Anesthetics Interfere with Axon Guidance in Developing Mouse Neocortical Neurons In Vitro via a γ-Aminobutyric Acid Type A Receptor Mechanism

RECENT epidemiologic studies have demonstrated a striking correlation between childhood exposure to general anesthetics (GAs) and subsequent learning and behavioral disorders,^1–5  raising concerns that GAs may adversely affect brain development.^6–10  It is difficult to dissociate the effects of surgery and anesthesia in clinical research. However, in developing rodent models, a combination of GAs administered at clinically relevant doses in the absence of surgery has been shown to cause persistent deficits in learning and memory.¹¹  The functional deficits observed in behavioral testing were accompanied by two distinct phenomena: first, GAs enhanced neuronal apoptosis throughout the forebrain; and second, they produced alterations in brain function, in the form of decreased long-term potentiation in the hippocampus.¹¹  Increases in apoptosis or apoptotic markers resulting from the administration of GAs have been verified in model systems ranging from the primate brain to organotypic rat brain slices to dissociated cell culture.^12–14  However, apoptosis is widespread and adaptive in normal brain development,^15,16  and so the question remains as to whether changes in the function of surviving neurons caused by GAs might be as important as the increased levels of cell death.

Normal brain development requires axons to navigate over long distances to form synapses with appropriate dendritic targets, and a range of severe cognitive defects are features of human developmental disorders that result from disrupted axon guidance.¹⁷  The axonal growth cone (AGC) is a highly motile specialized structure at the distal tip of the axon, which determines the direction of growth based on sensing of guidance cues that can be either attractive or repulsive.¹⁸  The semaphorins are a family of diffusible guidance cues initially determined to be chemorepulsive based on their ability to collapse growth cones.^19,20  Using Semaphorin-3A (Sema3A) and the rodent neocortex as a model, we tested the hypothesis that commonly used GAs might disrupt development of the nervous system by interfering with axon guidance, a process that is vital to circuit formation and normal brain function.

Care of animals adhered strictly to the guidelines of the National Institutes of Health and Columbia University, and institutional animal care committee approval was obtained for all experiments (Institutional Animal Care and Use Committee, Columbia University, New York, New York; and Institutional Animal Care and Use Committee, Mount Sinai School of Medicine, New York, New York). Dissociated neurons were prepared from embryonic day-18 to day-19 C57BL/6 mouse neocortex.^21,22  Neurons were plated at a density of 100 cells/mm² on coverslips coated with poly-d-lysine and laminin where noted (Sigma-Aldrich, St. Louis, MO). They were maintained in B-27/l-glutamine supplemented Neurobasal media (Invitrogen, Carlsbad, CA) and cocultured with a feeder layer of an immortalized astrocytic cell line (gift from James W. Jacobberger, Ph.D., Department of Molecular Biology and Microbiology, Professor, Case Western Reserve University, Cleveland, OH).²³  Experiments on dissociated neurons were performed at 2–4 days in vitro and represent at least three separate cultures, each with concurrent controls. For the slice overlay assay, 400-µm coronal slices of postnatal day–3 rat neocortex were allowed to settle on 0.4-µm Millicell inserts (Millipore, Bellerica, MA) for 4 h. Dissociated embryonic day–18 rat cortical neurons were then transfected with p_max-green fluorescent protein by electroporation using a Nucleofector (Lonza, Allendale, NJ) and applied at 7.5 × 10⁴ neurons per slice for 4 h before anesthetic treatment.^24,25  Experiments with slice overlay assays represent slices from at least four separate animals. There is some overlap between isoflurane and muscimol controls at 5 h, but in all cases controls were run concurrently to experimental conditions.

For isoflurane, sevoflurane, and desflurane treatment, dissociated neurons on coverslips or overlaid cortical slices were placed in airtight, humidified modular chambers (Billups-Rothenberg, Del Mar, CA) connected to an agent-specific calibrated vaporizer (Datex-Ohmeda, Madison, WI) that delivered the agent mixed with 5% carbon dioxide/95% air carrier gas at 12 l/min.²⁶  Nitrous oxide was delivered directly to the chambers from a tank containing 5% carbon dioxide, 25% oxygen, and 70% nitrous oxide. After a 15-min equilibration, the chambers were sealed and kept at 37°C. Gas composition was measured periodically using a 5250 RGM gas analyzer (Datex-Ohmeda). Overlaid slices were treated with isoflurane, muscimol, or control carrier gas for 5 or 8 h. Dissociated cultures were incubated with anesthetics or vehicle controls for 1 h, followed by 20 min with recombinant mouse Sema3A-Fc chimera or Netrin-1 (R&D Systems, Minneapolis, MN) to induce collapse. Pure propofol was diluted in dimethyl sulfoxide for delivery, and controls included equivalent amounts of this diluent. For branching experiments, cultures were incubated with isoflurane 1.2% and Netrin-1 for 8 h.

Slice cultures were fixed with 4% paraformaldehyde and immunolabeled with anti–green fluorescent protein antibody (Millipore). Neurons with clearly defined axons at least 10 μm long that fell on the cortical plate between 50 and 200 µm from the dorsal edge were visualized using an Axiophot fluorescence microscope (Carl Zeiss, Thornwood, NY) and traced using Neurolucida (MicroBrightField, Inc., Colchester, VT) for measurement of axon length and angle of trajectory. Axons were scored as either ventrally or dorsally targeted, and the odds ratio indicating the likelihood of a loss of appropriate ventral targeting is reported.

Following treatment, dissociated neurons were fixed with 4% paraformaldehyde and labeled with Texas Red–conjugated phalloidin (Invitrogen) for growth cone analysis. For these studies and for later studies on axonal arborization, the axons were identified as the longest neurite, a consistent morphologic determinant that is readily apparent at 2–4 days in vitro.²⁷  AGCs were classified morphologically by fluorescence microscopy as “extended” or “collapsed” based on standard criteria (collapsed growth cones are defined as lacking a full lamellipodia and/or exhibiting two or fewer filopodia in the direction of growth). The classification and counting of growth cones was performed by an investigator who was blind to the experimental condition. Data are reported as the mean percentage of collapsed AGCs per field, and the error bars denote SD. For branching analysis, dissociated neurons were fixed with 4% paraformaldehyde and immunolabeled with anti-βIII Tubulin (Millipore) and a fluorescently tagged secondary antibody. Axonal arbors were traced and analyzed using Neurolucida software (MicroBrightField) by an investigator blind to condition. Data are reported as mean branch number per 100 µm with error bars indicating SD.

The criterion for significance in all statistical tests was P < 0.05, and all statistical analysis was performed using GraphPad Prism 5 software (GraphPad Software, Inc., La Jolla, CA). All analyses represent complete data sets, with the exception of one overlaid slice in the muscimol group that was inadvertently destroyed during microscopy. For all slice overlay assays, a 2 × 2 table was constructed to test the hypothesis that a loss of appropriate ventral targeting is associated with the specified pharmacologic treatment. We used the Fisher exact test to assess for statistical significance. A Student t test was used to test for a statistically significant difference in mean axon outgrowth length between the control and isoflurane-exposed groups after an 8-h exposure. For all collapse assays, we tested the hypothesis that mean percentage of growth cones that collapsed in response to Sema3A or Netrin-1 was reduced by the specified pharmacologic treatment. We performed an ANOVA with Dunnett multiple-comparison post hoc tests to assess differences in mean collapse between groups. Similarly, in the analysis of axonal branching, we used an ANOVA with Dunnett multiple-comparison post hoc tests to test the hypothesis that axonal branching induced by Netrin-1 was reduced by isoflurane treatment.

To determine whether GAs interfere with axon targeting, we chose a model system in which axons follow a clear, easily assessed trajectory. In the “slice overlay” assay, the axons of dissociated neocortical neurons applied to the cortical plate of an early postnatal coronal neocortex slice are directed ventrally toward the subcortical white matter by a dorsoventral gradient of Sema3A.²⁵  We found that 85% of control axons take an appropriate ventral trajectory. In striking contrast, after 8 h of treatment with 1.2% isoflurane, only 58% of axons take a ventral trajectory, a statistically significant reduction. Figure 1 shows examples of Neurolucida tracings from a single slice for each condition (fig. 1, A and B) along with representative examples of individual neurons (fig. 1, C and D). A summary diagram depicting the axon paths of all neurons analyzed demonstrates that the trajectories are randomly oriented in the isoflurane-treated slice (compare fig. 1, E and F). Interestingly, there was also a modest but statistically significant reduction in mean axon outgrowth length (53.0 µm for controls vs. 43.5 µm for the isoflurane group), which is consistent with previous findings in dissociated culture.²⁸ 

To further explore the relationship between GA treatment and errors in Sema3A axon guidance, we examined growth cone collapse in dissociated primary neocortical neurons. In this assay, a switch in growth cone morphology from predominantly extended (fig. 2A) to predominantly collapsed (fig. 2B) correlates strongly with repulsive targeting.²⁹  Consistent with previous findings, cultures treated with Sema3A showed 60–70% AGC collapse (fig. 2C) as compared with 30–40% at baseline for all untreated controls (fig. 2C).²⁴  Isoflurane treatment alone does not alter growth cone morphology, but isoflurane treatment for 1 h caused a statistically significant inhibition of growth cone collapse induced by Sema3A (fig. 2C). The collapse response returned to pretreatment levels when the cells were allowed to recover for 3 h from isoflurane exposure before Sema3A application, indicating that the effect is reversible (fig. 2C). This inhibition of the AGC response to Sema3A is concentration-dependent across a spectrum of clinically relevant isoflurane concentrations, ranging from 0.6–2.4% (fig. 2D). An identical loss of the AGC collapse response to Sema3A is also obtained with 4% sevoflurane or 12% desflurane (fig. 2E). There is a smaller response to 70% nitrous oxide (fig. 2E).

To determine whether this effect is unique to volatile gas, we assayed AGC collapse in the presence of IV anesthetics of different classes. Propofol, the most commonly used GA alternative to the inhaled agents, exhibits a similar disruption of the AGC Sema3A collapse response, which is fully reversible and concentration-dependent between 1 and 3 µm (fig. 2, F and G). A barbiturate, sodium thiopental (50 µm), and a benzodiazepine, midazolam (1 µm), both cause a complete inhibition of Sema3A-induced AGC collapse as well (fig. 2H). In contrast, high doses of fentanyl (100 nm) and dexmedetomidine (40 nm) do not prevent AGC collapse, and ketamine (20 µm) has only a modest effect on the Sema3A response (fig. 2H). The known pharmacology of these GAs suggests a mechanism to explain our results. The principal molecular targets of propofol, barbiturates, and benzodiazepines are the γ-aminobutyric acid type A receptors (GABA_ARs),^30,31  and the potent volatile GAs also act on GABA_ARs.^30,32–34  In contrast, it is well established that fentanyl and dexmedetomidine have no effects at GABA_ARs, whereas ketamine and nitrous oxide may have some weak GABA_AR activity.^35,36 

To test the hypothesis that the Sema3A response in AGCs is inhibited by GA actions at GABA_ARs, we assayed collapse to Sema3A in the presence of a highly specific GABA_AR agonist, muscimol, at concentrations from the nanomolar to micromolar range. An inhibition curve was constructed demonstrating a clear concentration–response relationship for muscimol, with an IC₅₀ at 0.8 µm (fig. 3A). This action of muscimol is completely inhibited by the GABA_AR channel blocker picrotoxin (50 µm) (fig. 3B). The AGC collapse response to Sema3A is also fully preserved when picrotoxin is coadministered with 2.4% isoflurane (fig. 3C), consistent with the idea that the effect of isoflurane could be mediated by GABA_AR activation. Next, we sought to determine whether the inhibition of the growth cone responses to Sema3A by GABA_AR activation can translate into a disruption of axon targeting. We found that slice overlay preparations treated with muscimol (10 µm) had a statistically significant loss of appropriate ventral targeting relative to controls (fig. 3, D and E) (77% ventral for controls; 53% ventral for muscimol group; odds ratio, 2.94).

To determine whether the effect of anesthetics on the AGC is specific to Sema3A or generalized to repulsive guidance, we tested the effects of isoflurane on the response to another guidance cue with distinct receptors and signaling pathways from Sema3A. Netrin-1 is a chemoattractant in many contexts, but it functions as a chemorepellent to axons growing on a laminin substrate.³⁷  We found that isoflurane blocked collapse induced by Netrin-1 in conjunction with a laminin substrate, and that the collapse response was restored by coadministration of picrotoxin (fig. 4A). In addition to serving as a growth cone guidance cue, Netrin-1 can alter axonal morphology by inducing branching in cortical neurons.³⁸  To test whether anesthetics interfere with multiple functions of guidance cues, as opposed to having activity specifically at the growth cone, we assayed the effects of isoflurane on axonal branching in neurons exposed to Netrin-1. We found that Netrin-1 induces axonal branch formation (compare fig. 4B with fig. 4C, which are representative examples of single neurons; mean, 0.8 branches per 100 µm increased to 2.2 branches per 100 µm with Netrin-1) but that treatment with isoflurane neither inhibits nor enhances this effect (fig. 4D) (mean, 2.0 branches per 100 µm). Representative examples of fields of traced neurons are shown in figure 4, E–G, and results are summarized in figure 4H. Considered together, these data indicate that isoflurane generally inhibits repulsive guidance independent of the cue, but that the effects of isoflurane on guidance cue activity are specific to the growth cone.

To better understand the temporal relationship between isoflurane exposure and alterations in axon guidance that might be relevant to circuit formation, we examined the duration of exposure that was required to cause measurable effects. We found that an isoflurane exposure as short as 15 min, the minimum time to achieve equilibration in the closed chamber, was sufficient to render AGCs insensitive to Sema3A in the collapse assay (fig. 5A). In pilot experiments of the overlay assay, we found that control preparations require a minimum of 5 h to achieve consistent appropriate ventral targeting. Although the effects are less dramatic than at the 8-h time point, a 5-h treatment with isoflurane was sufficient to cause a statistically significant loss of ventral targeting (fig. 5, B and C) (odds ratio, 3.41).

Our data demonstrate that both isoflurane and a selective GABA_AR agonist can interfere with axon targeting. Furthermore, isoflurane and other agents that act on GABA_ARs block the response of AGCs to Sema3A, and sensitivity to Sema3A can be restored by coadministration of a GABA_AR antagonist. Many GAs used in clinical practice have activity at GABA_ARs, a ligand-gated ion channel with chloride selectivity that is composed of a variable combination of subunits. Interestingly, ethanol, which also acts on GABA_ARs, is known to inhibit Sema3A growth cone collapse in a similar fashion,³⁹  a finding that agrees with our results and suggests that they might generalize to a broad array of compounds with activity at GABA_ARs.

The putative link between GABA_AR, Sema3A, and the complex signaling and machinery system that effects rearrangement of the growth cone cytoskeleton is not yet clear. GABA has been shown to play an important role in neuronal development.⁴⁰  Also, tonic GABA activity can influence circuit formation in newly born dentate granule cells undergoing integration into existing circuits in the adult hippocampus. In this model, a depolarizing chloride current mediated by GABA_ARs is necessary for dendritic development to occur normally.⁴¹  Immature neurons are typically depolarized by GABA_AR activation because of an outwardly directed chloride gradient maintained by the developmentally prevalent sodium potassium chloride cotransporter 1.^42,43  Both Sema3A and Netrin-1 cause shifts in membrane potential, and repulsion can be blocked by electrically clamping the membrane potential.⁴⁴  Thus, the most parsimonious explanation for the ability of GAs to inhibit the AGC collapse response to both Sema3A and Netrin-1/laminin is that they do so by effectively clamping the membrane potential of the growth cone at the chloride equilibrium potential. It is also possible that GABA_AR-mediated depolarization causes calcium entry, which in turn disrupts the localized, asymmetric calcium transients on which axon guidance is dependent.^45–47 

Alternatively, the effects of GAs on axon guidance and growth cone function may be mediated by effects on the small GTPase RhoA. Isoflurane is known to activate RhoA,⁴⁸  which can result in changes in the actin cytoskeleton.¹³  RhoA regulates cytoskeletal dynamics downstream from both Sema3A and Netrin-1 signaling⁴⁹ ; however, our findings were replicated using GAs other than isoflurane, including agents that are chemically very dissimilar to the potent volatile anesthetics. It is possible that other GAs might have similar effects on RhoA, but such evidence is lacking at present. Furthermore, RhoA is known to promote axonal branching in developing cortical neurons.⁵⁰  The absence of any effects of isoflurane on branching induced by Netrin-1 supports a model in which isoflurane and other GAs that act on GABA_ARs fundamentally alter the motile properties of the growth cone, rather than interfering generally with signaling systems such as small GTPases that are shared between different functions of the cues.

Establishing the minimum duration of anesthetic exposure that is required to cause a lasting and clinically significant alteration in brain function is of key importance in research on putative developmental anesthetic neurotoxicity. We find that the effects of anesthetics on AGC responses to Sema3A occur acutely over the course of minutes, but a model more relevant to circuit formation such as the overlay assay has a different time scale. In this setting, the AGCs experience a Sema3A cue gradient that exerts an impact over a long period of growth, and an exposure on the order of hours is necessary for the effects of isoflurane to be manifested as errors in targeting. Our study is, of course, limited by the rodent model system in which brain development proceeds on the order of months rather than years, as in a human patient. In addition, our data indicate that the effects of anesthetics on growth cones responsiveness to cues does not outlast the exposure (fig. 2C), but it is unclear what the long-term fate of aberrantly directed projections laid down during the exposure would be.

The disruption of axon guidance by GAs is a novel and potentially important mechanism of anesthesia neurotoxicity, which must be added to already existing concerns that GAs can enhance neuronal apoptosis during development. Increased apoptosis is not likely to be a feature of the model presented here because the isoflurane exposure we used in primary cortical neuronal culture did not increase cell death measured by cell density per field or cellular metabolic rate, as measured by tetrazolium reduction (data not shown). In addition, recent data indicate that developmental exposure to both propofol and potent volatile anesthetics such as isoflurane alters dendritic spine density, again suggesting a perturbation of normal development distinct from direct toxicity to neurons via proapoptotic properties of the GAs.^51,52  However, these phenomena may all be related in the intact nervous system, as neurons that fail to make appropriate connections either through errors in axon guidance or problems with synaptogenesis are thought to be eliminated by apoptosis, presumably to protect the integrity of key circuits.¹⁶  Our data underscore the importance of examining the dose, type, and duration of GA exposure in clinical investigations of pediatric anesthetic neurotoxicity.

The authors thank Roxana Mesias, B.A. (Research Associate, Department of Neuroscience, Mount Sinai School of Medicine, New York, New York), for technical assistance; Sobiah I. Khan (Research Associate, Barnard College, New York, New York) for assistance with manuscript preparation; and H. T. Lee, M.D., Ph.D. (Professor, Department of Anesthesiology, Columbia University, New York, New York), for use of an anesthesia exposure system.