, Ardizzone et al. 2007). In order to determine if these agents altered the neurotoxicity of clotted blood in a controlled in vitro environment, cultures were concomitantly treated with either DFO or MK-801. The concentrations used were previously demonstrated to be tolerated in these cultures and to protect neurons from iron-dependent or excitotoxic injury (Regan & Guo 1999, Regan et al. 1998, Regan & Guo 1998, Regan & Rogers 2003). Neuronal death at 72 hours was reduced by approximately three-quarters by either agent (Fig. 6A). Similar protection was provided by the amphipathic antioxidant Trolox; however, ascorbate, at a concentration that protects neurons against hemoglobin neurotoxicity (Regan & Guo 1998), was ineffective. None of the agents tested had any significant effect on hemoglobin release (Fig. 6B).

Fig. 6.

Effect of MK-801, deferoxamine, Trolox and ascorbate on hemolysis and neurotoxicity. A) Bars represent mean PI fluorescence intensity (± SEM, n = 8–9/condition) in neuron/glia cultures after incubation for 72 hours with inserts containing clot alone, or clot plus MK-801 (10 μM), deferoxamine (DFO, 50 μM), Trolox (100 μM), or ascorbate (Asc, 100 μM) in the culture medium. Fluorescence is scaled to that in control sister cultures treated with 300 μM NMDA for 24 hours (=100), which kills neurons without injuring glia. PI fluorescence in sister cultures incubated with inserts containing culture medium only was subtracted from all values to quantify the fluorescence signal specific for blood clot toxicity. Since MK-801/DFO and Trolox/ascorbate were tested in separate experiments, fluorescence is compared with sister cultures treated concomitantly with clot only. B) Bars represent mean (± SEM, n=8/condition) oxyhemoglobin (OxyHb), methemoglobin (MetHb), hemichrome (HC) and total hemoglobin concentrations (expressed as heme content) in culture medium 72h after onset of incubation with 50μl blood clot. ***P< 0.001 vs. clot only condition, Bonferroni multiple comparisons test.

Malondialdehyde and heme oxygenase-1 (HO-1) expression are sensitive markers of oxidative injury in this culture system. Compared with control cultures incubated with a cell-free insert, malondialdehyde and HO-1 levels were increased 27-fold and 5.8-fold at 72 hours, respectively, in blood-treated cultures. Both markers were significantly reduced by DFO and Trolox, while ascorbate and MK-801 had no significant effect (Fig. 7). Immunostaining demonstrated that baseline HO-1 expression was predominantly localized to cells identified by tomato lectin staining as microglia, while diffuse expression was apparent after 72-hour blood clot co-culture (Fig. 8). Perl’s staining at this time point demonstrated increased iron deposition, particularly near degenerating neurons.

Fig. 7.

Effect of clot alone or with drug therapy on culture malondialdehyde and heme oxygenase-1 (HO-1) expression. A) Bars represent mean malondialdehyde (± SEM, n=11–21/condition) in neuron/glia cultures incubated for 72 hours with inserts containing culture medium or clot only, or clot plus deferoxamine (DFO, 50 μM), MK-801 (10 μM), Trolox (100 μM), or ascorbate (100–200 μM) in the culture medium. B) HO-1 expression in cultures treated as in A. Lane order in representative immunoblot corresponds with bar order. ***P < 0.001 vs. clot only condition, Bonferroni multiple comparisons test.

Fig. 8.

Heme oxygenase-1 expression and iron deposition after blood clot co-culture. A–D: Phase contrast (A, B) and fluorescence photomicrographs after anti-HO-1 and tomato lectin staining (C,D, merged images) of mixed neuron-glia cultures 72 hours after incubation with insert containing culture medium only (A,C, same field) or 50μl blood clot (B, D, same field). Most phase-bright neurons remain intact in control culture (A) but have degenerated with clot co-culture (B). HO-1 expression (red fluorescence) localizes with tomato lectin-stained microglia (green) in medium-treated control (C), but is diffuse after clot co-culture (D). White arrows indicate cells staining with both anti-HO-1 and tomato lectin.

E–F: Culture iron, as detected with enhanced Perl’s stain, is increased after blood clot co-culture (F) compared with medium-treated control (E) and is particularly prominent near degenerating neuron cell bodies (black arrows). Scale bar = 100μm.

Discussion

The present study accomplishes three ends. First, we have characterized an in vitro model that may facilitate investigation of cellular mechanisms of blood clot toxicity. Despite its inherent reductionism, this approach offers certain advantages, including a highly controlled extracellular environment, rapid and homogeneous drug delivery, and real time monitoring of injury, which can be reproducibly quantified using both morphological and biochemical endpoints. Second, we have identified relationships between clot and neural cells that may be relevant to ICH, but difficult to appreciate in vivo. Specifically, hemolysis is apparently accelerated by neurons and to a lesser extent by glia. Prior to hemolysis, the clot remains metabolically active, suggesting that it may compete with perihematomal cells for glucose and perhaps other substrates. Third, we have tested two drugs with highly variable effects on ICH-mediated injury in vivo (Kane et al. 1994, Nakamura et al. 2004, Ardizzone et al. 2007, Warkentin et al.). MK-801 and DFO were potently neuroprotective, indicating that both excitotoxic and iron-dependent mechanisms contribute to the inherent neurotoxicity of clotted blood. Other mechanisms mediating the effect of these compounds, such as inhibition of prolyl 4-hydroxylase by DFO (Siddiq et al. 2009), cannot be excluded.

Although erythrocyte rupture is a prerequisite for hemoglobin-mediated injury, it has not been quantitatively investigated in ICH models, and the time course of hemoglobin release and degradation has never been reported. Intravascular rodent erythrocytes have a mean life span of 40–60 days (Van Putten 1958). In subcutaneous and intramuscular hematomas, the appearance of erythrocyte “ghosts” lacking hemoglobin within one day suggests some loss of membrane integrity by that time point (Lalonde & Ghadially 1977, Lalonde et al. 1978). In the rodent CNS, deposits of non-heme iron have been observed in neurons and glial cells within three days of experimental ICH, consistent with hemolysis and hemoglobin breakdown (Wu et al. 2003). Accumulation of hemoglobin in the culture medium at 48–72 hours in the present study suggests a similar time course in this model. However, hemolysis is apparently not an inevitable consequence of clot formation, since clots cultured in the absence of other cells released little hemoglobin, and consumed glucose at a stable rate over a three day observation period. The mechanism by which neurons and glia accelerate hemolysis remains undefined, and seems worthy of further investigation as a potential therapeutic target. The present results suggest that it is neither iron-dependent nor oxidative, since neuroprotective concentrations of Trolox and DFO had no effect on culture medium hemoglobin levels. Hua et al. observed complement activation in the brain after experimental ICH (Hua et al. 2000), and hypothesized that this phenomenon may mediate erythrocyte necrosis via formation of the membrane attack complex (MAC). However, no published studies have directly investigated the relationship between complement activation and extravascular erythrocyte viability in the CNS. Complement proteins are synthesized by primary cultured neurons and glia (Speth et al. 2002), suggesting that in vitro testing of this hypothesis may be feasible.

Consistent with the instability of extracellular oxyhemoglobin at 37°C (Alyash 2006), methemoglobin was the most abundant hemoglobin species by 72 hours. Although methemoglobin has not been quantified in experimental ICH models, it is readily detected via magnetic resonance imaging 3–7 days after clinical ICH (Bradley 1993), and its distinct paramagnetic signal is routinely employed to estimate the age of a hematoma. The earlier appearance of methemoglobin in the present model may be an artifact of the different methods used, since a direct spectrophotometric assay is likely to be more sensitive than noninvasive imaging. Alternatively, more rapid hemoglobin oxidation may be due to the higher oxygen tension in these cultures compared with that adjacent to an intracerebral hemorrhage (Hemphill et al. 2005). The rate of methemoglobin formation may be a critical determinant of hemoglobin-mediated oxidative stress for two reasons. First, oxidation of heme moieties directly generates superoxide (Misra & Fridovich 1972). Second, oxidation also reduces their affinity for globin chains, facilitating release to sites vulnerable to free radical attack, such as membrane lipids (Bunn & Jandl 1968). If data from in vivo models demonstrate a slower rate of hemoglobin oxidation than in this cell culture model, a reduction in incubator oxygen concentration may be necessary to more closely mimic conditions in the intact CNS.

DFO has previously been reported to inhibit hemolysis produced by hemin, apparently by preventing hemin membrane binding (Sullivan et al. 1992). It also attenuated hemoglobin oxidation in intact erythrocytes treated with tert-butylhydroperoxide (Krukoski et al. 2009). However, DFO had no effect on either parameter in the present study. These results suggest that hemin binding and peroxide toxicity are not the predominant mechanisms mediating either hemolysis or methemoglobin formation in this model.

Recent studies have demonstrated that neurons express both α and β globin chains of hemoglobin (Biagioli et al. 2009), although it has not been established that these peptides assemble with heme into functional hemoglobin tetramers. Lysates of control untreated cultures used in these experiments were clear and colorless, suggesting that any contribution of neuronal hemoglobin to the total hemoglobin released into the culture medium is minimal and not readily detected by spectrophotometry.

Initial experiments demonstrated considerable glucose uptake by the clot, requiring supplementation to maintain physiologic concentrations during the three day observation period. The extraordinary capacity of erythrocytes for glucose has been extensively characterized, and is facilitated by enhanced expression of membrane glucose transporters (Montel-Hagen et al. 2009). The effect of clot formation on glucose consumption has not been reported, but the present results suggest that it is maintained at a high level. Since the only glucose available for an intracerebral hematoma diffuses from adjacent tissue, competition between erythrocytes and neural cells for glucose should be considered as a potential confounder in perihematomal metabolic studies, particularly when blood flow is reduced.

We have previously reported that purified hemoglobin produces an iron-dependent injury in this culture system that is attenuated by iron chelators and antioxidants (Regan & Guo 1998, Regan & Rogers 2003). The intact erythrocyte protects itself from its ~20 mM iron concentration with a multifaceted defense consisting of very high levels of enzymatic and low molecular weight antioxidants, most of which are cytosolic and therefore also released with membrane rupture (Minetti & Malorni 2006). The protective effect of Trolox and DFO demonstrates that endogenous antioxidants are insufficient to prevent iron-dependent oxidative neurotoxicity after hemolysis. Although both Trolox and ascorbate are potent free radical scavengers (Halliwell & Gutteridge 1999), the latter had no effect at concentrations that protect neurons from purified hemoglobin (Regan & Guo 1998). This disparity may reflect the greater membrane protection provided by Trolox due to its lipid solubility. Ascorbate can also have a paradoxical pro-oxidant effect under some conditions by reducing ferric to more reactive ferrous iron, thereby negating any benefit provided by radical scavenging (Minotti & Aust 1992). In addition, the antioxidant activity of ascorbate may be limited by mitochondrial injury produced by iron and excitatory amino acids (Liang et al. 2008), since mitochondria play a critical role in ascorbate recycling by reducing dehydroascorbate (Li et al. 2001). The inefficacy of ascorbate against blood neurotoxicity suggests that it is not the optimal antioxidant for in vivo testing.

MK-801 reduced cell membrane disruption as measured by PI staining, but had no effect on HO-1 expression or malondialdehyde levels in blood-treated cultures. The similar and quite robust protection in cultures treated with either MK-801 or DFO suggests a synergistic interaction of excitotoxic and iron-dependent injury pathways. Under conditions used in this series of experiments, pharmacologic inhibition of either pathway is sufficient to rescue most neurons. Since MK-801 does not attenuate hemoglobin neurotoxicity in these cultures (Regan et al. 1998), the excitotoxic component of injury is unlikely to be secondary to glutamate leakage from injured neurons per se. Other sources of excitotoxic stress that may be relevant to blood clot neurotoxicity include release of erythrocyte glutamate (Divino Filho et al. 1998) and upregulation of NMDA receptor function by Src-catalyzed phosphorylation (Ardizzone et al. 2007).

We have previously demonstrated that HO-1 is rapidly induced by hemoglobin in this culture system, but that HO-1 gene knockout has no effect on hemoglobin neurotoxicity (Chen-Roetling & Regan 2006). These results suggest that HO-1 induction, which is a sensitive and long-established marker of cellular oxidative stress (Tyrrell & Basu-Modak 1994), is not a direct cause of neuronal death in this model. However, mechanistic differences between the hemoglobin and blood toxicity models cannot be excluded. It is noteworthy that Wang and Doré have reported that HO-1 knockout mice sustain less injury than their wild-type counterparts after collagenase-induced ICH, as measured by injury volume, 8-deoxyguanosine levels, and neurological deficits, although edema was not altered (Wang & Doré 2007). HO-1 was primarily induced in cells expressing microglial markers, which were more numerous in cultures treated with blood in the present study. Assessment of the effect of HO-1 in this in vitro model would be most efficiently quantified using cultures prepared from HO-1 knockout mice, and seems a worthy topic of future investigation.

An intracerebral hemorrhage initiates injury cascades by both the mechanical effects of clot growth/retraction and release of cytotoxins (Xi et al. 2006). This in vitro model addresses only to the latter phenomenon. One limitation of this study is that the volume of culture medium used was only able to support a small clot, due to the high glucose consumption of blood in vitro. While this produced a moderate degree of neuronal injury that was largely prevented by either DFO or MK-801, it is possible that a larger clot would have generated lethal doses of both oxidative and excitotoxic stress, requiring combination therapy to observe any benefit. The existence of dual injury pathways mediating the neurotoxicity of clotted blood should be considered when interpreting negative results in vivo or in clinical trials. Inefficacy of iron chelators or glutamate receptor antagonists may not indicate that the target is irrelevant, but rather that monotherapy is insufficient.