Nanotechnology for in vitro neuroscience

Abstract

Neurons in vitro are different from any other cell types in their sensitivity and complexity. Growing, differentiating, transfecting, and recording from single neurons and neuronal networks all present particular challenges. Some of the difficulties arise from the small scale of cellular structures, and have already seen substantial advances due to nanotechnology, particularly highly fluorescent semiconductor nanoparticles. Other issues have less obvious solutions, but the complex and often surprising way that novel nanomaterials react with cells have suggested some revolutionary approaches. We review some of the ways nanomaterials and nanostructures can contribute to in vitro neuroscience, with a particular focus on emphasizing techniques that are widely accessible to many laboratories and on providing references to protocols and methods. The issues of nanotoxicology of greatest interest to cultured neurons are discussed. Finally, we present some future trends and challenges in nano-neuroscience.

---

Daniel R. Cooper

Daniel Cooper is a PhD student in Physics at McGill University. His thesis project involves the development of real-time fluorescent probes for cellular recording. His BS is in Physics and Physiology from McGill.

Jay L. Nadeau

Jay Nadeau is a Canada Research Chair in Nanocellular Neuroscience at McGill University's Department of Biomedical Engineering. Her research interests include nanoparticles for fluorescence and electron microscopy, genetically modified proteins for neuronal recording and engineering, and nanotoxicology. She did her PhD in theoretical physics at the University of Minnesota and a postdoc in neuroscience at the California Institute of Technology. She then spent three years at JPL's Center for Life Detection developing microscopic techniques for bacterial detection. She has been at McGill since 2004 with an adjunct position at the University of Maine, Orono Department of Physics.

---

1. Introduction: the challenge of neuroscience in vitro

Working with neuronsin vitro poses several challenges not seen with most other types of cell biology experiments. Neurons are terminally differentiated cells, and thus must be isolated as primary cultures or differentiated from precursors. The degree of differentiation is crucial to the outcome of nearly all studies. Because they do not divide, neurons are resistant to most forms of DNA transfection; viral vectors that transduce non-dividing cells such as lentivirus,¹ adeno-associated virus,² or herpesvirus³ are necessary for efficient gene delivery. Even with these methods, glial cells are often transduced more readily than neurons. Primary neurons in culture are extremely sensitive to physical shocks, such as exposure to an air–water interface and to excitotoxicity, require particular substrates for adhesion, and are susceptible to contamination by bacteria and fungi. Most labeling protocols (using dyes, beads, nanoparticles, or other reagents) reported for immortalized cell lines are inappropriate for primary neurons, leading to poor labeling, cell death, or both. Hence protocols for use in neurons must be developed and tested specifically for these cells.

Terminally differentiated neurons display complex features and behaviors that are difficult to study with traditional imaging methods. Even on a single-cell level, difficulties are seen with neurons that are not encountered with other cell types. For example, all types of cells show lateral membrane diffusion of receptors, and micron-sized beads bearing many fluorophores have traditionally been used to track this diffusion. The large number of fluorophores prevents the probe from photobleaching as it is being tracked.⁴ However, in neurons the small size of the synaptic cleft and of the spaces between neurons and glia demand the invention of smaller tools.

Similarly, the small size of features on neuronal dendrites and dendritic spines greatly restricts accessibility to labeled probes and resolution of such probes. Dendritic spines can be as small as 200 nm in diameter, near the limit of optical resolution, and with a total number of [Ca²⁺] ions on the order of 10²–10³. Imaging changes in such a small number of ions is difficult, especially in the context of the entire cell.⁵

Neuronal communication is both chemical and electrical. Chemical signals are transmitted viasynaptic vesicles, which may be newly generated at each signal or recycled. The ability to track individual vesicles is crucial to understanding these processes, and several dyes and genetically-encoded proteins have been designed for this purpose. The fluorescent styryl dyes, such as FM1-43, are highly effective for labeling vesicles.⁶ A pH-sensitive green fluorescent protein, called pHluorin, has also been fused to various synaptic vesicleproteins. Its fluorescence changes due to pH differences inside vs. outside the vesicle, allowing the movement of these proteins to be tracked.⁷ However, their limited signal-to-noise has made most of these techniques of limited value in mammalian brain cells due to the small size of the features involved.

Electrical signals involve action potentials, which are a change in transmembrane potential of >100 mV in ∼10 ms, as well as numerous subthreshold events: inhibitory and excitatory presynaptic and postsynaptic potentials, which can be smaller than 1 mV. The best method for resolving these events remains the patch-clamp technique, which can reduce electrical noise to the femtoamp level. Unfortunately, the technique requires considerable skill and is nearly always performed on only one cell (although simultaneous recordings from up to three synaptically coupled neurons have been reported⁸). A holy grail of neuroscience remains the ability to achieve patch-clamp level resolution on dozens to hundreds of cells at once. Commercial high-throughput patch-clamp devices have been developed, but these require cells in suspension that can be placed onto individual electrodes.⁹ They are not appropriate for studying neuronal connections, which require the cells to develop synapses over days to weeks.

A possible solution lies in optical recording using voltage-sensitive fluorescent labels. The electric fields associated with an action potential across a 5 nm membrane are substantial, but the field is located only within the lipid bilayer, meaning that any sensor of membrane potential must be at least in part lipophilic. The best results obtainable today make use of dyes such as di-4-ANEPPS and di-8-ANEPPS, which show high toxicity, low signal-to-noise, and broad emission spectra. Dyes are also diffusible and stain throughout a culture dish, so they cannot be used in individually selected cells the way microinjection can.¹⁰ Alternatives to voltage-sensitive dyes have been sought in the form of modified fluorescent proteins. The most successful approach to such a probe involves a fusion of the Shaker potassium channel to GFP, called FlaSh (not to be confused with the protein-labeling reagent FlAsH¹¹). While this fusion protein establishes a proof of principle, its emission changes with voltage are unfortunately very small (∼1%).¹² The greatest drawback of this type of approach is that the GFP is located outside of the membrane, so it is not directly exposed to the electric field. Attempts to improve upon FlaSh with multiply-labeled ion channels usually lead to poor expression and retention of the protein in the Golgi apparatus.¹³

Optical signals may allow recording from whole networks, but the ability to stimulate selected cells or types of cells within the network is also necessary for elucidating connections. Current approaches to coupled stimulation and recording involve multi-electrode arrays coupled with optical recording. The most sophisticated of these arrays consist of ∼100 electrodes and can record from a few hundred neurons extracellularly.¹⁴ An extracellular recording shows greater longevity but reduced resolution relative to an intracellularelectrode; distinction of action potentials is often complicated. Nonetheless, growing and maintaining neurons on these arrays is difficult for many reasons.¹⁵Glial cells often insulate the neurons from the electrode, reducing or eliminating the electrical signals from action potentials. The placement of cells onto the electrodes is imprecise, so that many cells are “missed” by the electrodes. Even in the best arrays, the networks are two-dimensional. Eventually, three-dimensional networks are desirable for a more realistic understanding of the connections that occur in the brain.

This review will discuss how nanotechnology can help advance neuroscience in all of the areas of difficulty mentioned. We begin with a brief introduction to a select group of nanomaterials and nanodevices of particular relevance to neurons. We then review the latest methods and results in each target area, finishing with a section on toxicity concerns and future outlook.

2. Background: nanotechnology, nanomaterials, and cells

Nanotechnology refers to engineered materials or devices that can be controlled physically and/or chemically on a scale of 100 nm or less. This small size allows for interactions with cells and subcellular structures in highly specific, often poorly-characterized ways¹⁶ (Fig. 1). As we begin to understand the physics and chemistry of interactions at this scale, we will be able to design materials for targeted applications. However, the current state-of-the-art remains largely empirical: new techniques in materials science and engineering allow us to create smaller-sized objects and devices, which then demonstrate surprising features of relevance in biology and/or unexpected interactions with cells.

Fig. 1 Spatial scales in neuroscience and nanotechnology. The images show neuronal structures visualized by appropriate methods for each spatial scale; the schematics show corresponding structures and their relationship to a typical fluorescent nanoparticle of ∼10 nm in diameter. (A) Single cells are on the scale of 10 µm (photo: fluorescence micrograph of a single rat hippocampal neuron in culture expressing green fluorescent protein [GFP]). At this scale, individual nanoparticles cannot be resolved, but appear as a uniform fluorescence over the entire cell. (B) Single dendrites have features on the scale of 1 µm (photo of same culture as in (A)). On this scale, individual particles cannot be resolved by light microscopy, but may be tracked by single-particle tracking methods. Small clumps of nanoparticles may block structures such as dendritic spine necks (arrow). (C) The synapse is on the order of 100 nm (photo: transmission electron micrograph of a Type 1 synapse from rat cerebral cortex. Image from H. Jastrow's electron microscopic atlas [http://www.uni-mainz.de/FB/Medizin/Anatomie/workshop/], used with permission). At this scale, individual nanoparticles are large enough so that their size within synaptic vesicles and the synaptic cleft must be taken into account. For example, a nanoparticle fits within a synaptic vesicle lumen, but not through “kiss-and-run” fusion pores. (D) The lipid bilayer and transmembraneproteins are 5–10 nm in size. The upper image shows a structure determined by protein tomography, a combination of cryo-TEM and image analysis (adapted by permission from Macmillan Publishers Ltd: Nature Methods,¹²⁴ copyright 2005). On this scale the nanoparticle is large: too large to pass through the great majority of channels and transporters and greater in diameter than the cell membrane. (E) Single molecules are on the order of 1 nm and smaller, requiring angstrom-level imaging to resolve (photo: electron diffraction pattern of aquaporin ion channel. Adapted by permission from Macmillan Publishers Ltd: Nature,¹²⁵ copyright 2005). On this scale, the nanoparticle ceases to be a featureless particle and becomes a crystal. Proteins with molecular weights of tens of kilodaltons, such as the 27 kD green fluorescent protein (GFP), are comparable in size. Small molecules such as dopamine (pictured) are still five times smaller.

2.1 Semiconductor and metal nanoparticles

A classic example of a nanotechnology that fits this description is quantum dots (QDs), which are a particular class of fluorescent semiconductor nanoparticles. The physics and chemistry of QDs had been characterized for many years before the publication of the first cellular-labeling studies in 1998.^17,18 Since that time, several thousand articles have been published describing biomedical applications of QDs. Most of these take advantage of their bright, tunable photoluminescence. For a semiconductor nanoparticle to be a quantum dot, it must have a band gap energy comparable to energies of photons of visible light. Numerous narrow band gap semiconductors, which may belong to group II–VI, group IV–VI, or group III–V, fit this criterion. Excitation by a photon slightly more energetic than the band gap leads to the creation of an electron–hole pair, or exciton. When the exciton recombines, it emits a photon slightly redder than that which was absorbed. As these semiconductors are reduced to the nanoscale, their excitons are confined to nearly infinite potential wells at the edges of the particle; this adds a particle-in-a-box energy to the band gap energy. For the lowest 1s state (the position of the first exciton absorbance peak), the total exciton energy is given by¹⁹where ε is the crystal dielectric constant, a_B is the Bohr radius of the exciton, µ is the reduced mass, and R is the radius of the particle. When the particle radius is smaller than the Bohr radius, the last term dominates, leading to significant shifts in the first exciton peak with changes in particle size. This is called the strong confinement regime, and corresponds to different colors for different particle sizes.

Other semiconductor nanoparticles are not “quantum dots” because their band gap energies are so large that even the smallest particles that can be synthesized have a radius larger than the Bohr radius. Such particles include TiO₂, ZnO, and ZnS. Like QDs, these particles are easily synthesized in a wide range of sizes and shapes for biological applications. Because of their highly energetic excitons, these particles are strong oxidants or reductants (the position of the band edges, as well as the band gap energy, determines redox potential)^20,21 (Fig. 2A). They are often used as photocatalytic agents and can also serve as “shuttles” of electrons to biomolecules.²²

Fig. 2 Energies and colors of nanoparticles. (A) Band gap energies and energetic positions of the band edges (vs.NHE) in aqueous solution for commonly used semiconductor materials. These values are for bulk materials; nanoparticles show corresponding energy shifts according to eqn (1). The positions of the band edges determine the ability of the electron to reduce or the hole to oxidize other molecules. Some common redox couples are shown on the same scale for comparison. (B) Emission spectra of commercially available QDs of CdS, CdSe, and PbS (note that the x-axis is compressed for PbS; the emission peaks of these particles are broad). (C,D) Images of different core sizes of CdSe and InPQDs in colloidal solution, illuminated with a 365 nm UV wand. Note that the reddest InP sample cannot be seen.

For CdSe, the most commonly used QD material, particle sizes of 2–6 nm show photoluminescence emission that spans the visible range. A cadmium-free alternative, InP, spans a similar range.²³ CdTe, with a slightly smaller bandgap, can range from mid-visible to the near IR. PbS and PbSe emit in the near to mid-IR range, and CdS covers the near UV (Fig. 2B–D).

A “bare core” QD made simply of its core semiconductor material is very sensitive to interactions with surrounding solvents and ions, particularly water and oxygen. The electron and hole from the exciton can transfer to the solvent, oxidizing or reducing the dot. These interactions, particularly oxidation, usually reduce fluorescence and eventually break down the nanoparticle by a process called anodic decomposition.^24,25 The exciton can be protected or passivated by the overlay of a material with a higher band gap energy. Provided that this material does not alloy well with the semiconductor core, its presence will affect the emission of the QD very little. The most common material used as a cap for CdSe and InP is ZnS.^26–28 This cap can be as thin as a single atomic monolayer, and particles capped in this fashion are called “core–shell” QDs.

Interestingly, CdTe nanoparticles can be highly fluorescent in water because the positions of their band edges prevent electron transfer to common molecules.²⁹ Most biological applications of CdTe QDs use bare-core QDs for this reason. However, the particles are still sensitive to oxidation and may break down. This has important implications for cytotoxicity, as we will discuss in Section 8, and means that toxicity studies using CdSe/ZnS cannot be directly compared with those using CdTe. Synthesis of core–shell CdTe/ZnS QDs has been reported, and the shell protects the particles from photobleaching in the same way as it does for CdSe.³⁰ However, core–shell CdTe has not yet become a standard biological reagent.

QDs, particularly core–shell QDs, have been developed for their desirable optical properties for biological imaging, particularly single-particle tracking and multicolor labeling. They demonstrate size-tunable, narrow emission spectra; resistance to photobleaching; high quantum yields; and large effective Stokes shifts (they absorb strongly at any wavelength more energetic than the bandgap). They are easily synthesized as size-homogeneous populations,²⁷ making color selection a simple matter of choosing the temperature of synthesis and time of ripening. These spectral properties also make them ideal fluorescence resonance energy transfer (FRET) donors^31,32 (they are poor acceptors because of their broad absorbance; it is difficult to find a wavelength at which they do not excite).

Nanoparticles of metals such as silver and gold also show size-tunable properties;^33,34 in this case it is the surface plasmon rather than the exciton that changes with reducing size. This changes the particle's absorbance peak and, if this peak is within the visible spectrum, its color. For example, Au particles smaller than 100 nm are red, and larger particles are blue, with variations due to the refractive index of the medium (Fig. 3). They can be used in cells for several types of non-fluorescence-based imaging:³⁵surface plasmon resonance imaging,³⁶Rayleigh scattering,³⁷ photothermal tracking,³⁸ and CT scanning.³⁹ The advantages are temporal stability (no photobleaching) and the ability to image deep inside tissues. For single-particle tracking studies, another advantage is that these types of imaging do not show the intermittency (“blinking”) demonstrated by fluorophores.

Fig. 3 Size-dependent effects in Au nanoparticles. (A–C) TEM images of different sizes of biologically-synthesized, protein capped Au nanoparticles. (D) Absorbance spectra of samples A–C, showing different positions of the plasmon peak. (E) Appearance of colloidal solutions of different fractions of biologically-synthesized nanoparticles, with the size range given below each vial.

Nanoparticles can be rendered water soluble by coating with amphiphilic molecules. However, this often leaves the particle vulnerable to attack by water and oxygen; the nature and degree of the surface coat determines the rates of particle breakdown.⁴⁰ Interactions with biological systems make this picture even more complex, as the interface now consists of the particle core material, its solubilizing and/or functionalizing coat, and the cell. Particle oxidation leads to changes in fluorescence (either a decrease or an increase followed by a decrease), loss of water-solubility, and often cytotoxicity due to released heavy metals and other, less well characterized effects (Fig. 4). Unexpected increases and decreases in fluorescence intensity made early attempts to use QDs for quantitative applications impossible.⁴¹ As a result of these issues, many dozens of solubilization and functionalization schemes have been designed for nanoparticles: coating with polymers, Si shells, lipid micelles, bidentate thiols, and other compounds. These methods have been thoroughly reviewed.⁴² Each functionalization technique leads to QDs that show unique biomolecular interactions, from fluorescence stability to toxicity to sub-cellular localization. Thus, there is no “best” approach to nanoparticle functionalization, but the method should be chosen according to the goals of the experiment and its required sensitivity and duration as well as its target cells.

Fig. 4 Conceptual diagram of CdSe quantum dot (QD, green spheres with small sphere conjugate “c” attached) interactions with a neuron. Bioconjugated QDs may bind to specific cell receptors (1). However, they may be broken down into bare QDs and conjugates before they reach the receptor (2). Breakdown may liberate free Cd²⁺ ions. Uptake into cells is often by receptor-mediated endocytosis (3). Breakdown of QDs to cores, conjugates, and ions may occur inside of endosomes or in the cytoplasm (4). QDs themselves rarely penetrate into the nucleus, but products of degradation such as ions and conjugates may cross the nuclear membrane (5). QDs may be axonally transported after nonspecific or receptor-mediated uptake (6).

2.2 Carbon nanomaterials

There are many types of carbon nanomaterials, including what is probably the easiest form of nanoparticle to manufacture: carbon black, a byproduct of incomplete combustion of organic materials such as bones, vines, or petroleum products. However, it is the newer, more sophisticated forms of nanoscale carbon that have demonstrated unusual optical and electronic properties alone and in the presence of cells. Those that have had the most impact on neuroscience are single-walled carbon nanotubes (SWNTs) and spherical fullerenes (usually C₆₀).

SWNTs are rolled-up carbon (graphene) sheets one atom thick that may or may not be capped at the ends by a hemisphere. The vector describing the rolling determines the tube's structure and electronic properties (Fig. 5A, B). All the current carriers flow at the surface, so their conductance is highly sensitive to any local electrochemical perturbations. Biomolecules can interact with these carriers in a myriad of ways: by electrostatic gating, doping, reducing carrier mobility, or electron transfer. It is their electronic properties that make these materials most useful, although they also have desirable physical properties, notably great tensile strength. SWNTs can be purchased commercially as aggregates that are not soluble in any solvent, though they may be dispersed in detergent. Biofunctionalization is not trivial, as the functionalization mostly adheres to the tube end. However, highly effective methods have been developed that are straightforward to implement, involving 1,3 dipolar cycloaddition.^43,44

Fig. 5 Carbon nanotubes and buckyballs. (A) The different forms of carbon nanotubes can be defined as a chiral vector (called c) defined by the components m and n. The chiral vector indicates how an infinite graphene sheet would be rolled up to make the tube (even though they are not synthesized this way). If n = m, the tube is metallic, otherwise it is a semiconductor. (B) Appearance and names of the different types of nanotubes are defined by their chiral vector. (C) Different types of buckyballs.

Once functionalized, the SWNTs can be assembled into electrically active surfaces that are conducive to cell growth. Two distinct types of arrangement of the NTs are used: mesh deposition (essentially random association on a surface), or vertical alignment.⁴⁵ Mesh deposition is as simple as pipetting the SWNTs onto a substrate such as glass, which may have a metal electrode to permit electrical recording. Vertical alignment is more difficult, requiring chemical-vapor deposition to define sites for nanotube placement. Vertical nanotubes are also fragile, and collapse in liquid–air interfaces. They must thus be functionalized or embedded in order to retain their order.

Fullerenes or “buckyballs” are nanospheres or nanoellipses of the same graphite structure found in carbon nanotubes (Fig. 5C). The most common and smallest fullerene is C₆₀, but they can range up to C₂₀₀. Like nanotubes, they are insoluble before functionalization, but may be biofunctionalized in numerous ways, many of them noncovalent “wrapping” methods that take advantage of the thermodynamic drive to eliminate the hydrophobic interface between the fullerenes and water. Many different types of polymers may be used, including polystyrene, SDS, and DNA. Buckyballs may also be wrapped in phospholipids. The different solubilization methods and the possible interactions with cells are well reviewed.^46,47 We include these materials here primarily in the discussion of toxicity, because the study of their pharmacology has been plagued by inconclusive or contradictory results. Fullerenes are alternately reported as being anti-oxidants that protect cells⁴⁸ and as being potent generators of reactive oxygen species (ROS).⁴⁹ A recent study convincingly concluded that water-stabilized but unconjugated C₆₀ did not generate ROS, but that it oxidized bacterial membrane proteins and thus interfered with oxidative respiration.⁵⁰ These findings have particular interest to the study of neurons, where excitotoxicity is a unique issue particularly in certain cell types such as motor neurons.

2.3 Silica and polymernanoparticles

Silica is a natural component of sand and glass that is considered harmless to human health; some people ingest it as a dietary supplement. Ultrafine silica particles can be synthesized as size-homogeneous populations ranging from approximately 10 nm to more than 80 nm in diameter. Silicon is an indirect bandgap semiconductor, so in its bulk state is a poor emitter of light. However, Si particles that are strongly quantum confined (1–5 nm) show size-dependent photoluminescence in the visible range, though usually with quantum yields significantly lower than that of CdX QDs. The biggest obstacle to their use has been the difficulty of obtaining size homogeneous populations within this small size range. However, photocatalyzed acid etching can result in a homogeneous distribution of sizes and thus a relatively narrow emission spectrum. Si nanoparticles emitting between green and red have been obtained in this fashion, with some colors showing quantum yields of 60%⁵¹ (Fig. 6A, B).

Fig. 6 Silica nanoparticles. (A, B) Inherently fluorescent Si nanocrystals (adapted with permission from ref. 51, copyright 2007 American Chemical Society). (A) Simulated distribution function during Si nanoparticle suspension dissolution in the acid mixture: (1) initial distribution and (2) 10 min, (3) 30 min, (4) 60 min, (5) 100 min, (6) 160 min, and (7) 250 min after acid addition. (B) Photoluminescence spectra recorded during photocatalyzed Si nanoparticle dissolution: (1) prior to acid introduction and (2) 30 min, (3) 50 min, (4) 80 min, (5) 110 min, (6) 130 min, (7) 170 min, and (8) 200 min after acid addition. Inset: appearance of selected fractions under 360 nm illumination. (C–E) “Cornell dots” (adapted with permission from ref. 52, copyright 2005 American Chemical Society). (C) Schematic representation of synthesis of core–shell particles. (D) Appearance of final, 30 nm particles by TEM. (E) Cornell dots with different organic dyes incorporated, from left to right: Alexa 350, N-(7-(dimethylamino)-4-methylcoumarin-3-yl), Alexa 488, fluorescein isothiocyanate, tetramethylrhodamine isothiocyanate, Alexa 555, Alexa 568, Texas Red, Alexa 680, and Alexa 750.

Si nanomaterials are porous and can thus also act as carriers for labeling agents. A wide variety of molecules, such as fluorophores or chemotherapeutic agents, can be encapsulated within the matrix. Usually this arrangement is unstable, but a core–shell synthesis procedure involving a dye-rich core surrounded by a denser silica network has resulted in ultrabright, stable nanoparticles that are referred to as “Cornell dots” (Fig. 6C–E).⁵² The spectrum of the encapsulated dye determines the dot spectrum and so retains the broad emission drawbacks characteristic of organic dyes, but photostability is greatly improved because the molecules are protected from the solvent.

So-called organically modified silica nanoparticles or ORMOSILs incorporate silanes into the synthesis procedure, making the resulting particle more porous and softer than unmodified particles. This helps in the design of structures intended to release cargo. Other advantages over unmodified particles include the presence of both hydrophilic and hydrophobic groups, so they can be loaded with polar or nonpolar drugs or dyes; the fact that they can be prepared without harsh solvents or ultracentrifugation; and the ability to further modify the organic groups for biological targeting.⁵³

Nanoparticles and nanotubes can be made out of many polymer materials, usually by wet synthesis methods in which the core of a macromolecule is condensed or cross-linked into a solid core. The materials remain porous and small molecules (DNA, dyes, drugs) can be encapsulated within them. There is a large body of literature on biomedical applications of these particles, which have tremendous applications in vivo for drug delivery and cellular interfacing. Many excellent reviews exist on the subject,^48,54,55 and we will touch on them only briefly here.

2.4 Manufactured nanopatterns and scaffolds

Growth and differentiation are truly nanoscale processes, relying upon topographic features and local concentrations or gradients of growth factors. Recent advances in nanoscale patterning have allowed for the design and manufacture of substrates that emulate the extracellular matrix (ECM), the three-dimensional microenvironment that serves primarily as structural support for nearly all animal tissue. The components of the ECM are secreted by specialized cells in vivo, and various compositions of fibrous proteins and glycosaminoglycans (GAGs) confer distinct chemical and physical properties, which coalesce to orchestrate the growth of cells. The vast majority of cell culture is performed on two-dimensional substrates which do not physically reflect the form of the ECM and can result in atypical behavior of cells. In the design of an artificial substrate, there are several physical attributes of the ECM that must be taken into account: a large surface area for cell adhesion, consisting of nanoscale fibers and interconnected pores that support the cells in three dimensions while preventing small molecules from rapidly diffusing out of the microenvironment. Though there are many materials that satisfy these basic criteria, it is considerably more challenging to replicate the chemical signals and molecular gradients that influence cell growth. Necessarily, cytotoxicity and biocompatibility of the substrate are critical concerns. The important factors involved in the design of a nanosubstrate can thus be described as: dimensionality, directionality, and biocompatibility.

Synthetic biopolymers for cell culture have been studied extensively over the past few decades, but many processed synthetic polymer networks have large fibers and pore sizes, resulting in an environment that has a much larger scale than the ECM. Fibers that are tens of µm in diameter (similar to the scale of the cell body), approximate to a 2D surface depending on the fiber curvature. In addition, many synthetic polymers, though biocompatible, are not sufficiently hydrophilic for biological use. Concerns about network scale can be addressed by using natural biomaterials, which are characterized by nanolength scales. These include collagens, gelatins, and laminin; Matrigel™ is a commercially available gelatinous protein mixture that is secreted by mouse tumor cells and is available in several different compositions. Unfortunately, because these biomaterials are derived from the original ECMin vivo, they often contain ill-defined constituents that make them difficult to characterize and control, with marked variation between batches. The most successful scaffold designs exploit the advantages of both synthetic polymers and natural biomaterials by combining them through covalent binding, physical adsorption or blending in solution.

Several methods have been investigated to fabricate nanoscale scaffolds, including phase separation, self-assembly, and template synthesis,⁵⁶ with varying degrees of success and utility. The most widely used and arguably most practical technique for nanofiber production is electrospinning, a simple fabrication method that uses electric charge to form ultra-fine fibers from polymer solutions (Fig 7A), and has been demonstrated with a wide variety of synthetic polymers such as polycaprolactone, poly(L-lactic acid), poly(glycolic acid) and poly(DL-lactic-co-glycolic acid), natural polymers such as collagen, as well as blended polymers. The distinct advantage of electrospinning over other methods is that it provides substantial control over fiber alignment and size, as well as network porosity, with a fairly simple experimental setup. Electrospinning has gained popularity over other methods that produce similar networks, such as phase separation, which requires careful temperature control.

Fig. 7 Techniques for patterned substrate generation. (A) Diagram showing the basic principle of the electrospinning technique. The polymer solution is passed through a tip where it is subjected to high voltage, typically tens of kV, which charges the conductive liquid. The liquid droplet is stretched beyond the expected shape by electrostatic repulsion, and the resulting surface is known as the Taylor cone. At a critical point, the liquid erupts in a stream. As the jet dries, charge migrates to the surface of the fiber and the mode of current flow changes from ohmic to convective. The jet is then elongated by whipping caused by electrostatic repulsion and is captured on a grounded collector surface. Deposition of the stream onto the edge of a rotating disk produces aligned nanofibers. (B) Schematic of the steps in microcontact printing, showing the creation of a PDMS stamp from a mask and the deposition of biomolecules onto a substrate using the stamp.

While electrospinning can create fairly uniform networks of aligned nanofibers, top-down nanofabrication techniques such as chemical etching and photolithography allow for exceptionally well controlled patterning, with some methods such as electron beam lithography providing a resolution of a few nanometres. These techniques can be used to produce a range of substrates from a simple “roughening” of a silicon surface to elaborately etched designs, though they are often time-intensive and require rather sophisticated facilities. One of the more widely accessible techniques is soft lithography or microcontact printing.⁵⁷ In the first step, a “hard stamp” on a Si wafer is created from a printed photolithographic mask containing the pattern of interest. Then a soft stamp made from poly-dimethylsiloxane (PDMS) is created from this mask. The PDMS may be “inked” with proteins, growth factors, or other molecules and used as a rubber stamp to transfer the pattern onto a slide (Fig. 7B).

Self-assembled peptide networks provide another distinct approach to nanoscaffold design. Peptide amphiphiles have been designed with a sequence similar to collagen,⁵⁸ such that self-assembly into fibers occurs through ionic and hydrophobic interactions under physiological conditions. While the peptides have high purity and self-assembly makes the fabrication process relatively simple, it is more difficult to control the features of the network.

3. Transfection

Most transfection techniques deliver plasmidDNA to endosomes or at best the cell cytoplasm. When the cell prepares to divide and its nuclear envelope breaks down, the DNA is able to access the nuclear machinery and thus be expressed. In the absence of cell division, unless it can diffuse across the nuclear membrane, the plasmid remains unexpressed and is eventually broken down. As a result, chemical (liposome⁵⁹) or physical (gene gun⁶⁰) transfection methods are virtually ineffective in post-mitotic primary cells such as neurons. Neurons also show reduced release of cargo from endosomes relative to other cell types, and their extended processes mean that retrograde transport of endocytosed vectors is necessary for efficient transduction. A tremendous amount of work has been done on the development of vehicles to carry plasmidDNA into neurons (Fig. 8A); these techniques have been well reviewed.⁶¹ At present, neurotropic viruses are by far the most efficient technique, but cloning of the viral constructs can be difficult, their capacity is limited, and they often require biosafety containment.

Fig. 8 Transfection of cultured primary neurons. (A) Cartoon (not to scale) depicting a neuron (blue) with the nucleus shown in purple. Genetic material is represented as dark red squiggles. Three methods of transfection are shown: magnetically-modified carbon nanotubes (grey) that are used as “spears” to physically drive genetic material into the cell nucleus; a neurotropic virus (red hexagon) with surface peptides (blue) that delivers its contents to the cell in a very efficient fashion; and a generic nanoparticle (shown in beige, with many possible compositions) functionalized with ligands that allow access to the cell nucleus, as well as the genetic material to be delivered. (B) Detail of lipid-encapsulation of CdSe/ZnS QDs and covalent linking of plasmidDNA for transfection. Briefly, trioctylphosphine oxide (TOPO) coated QDs are mixed with lipids in some combination to create a water soluble shell. Maleimide functionalized lipids allow bonding to thiol groups on modified plasmidDNA, and the whole assembly is taken into cell nuclei (adapted with permission from ref. 62, Copyright 2009 American Chemical Society). (C) Dopaminergic neurons from the substantia nigra pars compacta (SNc) of adult mice, transfected with organically modified silica (ORMOSIL) nanoparticles complexed with plasmidDNA expressing EGFP, resulting in green fluorescence. Red is due to TH immunostaining (used with permission from ref. 63, Copyright 2005, National Academy of Sciences, USA). (D) Carbon nanotube spearing. SEM image of a MCF-7 cell membrane after spearing. The dashed ovals indicate the nanotubes (used by permission from Macmillan Publishers Ltd: Nature Methods,⁶⁷ copyright 2005).

The use of nanoparticles to condense, package, and deliver nucleic acids (DNA and RNA) and proteins is a well-established field.⁶² Many types of nanoparticles have been used for in vitro cell transfection, including polymers, Au nanoparticles, QDs (Fig. 8B), and silica particles. Bharali et al. showed that ORMOSIL could transduce neurons in mouse brain after direct injection (Fig. 8C).⁶³ However, very few studies report successful, efficient transfection of primary neurons in culture. Even when particles are taken up by neurons, they are often sequestered and their carried DNA is never expressed.⁶⁴ Those that appear promising target specific receptors or modes of uptake other than endocytosis. For example, Hasadsri et al. used polybutylcyanoacrylate (PBCA) nanoparticles to deliver functional proteins into primary rat hippocampal neurons.⁶⁵ They confirmed the hypothesis that these particles are taken up by the low density lipoprotein (LDL) receptor, a property which also allows them to cross the blood brain barrier. The proteins delivered were β-galactosidase, rhoG (which mediates differentiation), and a specific antibody. The particles were easy to synthesize and protein loading was by simple adsorption.

The use of cell-penetrating peptides on the surface of DNA-bearing nanoparticles may also improve transfection. Suk et al. coated polyethylenimine (PEI)-DNAnanoparticles with either the HIV Tat peptide (YGRKKRRQRRR) or the RGD peptide (H-Arg–Gly–Asp–Cys–OH) and obtained up to 14-fold greater transfection in primary cells than with the nanoparticles alone.⁶⁶ The cells used were not primary cultures, but a SH-SY5Y human neuroblastoma cell line used as is or differentiated.

One of the most promising emerging techniques is what Cai et al. term “carbon nanotube spearing”.⁶⁷SWNTs with ferromagnetic nickel in the tips respond to magnetic fields, and can be accelerated with different amounts of force into cells and even cell nuclei (Fig. 8D). The authors bound plasmidDNA encoding green fluorescent protein (GFP) to the carboxylates of the NTsvia carbodiimide coupling. Efficient transfection of cortical neurons required two steps: agitation of the nanotubes for 20 min in a rotating magnetic field to cause initial cell spearing, then application of a static field for an additional 20 min to pull the tubes deeper into the cells. After 48 h, GFP was expressed in 80% of cells without loss of cell viability, although toxicity was not fully quantified and neurite retraction was seen.

Unlike other magnetic nanomaterial based methods, such as magnetofection, carbon nanotube spearing does not rely upon endocytosis but delivers the cargo directly into the cell cytoplasm or nucleus. This accounts for the increased efficiency and the low amount of plasmidDNA needed, as most endocytosed DNA is destroyed in the endosomes. The authors synthesized the SWNTsviaplasma-enhanced chemical vapor deposition (PECVD), which is not available to all labs, and not all the nanotubes produced could be magnetically driven. However, once synthesized, the methods were straightforward and used only a magnetic stir plate to generate the rotating field.

4. Fluorescence imaging

The stable, bright fluorescence of QDs makes them ideal for imaging cellular events at the single-molecule level. However, delivery of the particles to intracellular targets can be challenging. Preparing conjugates to untested molecules also requires a good deal of characterization and testing and may be hindered by complications such as aggregation or fluorescence quenching. Thus, most of the work on QD imaging with neurons has focused on extracellular events with commercially available QD conjugates: QD–antibodies or QD–streptavidin. Pathak et al. developed staining methods to target neurons and glia specifically with QD–antibodies, which helped to remedy the nonspecific binding seen in many early studies.⁶⁸

QD–antibodies or QD–streptavidin plus antibodies have proven a very useful tool for probing synaptic signaling. Synaptic molecules such as receptors turn over on a much faster timescale than development or plasticity, so that valuable information can be obtained from imaging on scales from seconds to a few hours. Insight into receptor trafficking and turnover is obtained from single particle tracking (SPT),^69,70 which observes the diffusion of single particles in a medium such as the cytoplasm or cell membrane. The trajectories and diffusion coefficients of the particles, which can be labeled receptors or other proteins, give insights into the life-cycle of the molecules and the diffusion properties of the medium. The major limitation of SPT is photobleaching. The longer the fluorophore lasts, the longer it can be tracked. A specific type of “sandwich” construct consisting of a primary antibody to a cellular target, biotinylated Fab fragments, and finally QD–streptavidin has been used in several SPT studies (Fig 9A). The large size of the sandwich is offset by its great chemical and optical stability relative to less fully-functionalized QDs. The first demonstration that these relatively large fluorophores could penetrate the synaptic cleft was given by Dahan et al., who targeted glycine receptors in cultured spinal neurons.⁷¹ The authors were able to track single receptors for 40 min and calculate the diffusion coefficients of the receptors in the membrane, and provided good evidence that the QDs did not significantly slow the receptor kinetics. A similar approach was used by Bats et al. to target AMPA receptors in normal hippocampal neurons as well as those transfected with a stargazin mutant.⁷² This showed that the interaction between stargazin and the protein PSD-95 was necessary for transient stabilization of AMPA receptors (Fig. 10A, B). This stabilization is responsible for the concentration of AMPA receptors at synaptic sites, a phenomenon that had been observed more than ten years earlier by electron and optical microscopy. Similarly, Lévi et al. used the QD–antibody sandwich to track glycine receptors in inhibitory synapses of the spinal cord, which are usually of a mixed GABA/glycinephenotype . They found that there was an excitatory activity-based decrease in lateral diffusion rates of glycine receptors but not GABA receptors. This suggested a mechanism for rapid regulation of the glycinergic component in response to excitatory transmission.⁷³ Bannai et al. have published a protocol for the preparation of these “sandwich” conjugates, optimization of targeting to cell-surface proteins, and image acquisition and analysis.⁷⁴

Fig. 9 Schematic of approximate relative sizes and configurations of commonly used QD conjugates. (A) “Sandwich” construct consisting of a QD core (dark gray), polymer solubilizing and stabilizing shell (light gray), streptavidin (showing 4 active sites), biotinylated Fab, and primary antibody. Only one of the Fab/antibody complexes is shown bound to the QD, even though typically the particle would be coated with them. (B) QD–streptavidin bound to a biotinylated protein, in this case NGF shown approximately to scale. (C) A compact QD consists of a much thinner solubilization shell and a single monovalent streptavidin. The hydrodynamic radius can be as small as 11 nm rather than 30 nm or more as in the sandwich construct.

Fig. 10 The use of single-particle tracking (SPT) to elucidate protein clustering and turnover. (A, B) Measuring stargazin diffusion in neurites (adapted from ref. 72 with permission from Elsevier). (A) Sample images of a neurite coexpressing PSD-95::GFP and Stargazin WT tagged with an N-terminal extracellular HA epitope and QDs. The neurite is visualized in the fluorescence channel of QDs and GFP. The overlay shows the PSD-95::GFP clusters and Stargazin::HA-coupled QDs trajectories. Confined and diffusive periods of movement are shown respectively in red and blue. Scale bar, 2 µm. (B) Mean values ± SEM of the percentage of immobile Stargazin (left) and median diffusion of mobile Stargazin (right). These data allowed the authors to conclude that the fraction of immobile stargazin was greater on PSD-95 clusters, and that the median diffusion of mobile stargazin was also decreased on these clusters. (C, D) Analysis of diffusion of QD-tagged mutant ion channels in a HEK cell (adapted with permission from ref. 77). (C) The QD single-particle track used to generate the mean-square displacement plot indicated in (D).

Another approach, which minimizes the size of the construct somewhat, is to biotinylate a ligand of choice in order to attach it to QD–streptavidin (Fig. 9B). Biotinylation of proteins is straightforward using a number of commercially-available reactive biotins such as biotin hydrazide (aldehyde-reactive), biotin sulfosuccinimidyl ester (amine-reactive), biotin maleimide (thiol-reactive), and others. If the ligand used is endocytosed by neurons, the resulting QD conjugate will be taken up efficiently as well. The conjugated protein (or other molecule) is also likely to retain its biological activity unless the QD is conjugated directly to the active site. An example that has yielded several interesting studies is QD–nerve growth factor (NGF) conjugates. These conjugates are alternatives to radio-labeled (¹²⁵I) NGF and rhodamine–NGF, whose limited temporal and spatial resolution did not permit visualization of how NGF was transported or in which type of organelles (early endosomesvs. more complex structures such as multivesicular bodies). In the first study using QD–NGF, Cui et al. reported that the conjugate retained bioactivity and that it was retrogradely transported from the axon terminal to the cell body inside endosomes.⁷⁵ The majority of endosomes were small vesicles containing only one QD–NGF. The speed of transport and its stop-and-go nature was imaged using single-particle tracking of these endosomes. QD–NGF has also been used to image trafficking of NGF-responsive TrkA receptors.⁷⁶

An interesting and useful twist to the biotinylation approach is that a specific peptide sequence (GLNDIFEAQKIEWHEAR) serves as a “biotin acceptor” for the E. colibiotinligase , BirA (the boldfaced lysine is biotinylated). Thus any protein can be biotinylated at a specific location by inserting this sequence, making it possible to biotinylate membrane proteinsin situ. This approach has been successfully used by Tamkun et al. to biotinylate the Kv2.1 delayed-rectifier ion channel, which is found in the somato-dendritic region of hippocampal neurons, and label it with QD–streptavidin.⁷⁷ The authors tracked channel diffusion in HEK cells, which show channel patterns similar to those in neurons, and concluded that Kv1.2 clustering was due to sub-membrane cytoskeletal structures (Fig. 10C, D).

Native mammalian biotinproteinligases also exist, and commercial cloning vectors are available which link the target sequence to a protein of choice (BioEase, Invitrogen). However, the tag is significantly longer than that needed for biotinylation by bacterial enzymes (72 amino acids), and thus may disrupt the function of complex proteins such as ion channels.

QDs conjugated to small molecules, or coated with a minimal layer of solubilizing ligands without conjugation, are much more sensitive than the constructs described above to chemical and photo-induced degradation. They also show variations in fluorescence intensity with pH, redox potential, and other variables. However, they can have hydrodynamic diameters of ≤15 nm in aqueous solution (depending upon the manufacturer), and their sensitivity can be used as a feature to detect specific processes. Zhang et al. used unconjugated, carboxylate-terminated QDs to distinguish between kiss-and-run and full-collapse fusion of synaptic vesicles in rat hippocampal neurons.⁷⁸ Deacidification of vesicles upon fusion caused QD fluorescence to increase by a detectable amount (Fig. 11). The QDs were small enough to be loaded into vesicles without changing the vesicle size or trafficking, but large enough to remain inside the vesicle during kiss-and-run.

Fig. 11 The use of pH-dependent QD photoluminescence to probe vesicular recycling. The QDs show a 15% decrease in emission intensity when passing from a neutral cytoplasmic pH to an acidic vesicular pH. The cartoons show hypothetical QD signals arising from pH dependence for kiss-and-run (K&R) vs. full-collapse fusion (FCF). (Reprinted from ref. 78 with permission from AAAS).

To further reduce QD size, Howarth et al. have developed “compact QDs” consisting of a CdSe–ZnCdS shell and a thin functionalizing layer to solubilize QDs and prevent nonspecific binding while maintaining the hydrodynamic diameter ∼11 nm (Fig. 9C).⁷⁹ The authors showed that these QDs penetrated more effectively into synapses than commercial QDs. They also developed a simple method for isolating monovalent QDs bearing a single streptavidin conjugate. The method involves separation of conjugates by agarose gel electrophoresis and should be applicable to any protein. The resulting conjugates had hydrodynamic diameters only 1.2 nm greater than the unconjugated QDs.

Regardless of conjugate, all commonly-used QDs show fluorescence intermittency, or “blinking”—stochastic periods of emissive (“on”) and dark (“off”) states. This poses a problem for single particle tracking, since a particle that is dark cannot be tracked. (Interestingly, organic fluorophores also show intermittency, but this is usually not observed since the single molecules are not bright enough to track and the blinking occurs on rapid time scales.) Currently the best approaches to dealing with blinking are analytic—that is, using data analysis techniques to separate the intensity fluctuations due to particle photophysics from those due to motion. Bachir et al. show that k-space image correlation spectroscopy, called kICS, can be used to extract accurate diffusion coefficients and QD blinking probability distribution functions.⁸⁰

QDs made of CdSe with thick CdS shells have been reported to show greatly reduced blinking.⁸¹ The diameter of these particles was ∼13 nm. Even more recently, non-blinking nanocrystals of an alloyed core–shell material (CdZnSe–ZnS) were reported;⁸² the particles were only slightly larger than ordinary core–shell QDs (Fig. 12). In both cases, fluorescence imaging was performed in organic solvents, and biological applications of these materials have not yet been established.

Fig. 12 Blinking and non-blinking QDs. Typical intensity vs. time traces for commercial CdSe–ZnS in water (A) and alloyed CdZnSe–ZnS in toluene (B) (adapted by permission from Macmillan Publishers Ltd: Nature, ref. 82 copyright 2009).

Non-fluorescent imaging techniques using other types of nanoparticles deserve mention. Latex beads and metal nanoparticles used for Rayleigh intensity scattering are too large for most neuroscience applications (40 nm for Au particles; micron sized for beads). An alternative approach exploits the photothermal effect of metal nanoparticles—that is, that most light absorbed by these particles is converted to heat. Using a photothermal interference contrast technique, Lasne et al. were able to imagine AMPA receptors in live neurons after labeling with 5 nm gold particles.³⁸

5. Electron microscopy and correlated electron-fluorescence imaging

Although not so electron-dense as metal nanoparticles, CdSe possesses sufficient electron density to permit visualization using transmission electron microscopy (TEM) (Fig. 13A). Unstained QDs can be difficult to distinguish from structures like ribosomes, but standard methods can be used to improve contrast. Dahan et al. used a silver enhancing kit to observe QD-tagged glycine receptors.⁷¹ Such kits are usually used to enhance immunogold labeling by exploiting the ability of metals to transfer electrons to silver ions and thus precipitate metallic silver. Because semiconductors are capable of the same reaction, QDs are also enhanced by silver; the stain may even distinguish photooxidized from undamaged QDs.

Fig. 13 Appearance of nanoparticles under different types of electron microscopy. (A) TEM of ultrathin section of a multilamellar lipid vesicle containing CdSe core QDs. The section is stained with uranyl acetate and osmium tetroxide, making the lipid layers appear gray. The hydrophobic QDs (arrows indicate some examples) appear nearly black under these conditions. (B) SEM image of unstained Staphylococcus aureus bacteria coated with CdSe–ZnS QDs. (C, D) STEM images of E. coli cells with (C) CdTe QDs and (D) TiO₂ nanoparticles. The samples are stained with uranyl acetate and osmium tetroxide. (Images courtesy R. Mielke.)

An excellent technique for the unambiguous detection of QDs is energy-filtered TEM (EFTEM), also called electron spectroscopic imaging (ESI). EFTEM generates maps of elemental distributions and can distinguish elements that have different ionization edges. The M_IV,V ionization edge of cadmium permits it to be distinguished from nitrogen. Nisman et al. have used this technique to illustrate subcellular QD distribution in fixed, permeabilized, thin-sectioned cells labeled with a QDantibody.⁸³ It can also be used to map the subcellular distribution of naturally-occurring elements. Aronova et al. used this method to measure mitochondrial calcium concentrations in recently-stimulated, freeze-dried neurons.⁸⁴

Correlative fluorescence and electron microscopy, in which the same structures are imaged by both methods, remain challenging for several reasons. Fluorescent probes may be rendered electron-dense by sequential labeling, diamino-benzidine oxidation, or silver-intensification. However, these methods require multiple steps and may be unreliable. The features that allow for multiple labeling in fluorescence studies (i.e., different colors) are not carried over to electron microscopy, so simultaneous labeling of multiple targets is very difficult. In a study using 80 µm sections of mouse cerebellum, Giepmans et al. demonstrated that different colors of QDs could be used for correlated, multi-target imaging using fluorescence and TEM.⁸⁵ Because the different colors of QDs are not only different sizes but also different shapes (the reddest emitting particles are rod-shaped), fluorescence images could be readily compared with EM ultrastructure. They illustrated neuronal vs. glial staining using double labeling with QDantibodies to glial fibrillary acidic protein (GFAP) and to the Purkinje cell marker IP3R. They also illustrated triple labeling using GFAP, neurofilamentprotein 68 (NFP68), and the cerebellar protein Cx43. An important caveat in this work was that the authors found a complete destruction of QD fluorescence by osmium tetroxide. Thus osmication cannot be used on samples for correlated imaging.

Semiconductor and metal nanoparticles may also be used for biological labeling using scanning electron microscopy (SEM). Under SEM, the particles appear bright relative to surrounding biological tissue (Fig. 13B). Pavone et al. used immunogold to image the distribution of serotonin transporters in buffalo brain.⁸⁶ It is easy to mistake particles for cellular structures such as villi with typical SEM, especially because individual QDs or immunogold particles are too small to resolve.

There are methods to scan through thick biological samples without the need for thin sectioning. Aoyama et al. demonstrated the usefulness of scanning transmission electron microscopy (STEM) tomography of uranyl acetate stained, Epon-embedded hippocampal neuron slices as thick as 1 µm.⁸⁷ Under these conditions, nanoparticles appear bright (Fig. 13C, D). Another approach is to use ion abrasion in conjunction with SEM. Heymann et al. were able to observe subcellular structures labeled with nanogold and QDs using this method.⁸⁸

Finally, with the appropriate instrumentation electron microscopy techniques can be applied to living, or at least fully-hydrated samples. Environmental scanning electron microscopy (ESEM) is performed in a chamber with up to 8 mbar of water pressure, preserving hydration. Bacteria and plant cells are readily imaged with ESEM, as they do not collapse when the medium is removed. Mammalian cells, particularly neurons, pose a greater challenge; however, this technique is ideal for some types of organotypic neuronal culture. Uroukov and Patton showed surface features of chick embryo brain spheroids using this technique.⁸⁹ For STEM, the design of thin environmental chambers can permit nanoscale resolution of cells in liquid.⁹⁰

6. Electrical recording

Arrays of dozens to hundreds of electrodes interfaced with sparse cultures, dense cultures, or brain slices are used to elucidate neuronal networks in vitro and to screen neuroactive drugs. These are called multielectrode arrays (MEAs) (Fig. 14A). The electrodes in MEAs must be capable of both stimulation and recording from neurons or tissue. The sizes of the electrodes range from millimetres to tens of microns, and ideally permit recordings for weeks at a time. Two major problems hinder these experiments. First is low signal to noise caused by large impedance electrodes. The smaller the size of conventional metal electrodes, the larger the impedance, hence the higher the thermal noise. This often makes resolution of even action potentials a challenge. Second is the biological–electrode interface, which changes in neuronal cultures as they differentiate and as glial cells proliferate. Glia often “shield” neurons from the electrodes and prevent recording. The ways in which nanotechnology promises to improve these recordings can thus be grouped into two primary categories: creation of small, low-noise electrodes; and improvement of the neuron–electrode interface by the use of novel materials or nanoscale patterning.

Fig. 14 Substrates for electrical stimulation of and recording from cultured neurons. (A) Commercial multielectrode array (Multichannel systems) plated with a sparse culture of GFP-transfected rat hippocampal neurons. (B) Neuron grown on a SWNT mesh (image courtesy L. Ballerini, http://www.neuronano.net). (C) Close-up of neurite on SWNT mesh (image courtesy G. Cellot and F. Toma). (D) Neurons and glia on a carbon nanotube mesh electrode (from ref. 93, reproduced by permission of The Royal Society of Chemistry). (E) PC12 cells grown on a vertically aligned carbon nanotube network (image used with permission from ref. 95, copyright 2007 IEEE). (F) System for photostimulation of neurons using HgTe nanoparticles (adapted with permission from ref. 97, copyright 2007 American Chemical Society). Left, schematic of a neuron growing on a tightly-packed, layer-by-layer deposited composite film consisting of nanoparticles, polylysine, and RGD peptides. (Right) AFM image of the substrate on a Si wafer.

Carbon nanomaterials, particularly SWNTs, have shown considerable promise as nanoscale electrodes for stimulation and recording of neural tissue. Mesh deposition of SWNTs is as simple as pipetting hepatanal/sarcosine-functionalized NTs onto the substrate. Liopo et al. showed that mesh-deposited SWNTs supported attachment and growth of glioma cells, and investigated the biocompatibility of a number of different surface coatings.⁹¹ They were the first to demonstrate that individual cells could be electrically stimulated by the SWNTs, although they did not develop a model of cell–nanotube coupling. Since then, their findings have been confirmed and an increase in signaling efficacy of neuronal networks on nanotube matrices has been observed. Several groups have made efforts to model the electrical coupling between the NTs and cells that explain these observations. Mazzatenta et al. cultured hippocampal neurons on SWNT-treated glass coverslips (Fig. 14B, C) and performed single-cell patch-clamp recordings as well as SWNT stimulations via an Ag electrode coupled to the nanotube film.^92a This led to electrical equivalent models of the nanotube–cell interface that suggested a direct resistive coupling. Cellot et al. also support the idea of direct couplings based upon TEM imaging of interfaces and on observed short-circuiting of neuronal electrical signals through a SWNT network.^92b These findings have been implemented in the design of multielectrode arrays consisting of SWNT mesh deposited onto micropatterned electrodes (Fig. 14D).⁹³

Vertically-aligned NTs may offer certain advantages relative to mesh. They may be able to couple more effectively to cells without creating short-circuit pathways, greatly simplifying electrical modeling. Massobrio et al. presented a model for a vertical NT/ion-sensitive field-effect transistor (ISFET)–neuron interface under different coupling conditions, and predicted that under strong coupling, recordings of quality approaching that of intracellular recordings might be obtained.⁹⁴ A significant problem is that vertically-aligned SWNTs are fragile and break under biological conditions. Nguyen-Vu et al. reported a vertical nanofiber array stabilized by polypyrrole, where PC12 cells grew on collapsed microbundles of the fibers⁹⁵ (Fig. 14E).

Nanoparticles of varying compositions have also been used to stimulate neurons by taking advantage of their photocatalytic properties. Photo-induced stimulation of neurons could be an alternative to electrical stimulation that is less invasive and more subject to spatial control. Zhao et al. created nanostructured electrodes containing PbSe particles that were excited with 830 nm light.⁹⁶ When placed into hippocampal or olfactory brain slices (in the same way as a patch electrode) and illuminated with a Ti-sapphire laser, the electrodes stimulated the cells in a similar way to direct current injection. The nanoparticles could be placed inside the electrode so that there was no direct contact with the neurons, suggesting this was an effect of electric field rather than electron transfer. Similarly, Pappas et al. used a thin film of HgTe nanoparticles for photostimulation of a neuronal cell line (NG108)⁹⁷ (Fig. 14F). The nanoparticles were stimulated with 532 nm laser light and a resistive coupling of the photocurrent to cell stimulation was observed, with generation of action potentials.

7. Differentiation and patterned growth

Neuronal precursors at every level, from stem cells to neuroendocrine cancer cells, are differentiated in vitro for a variety of applications. Numerous suppliers offer cells that can be differentiated into primary neurons, and new methods of immortalizing human CNS cells are continually being developed. The most obvious advantage of these cells over primary cultures is to permit the use of immortalized cell lines in experiments, which obviates the need for animals, greatly reducing cost and required expertise. The ability to differentiate dividing precursors also allows transfection, axon guidance, or other manipulations to be done before the cells become post-mitotic. Finally, study of the differentiation process is key to understanding neuroregeneration.

Regardless of their position along the differentiation axis, cell lines are difficult to turn into primary neurons, and differentiation remains a hit-or-miss procedure. Special media containing trophic factors are sold to aid the process, but this is a very poor model for what happens in vivo.

Culturing cells on thoroughly characterized, ECM-like nanopatterned substrates in vitro may be able to provide useful clues into how nanoscale chemical and physical cues direct neural differentiation. It is therefore of the utmost importance to know what signals are being presented to and recognized by the cells. Self-assembled designer peptide scaffolds were used by Gelain et al. to form 3D cell culture systems modified with a number of motifs to impart a range of functionalities such as cell adhesion, differentiation and bone marrow homing signals⁹⁸ (Fig. 15A). The self-assembly process could be controlled to some degree by ionic concentration in solution. This design, with individual peptides modified with single functionalities, allowed for a modular network with quantified fibers and functionalities, providing a distinct advantage over animal-derived biomaterials. By using molecular markers and monitoring gene expression, it was possible to investigate the role of network design on neural development.

Fig. 15 (A) A nanofibrous network consisting of self-assembled designer peptides modified with bone marrow homing factor, self-assembled in PBS. The nanofibers are ∼10 nm and the pores ∼5–200 nm. Addition of functional motifs does not prevent self-assembly or result in aggregation, as can be seen in Matrigel™ (image from ref. 98, used with permission). (B) Schwann cells on aligned PCL–gelatin nanofibers obtained after several days of cell culture. Cell proliferation was somewhat improved in randomly oriented versus aligned nanofiber networks, perhaps due to differences in roughness or porosity (reprinted from ref. 100 with permission from Elsevier). (C) Light microscope images of microcontact printed polylysine–fluorescein with and without neurons. The stripes are 10 µm apart. Upper left: substrate under brightfield. Upper right: substrate under epifluorescence. Lower left: fluorescently labeled dorsal root ganglion neurons (DRGs) on substrate. Lower right: labeled Schwann cells on substrate (images courtesy P. Grutter and H. Bourque). (D) Mouse ganglia growing on a nanoimprinted positive pattern in a silicon wafer. The ridges are 100 nm high and 100 nm wide (approximately the dimensions of the axons), and it can be seen that axonal outgrowth exhibits a preference for the elevated ridges rather than the grooves between (reprinted from ref. 104 with permission from Elsevier).

Koh et al. used electrospinning to synthesize nanoscale scaffolds of poly(L-lactic acid) (PLLA), a synthetic polymer that is biocompatible, but hydrophobic and lacking biological recognition sites for cellular adhesion.⁹⁹ To combat these substantial limitations, they coupled laminin to the PLLA scaffold in one of several ways to improve hydrophilicity and to serve as a biochemical beacon for neurite outgrowth. They found that PLLA–laminin scaffolds supported PC12 cell adhesion and that nerve growth factor (NGF) could be used to incite neurite outgrowth. The authors determined that blending PLLA and laminin prior to electrospinning was more conducive to neuronal growth than either covalent bonding or physical adsorption of laminin.

Electrospinning can also be used to create aligned nanofiber scaffolds for directed growth. Gupta et al. used this technique with blended polycaprolactone (PCL)–gelatin (Fig. 15B) to provide a substrate for the growth of Schwann cells.¹⁰⁰ They showed that while cell proliferation was somewhat reduced on aligned versus random PCL–gelatin fibers, either arrangement was a considerable improvement over fibers containing PCL alone, reinforcing the notion of blending synthetic polymers with biomaterials for a synergistic approach. In this study, great care was taken in the characterization of the networks, including determination of porosity, tensile properties, hydrophilicity, and alignment. Cell viability was also well characterized, with proliferative, morphological, and metabolic investigation.

Printed patterns are less physiologically relevant than 3D nanofibrous networks, but are more standardized in design. Microcontact printing of cell-adhesive peptides or polylysine can restrict the growth of neurons to particular regions of a substrate in both two¹⁰¹ and three¹⁰² dimensions (Fig. 15C). In order to fully study the interactions of proteins and cells, whole proteins can be used in the place of short adhesive peptides.¹⁰³

Along with chemical cues, topographical features of the local cellular environment have been shown to play a role in neurite outgrowth. Johansson et al. employed nanoimprint lithography and electron beam lithography to create nanopatterns etched into silicon wafers.¹⁰⁴ These patterns consisted of parallel grooves of various nanoscale widths, depths and spacings, and their effects on axonal growth of neurons in Matrigel™ were explored. Upon stimulation of outgrowth with NGF, the authors found that neurons exhibited contact guidance on all patterns, and the neurites preferred to extend along elevated portions of the substrate rather than in the grooves (Fig. 15D). Axons have a much more meandering advance than whole cells, making it more challenging to evaluate the extent of guidance. FFT analysis showed differences in neurite alignment on different patterns, and certain preferences in pattern dimension and “elevation” for axonal outgrowth of autonomic ganglia and sensory nerve fibers. In particular, processes were best supported by ridges of width >100 nm.

Both nanopatterning and electrical recording techniques can be combined to create a defined neuronal network with individually-addressable cells (or even axons and dendrites). Patolsky et al. used silicon nanowire field effect transistors (FETs), with active junction areas of about 0.01 to 0.02 µm², to record from cultured neurons.¹⁰⁵ The devices were passivated with silicon nitride to prevent corrosion under cell-culture conditions, and then directed growth was obtained by lithographic patterning of polylysine in 2 µm wide channels. Electrical recording showed reproducible spikes and the resolution of signal propagation in both axons and dendrites (Fig. 16). The same group has had similar success with myocytes, and describe a modular approach to interfacing FETs with living cells that involves cell culture on polydimethylsiloxane (PDMS) and subsequent interfacing with the electronic device.¹⁰⁶ This simplified method should make combined patterning and electrical recording accessible to many laboratories.

Fig. 16 Nanowire (NW) recording and stimulation of neuronal axon signals from patterned arrays (adapted from ref. 105 with permission from AAAS). (A) (Top) Optical image of a high-density device array, which is located within the open blue rectangle, with a polylysine pattern consisting of two diamonds connected by a horizontal linear stripe that runs across the center of the array. The linear array consists of 150-devices with a 400 nm device pitch. The blue rectangle is 127 µm × 35 µm. (Bottom left) Low magnification SEM image of the 150-device linear array; the scale bar is 8 µm. (Bottom right) High resolution SEM image showing 4 sequential NW devices; scale bar is 400 nm. The image was recorded from the region in the low-resolution image indicated by the open red rectangle. (B) Optical images recorded at increasing magnifications (top to bottom) of a rat cortical neuron with its axon aligned on a 150-device NW linear array and the soma aligned along the lower right edge of a diamond. The open green rectangle is 52 µm × 15 µm. (C) (Top) Intracellular (IC) potential of an aligned cortex neuron (after 6 days in culture) during stimulation with a 500 ms long current injection step of 0.1 nA. (Bottom) Time-correlated signal from axon measured with a p-type silicon nanowire device. (D) Local nanowire-axon stimulation and correlated IC electrical recording of a cortex neuron. IC plots were recorded following a rectangular biphasic train of stimuli with amplitudes of (top) 0.5 V, (middle) 0.3 V, and (bottom) 0.5 V after bath application of 0.5 µM tetrodotoxin (TTX). Green box highlights the train of rectangular biphasic voltage pulses.

8. Toxicity concerns specific to neurons

The toxicology of nanomaterials to cells in vitro has been widely studied and reviewed, and includes oxidative mechanisms both direct and indirect as well as physical damage to cell structures.^107–109 In all quantum dot toxicity studies, it is crucial to distinguish between bare core and core–shell QDs. Breakdown of core QDs that contain cadmium leads to the release of Cd²⁺ ions and resulting massive cytotoxicity,¹¹⁰ whereas breakdown of core–shell particles capped with ZnS leads only to relatively nontoxic zinc and sulfur.¹¹¹ In all nanotoxicology studies, the importance of synthesis byproducts and solvents must be carefully considered as possible confounding factors.

In this review we will focus only on specific changes seen in cultured neurons, as sub-lethal toxicity can greatly affect the properties of a neuronal culture, from action potential firing to synapse formation to plasticity. Even agents that appear to increase cell viability can be suspect, as they may be suppressing excitatory transmission. A classic example of this is the effect of dimethylsulfoxide (DMSO) on motor neurons. At concentrations of >0.02%, fifty times lower than is commonly used to deliver drugs, DMSO has a profound effect on glutamatergic neurotransmission.¹¹² Thus any time an agent is introduced into the cell for labeling or transfection, it is important to make sure that it does not substantially alter the neuron's behavior. Different types of neurons show different levels of susceptibility to such changes, so each new nanomaterial will need to be tested in the cultures under study before its innocuousness can be assumed. The existing data are very incomplete, but a few studies exist that suggest the types of effects nanomaterials may have on cultured neurons.

Tang et al. reported that bare core CdSe QDs caused dose-dependent death in cultures of rat hippocampal neurons.¹¹³ Levels of cytoplasmic calcium were elevated, caused both by extracellular calcium influx and internal calcium release from the endoplasmic reticulum. Interestingly, the authors also reported effects on voltage gated sodium channels (VGSCs): enhanced activation and inactivation of I_Na, prolonged the time course of activation, slowed I_Na recovery, and a reduced fraction of available VGSCs. In a subsequent paper, the same group used specific antagonists to show that the QDs caused extracellular calcium influx through N-type voltage-gated calcium channels (VGCCs) and VGSCs.¹¹⁴Extracellular Na⁺ could also influx through VGSC, further triggering Ca²⁺ release from mitochondria.

Zhao et al. reported similar effects when acutely dissociated rat CA3 neurons were exposed to nanoscale ZnO.¹¹⁵ The particles caused an influx of Na⁺viaVGSCs, causing calcium influx, reversed Na⁺–Ca²⁺ exchange, potassium efflux, and glutamate release. This caused cell death by excitotoxicity as well as apoptosis caused by the loss of cytoplasmic potassium.

Although toxic effects of these highly reactive particles are not surprising, these results give some idea of the types of measurements that need to be done to quantify effects on neurons. The exact interactions that cause ion channel malfunction have not yet been explored. They may be manifestations of oxidative stress as is found in other cell types exposed to QDs, or they may represent as yet undescribed interactions between the particles and the neuronal membrane.

The controversy over the effects of C₆₀ illustrates some of the difficulty in neurotoxicology studies. It is certain that C₆₀ can cause oxidative damage to cells, even if ROS generation and lipid peroxidation do not occur. Nonetheless, water-solubilized fullerenes are potent free radical scavengers. Dugan et al. were the first to report that carboxyfullerenes could protect mouse cortical neuronsin vitro from excitotoxic shock due to application of AMPA or NMDA.¹¹⁶ The cells exposed to the fullerenes also showed reduced apoptosis in response to serum deprivation or β-amyloid. The authors attributed the effect to the removal of hydroxyl radicals and superoxide. However, Jin et al. found a similar effect in fetal rat brain cultures, which they attributed to direct blockade of glutamate receptors.¹¹⁷ No similar effect on GABA receptors was observed. Thus the mechanisms of cytotoxicity and of cytoprotection of fullerenes remain to be fully elucidated.

Similarly, Schubert et al. reported that cerium and yttrium oxidenanoparticles protected a rodent neuronal cell line from glutamate-induced excitotoxic stress.¹¹⁸ They attributed this to direct anti-oxidant effects, although direct measurements of redox reactions with ROS and particles were not performed. The ROS measurements inside cells were also done with dichlorofluorescein-di-acetate, which we have found can be unpredictable in the presence of nanoparticles. Although all of these studies are intriguing, firm conclusions about the potential of nanoparticles for protectingneurons cannot yet be drawn.

9. Conclusions and future directions

Manufactured nanomaterials have been shown to have unexpected, potentially very useful interactions with cells in general and neurons in particular. As we begin to understand these interactions, we will be able to design the materials for specific applications, and possibly obtain new sensing and recording techniques that will truly enable us to quantify neuronal networks in vitro.

Working with primary neurons still remains highly challenging from the standpoints of both biology and physics. Gene transfer and transfection in dissociated cultures and slices are far from routine. Most investigators make viral constructs or use microinjection, both of which require significant skill and infrastructure. Protein delivery is even more difficult. Given the strides nanotechnology has made in these directions, rapid improvement in these areas is hopeful, especially for specific types of cells or cultures.

A revolution is needed in neuronal recording techniques if we are to truly elucidate network structure and plasticity. There is still no replacement for patch-clamp electrodes for low-noise recording. A method for automatic, low-noise recording from dozens to hundreds of cells, with time courses of hours to days, would mark a new era in our ability to investigate networks. SWNTs have shown promise for supporting neuronal growth and permitting electronic coupling, but their noise levels are still too high for resolution of single synaptic potentials.

It is possible that optical recording techniques will soon allow us to sense membrane potential with at least action potential sensitivity. Although QDs have a large Stark effect, this is not particularly useful for this purpose as it is difficult to see in ensembles of QDs at room temperature, and it would require embedding the QDs within the cell membrane. Since the QDs are at least as large as the membrane, this is difficult to without killing cells. A FRET- or electron-transfer based system that could sense membrane potential by means of a small tail within the membrane, leaving the nanoparticle external, is ideal. If the QDs act as FRET acceptors rather than donors, they do not need to feel the electric field directly, but can rely upon the voltage-sensitivity of a donor dye. Kloepfer et al. showed that the voltage sensitivity of the dye di-4-ANEPPS could be enhanced by QDs acting as FRET acceptors, but this system has not yet been proven to work in cells.¹¹⁹

Another very useful tool would be sensors for specific ions, particularly calcium, that could work for many days at a time. This would allow for elucidation of more chronic excitotoxicity processes such as those that occur in motor neurons. QDs with sensitivity to specific ions have been created, and a general model for their construction based upon FRET principles was proposed by Yildiz et al.¹²⁰ Such probes would need to be delivered into the cell cytoplasm and not be trapped in endosomes. This can be accomplished by microinjection; however, this is a tedious and invasive process. QDs coated with certain molecular patterns have been shown to pass into cells and bypass endosomes,¹²¹ but this functionalization would have to be coupled with the sensing system in order to work.

On the imaging front, the full potential of QDs and other nanoparticles has not been exploited. The use of these particles for multiplexed imaging is an important future avenue for correlating ultrastructure and light microscopy. The ability of QDs to emit cathodoluminescence might also be used to obtain completely correlated, simultaneous fluorescence and electron images, but so far this has not been demonstrated in biological samples.¹²²

The constant discovery, fabrication, and characterization of new nanomaterials will also lead to advances as yet unforeseen. Novel shapes (quantum rods, carbon nanohorns), allotropes (nanodiamond), sizes (Au particles <1 nm), and multifunctional composites have all showed promise for biomedical applications.¹²³

Acknowledgements

We thank the CIHR Nanomedicine and Regenerative Medicine Seed Grant, the NSERC Individual Discovery program, and the U.S. EPA for funding.

Notes and references

1. P. Salmon and D. Trono, Curr. Protoc. Neurosci., 2006, ch. 4 , Unit 4.21.
2. T. J. McCown, Curr. Gene Ther., 2005, 5, 333–338.
3. B. K. Berges, J. H. Wolfe and N. W. Fraser, Mol. Ther., 2007, 15, 20–29.
4. M. J. Saxton and K. Jacobson, Annu. Rev. Biophys. Biomol. Struct., 1997, 26, 373–399.
5. M. Segal, J. Neurosci. Methods, 1995, 59, 25–29.
6. A. M. Aravanis, J. L. Pyle and R. W. Tsien, Nature, 2003, 423, 643–647.
7. S. J. Royle, B. Granseth, B. Odermatt, A. Derevier and L. Lagnado, Methods Mol Biol, 2008, 457, 293–303.
8. G. Q. Bi and M. M. Poo, Journal of Neuroscience, 1998, 18, 10464–10472.
9. C. J. Milligan, J. Li, P. Sukumar, Y. Majeed, M. L. Dallas, A. English, P. Emery, K. E. Porter, A. M. Smith, I. McFadzean, D. Beccano-Kelly, Y. Bahnasi, A. Cheong, J. Naylor, F. N. Zeng, X. Liu, N. Gamper, L. H. Jiang, H. A. Pearson, C. Peers, B. Robertson and D. J. Beech, Nat. Protoc., 2009, 4, 244–255.
10. P. Schaffer, H. Ahammer, W. Muller, B. Koidl and H. Windisch, Pfluegers Arch., 1994, 426, 548–551.
11. B. A. Griffin, S. R. Adams, J. Jones and R. Y. Tsien, Methods Enzymol., 2000, 327, 565–578.
12. G. Guerrero, M. S. Siegel, B. Roska, E. Loots and E. Y. Isacoff, Biophys. J., 2002, 83, 3607–3618.
13. B. J. Baker, H. Lee, V. A. Pieribone, L. B. Cohen, E. Y. Isacoff, T. Knopfel and E. K. Kosmidis, J. Neurosci. Methods, 2007, 161, 32–38.
14. E. Claverol-Tinture, M. Ghirardi, F. Fiumara, X. Rosell and J. Cabestany, J. Neural Eng., 2005, 2, L1–7.
15. M. D. Boehler, B. C. Wheeler and G. J. Brewer, Neuron Glia Biol., 2007, 3.
16. G. A. Silva, Nat. Rev. Neurosci., 2006, 7, 65–74.
17. W. C. Chan and S. Nie, Science, 1998, 281, 2016–2018.
18. M. Bruchez, Jr., M. Moronne, P. Gin, S. Weiss and A. P. Alivisatos, Science, 1998, 281, 2013–2016.
19. L. E. Brus, J. Chem. Phys., 1984, 80, 4403–4409.
20. A. Hagfeldt and M. Gratzel, Chem. Rev., 1995, 95, 49–68.
21. R. Vogel, P. Hoyer and H. Weller, J. Phys. Chem., 1994, 98, 3183–3188.
22. C. Burda, T. C. Green, S. Link and M. A. El-Sayed, J. Phys. Chem. B, 1999, 103, 1783–1788.
23. S. Xu, J. Zieler and T. Nann, J. Mater. Chem., 2008, 18, 2653–2656.
24. J. Jasieniak and P. Mulvaney, J. Am. Chem. Soc., 2007, 129, 2841–2848.
25. C. Wang, M. Shim and P. Guyot-Sionnest, Science, 2001, 291, 2390–2392.
26. M. A. Hines and P. Guyot-Sionnest, J. Phys. Chem., 1996, 100, 468–471.
27. B. O. Dabbousi, J. RodriguezViejo, F. V. Mikulec, J. R. Heine, H. Mattoussi, R. Ober, K. F. Jensen and M. G. Bawendi, J. Phys. Chem. B, 1997, 101, 9463–9475.
28. A. R. Kortan, R. Hull, R. L. Opila, M. G. Bawendi, M. L. Steigerwald, P. J. Carroll and L. E. Brus, J. Am. Chem. Soc., 1990, 112, 1327–1332.
29. S. F. Wuister, C. Donega and A. Meijerink, J. Phys. Chem. B, 2004, 108, 17393–17397.
30. J. M. Tsay, M. Pflughoefft, L. A. Bentolila and S. Weiss, J. Am. Chem. Soc., 2004, 126, 1926–1927.
31. I. L. Medintz, A. R. Clapp, H. Mattoussi, E. R. Goldman, B. Fisher and J. M. Mauro, Nat. Mater., 2003, 2, 630–638.
32. I. L. Medintz and H. Mattoussi, Phys. Chem. Chem. Phys., 2009, 11, 17–45.
33. P. V. Kamat, J. Phys. Chem. B, 2002, 106, 7729–7744.
34. M. A. El-Sayed, Acc. Chem. Res., 2001, 34, 257–264.
35. Z. X. Wang and L. N. Ma, Coord. Chem. Rev., 2009, 253, 1607–1618.
36. P. K. Jain, X. Huang, I. H. El-Sayed and M. A. El-Sayad, Plasmonics, 2007, 2, 107–118.
37. C. Cao and S. J. Sim, Lab Chip, 2009, 9, 1836–1839.
38. D. Lasne, G. A. Blab, S. Berciaud, M. Heine, L. Groc, D. Choquet, L. Cognet and B. Lounis, Biophys. J., 2006, 91, 4598–4604.
39. Q. Y. Cai, S. H. Kim, K. S. Choi, S. Y. Kim, S. J. Byun, K. W. Kim, S. H. Park, S. K. Juhng and K. H. Yoon, Invest. Radiol., 2007, 42, 797–806.
40. J. Aldana, Y. A. Wang and X. G. Peng, J. Am. Chem. Soc., 2001, 123, 8844–8850.
41. Y. Xiao, W. G. Telford, J. C. Ball, L. E. Locascio and P. E. Barker, Nat. Methods, 2005, 2, 723.
42. I. L. Medintz, H. T. Uyeda, E. R. Goldman and H. Mattoussi, Nat. Mater., 2005, 4, 435–446.
43. C. Cioffi, S. Campidelli, F. G. Brunetti, M. Meneghetti and M. Prato, Chem. Commun., 2006, 2129–2131.
44. V. Georgakilas, K. Kordatos, M. Prato, D. M. Guldi, M. Holzinger and A. Hirsch, J. Am. Chem. Soc., 2002, 124, 760–761.
45. J. J. Pancrazio, Nanomedicine, 2008, 3, 823–830.
46. M. Satoh and I. Takayanag, J. Pharmacol. Sci., 2006, 100, 513–518.
47. P. C. Ke and R. Qiao, Journal of Physics-Condensed Matter, 2007, 19.
48. J. L. Gilmore, X. Yi, L. Quan and A. V. Kabanov, J. Neuroimmune Pharmacol., 2008, 3, 83–94.
49. E. M. Hotze, J. Labille, P. Alvarez and M. R. Wiesner, Environ. Sci. Technol., 2008, 42, 4175–4180.
50. D. Y. Lyon and P. J. J. Alvarez, Environ. Sci. Technol., 2008, 42, 8127–8132.
51. J. Choi, N. S. Wang and V. Reipa, Langmuir, 2007, 23, 3388–3394.
52. H. Ow, D. R. Larson, M. Srivastava, B. A. Baird, W. W. Webb and U. Wiesner, Nano Lett., 2005, 5, 113–117.
53. I. Roy, T. Y. Ohulchanskyy, D. J. Bharali, H. E. Pudavar, R. A. Mistretta, N. Kaur and P. N. Prasad, Proc. Natl. Acad. Sci. U. S. A., 2005, 102, 279–284.
54. V. P. Torchilin, Adv. Drug Delivery Rev., 2006, 58, 1532–1555.
55. T. Govender, E. Ojewole, P. Naidoo and I. Mackraj, Drug Delivery, 2008, 15, 493–501.
56. J. Venugopal, S. Low, A. T. Choon and S. Ramakrishna, J. Biomed. Mater. Res., Part B, 2008, 84b, 34–48.
57. R. S. Kane, S. Takayama, E. Ostuni, D. E. Ingber and G. M. Whitesides, Biomaterials, 1999, 20, 2363–2376.
58. P. Berndt, G. B. Fields and M. Tirrell, J. Am. Chem. Soc., 1995, 117, 9515–9522.
59. E. C. Ohki, M. L. Tilkins, V. C. Ciccarone and P. J. Price, J. Neurosci. Methods, 2001, 112, 95–99.
60. J. A. O'Brien and S. C. R. Lummis, Trends Biotechnol., 2007, 25, 530–534.
61. J. M. Bergen, I. K. Park, P. J. Horner and S. H. Pun, Pharm. Res., 2008, 25, 983–998.
62. M. A. Mintzer and E. E. Simanek, Chem. Rev., 2009, 109, 259–302.
63. D. J. Bharali, I. Klejbor, E. K. Stachowiak, P. Dutta, I. Roy, N. Kaur, E. J. Bergey, P. N. Prasad and M. K. Stachowiak, Proc. Natl. Acad. Sci. U. S. A., 2005, 102, 11539–11544.
64. J. S. Suk, J. Suh, S. K. Lai and J. Hanes, Exp Biol Med (Maywood), 2007, 232, 461–469.
65. L. Hasadsri, J. Kreuter, H. Hattori, T. Iwasaki and J. M. George, J. Biol. Chem., 2009, 284, 6972–6981.
66. J. S. Suk, J. Suh, K. Choy, S. K. Lai, J. Fu and J. Hanes, Biomaterials, 2006, 27, 5143–5150.
67. D. Cai, J. M. Mataraza, Z. H. Qin, Z. P. Huang, J. Y. Huang, T. C. Chiles, D. Carnahan, K. Kempa and Z. F. Ren, Nat. Methods, 2005, 2, 449–454.
68. S. Pathak, E. Cao, M. C. Davidson, S. H. Jin and G. A. Silva, J. Neurosci., 2006, 26, 1893–1895.
69. M. J. Saxton, Nat. Methods, 2008, 5, 671–672.
70. A. Triller and D. Choquet, Neuron, 2008, 59, 359–374.
71. M. Dahan, S. Levi, C. Luccardini, P. Rostaing, B. Riveau and A. Triller, Science, 2003, 302, 442–445.
72. C. Bats, L. Groc and D. Choquet, Neuron, 2007, 53, 719–734.
73. S. Lévi, C. Schweizer, H. Bannai, O. Pascual, C. Charrier and A. Triller, Neuron, 2008, 59, 261–273.
74. H. Bannai, S. Levi, C. Schweizer, M. Dahan and A. Triller, Nat. Protoc., 2007, 1, 2628–2634.
75. B. X. Cui, C. B. Wu, L. Chen, A. Ramirez, E. L. Bearer, W. P. Li, W. C. Mobley and S. Chu, Proc. Natl. Acad. Sci. U. S. A., 2007, 104, 13666–13671.
76. S. S. Rajan, H. Y. Liu and T. Q. Vu, ACS Nano, 2008, 2, 1153–1166.
77. M. M. Tamkun, K. M. S. O'Connell and A. S. Rolig, J. Cell Sci., 2007, 120, 2413–2423.
78. Q. Zhang, Y. L. Li and R. W. Tsien, Science, 2009, 323, 1448–1453.
79. M. Howarth, W. Liu, S. Puthenveetil, Y. Zheng, L. F. Marshall, M. M. Schmidt, K. D. Wittrup, M. G. Bawendi and A. Y. Ting, Nat. Methods, 2008, 5, 397–399.
80. A. I. Bachir, D. L. Kolin, K. G. Heinze, B. Hebert and P. W. Wiseman, J. Chem. Phys., 2008, 128, 225105.
81. B. Mahler, P. Spinicelli, S. Buil, X. Quelin, J. P. Hermier and B. Dubertret, Nat. Mater., 2008, 7, 659–664.
82. X. Y. Wang, X. F. Ren, K. Kahen, M. A. Hahn, M. Rajeswaran, S. Maccagnano-Zacher, J. Silcox, G. E. Cragg, A. L. Efros and T. D. Krauss, Nature, 2009, 459, 686–689.
83. R. Nisman, G. Dellaire, Y. Ren, R. Li and D. P. Bazett-Jones, J Histochem Cytochem, 2004, 52, 13–18.
84. M. A. Aronova, Y. C. Kim, N. B. Pivovarova, S. B. Andrews and R. D. Leapman, Ultramicroscopy, 2009, 109, 201–212.
85. B. N. Giepmans, T. J. Deerinck, B. L. Smarr, Y. Z. Jones and M. H. Ellisman, Nat. Methods, 2005, 2, 743–749.
86. L. M. Pavone, S. Tafuri, V. Mastellone, R. D. Morte, P. Lombardi, L. Avallone, V. Maharajan, N. Staiano and G. Scala, Anat Rec (Hoboken), 2007, 290, 1492–1499.
87. K. Aoyama, T. Takagi, A. Hirase and A. Miyazawa, Ultramicroscopy, 2008, 109, 70–80.
88. J. A. W. Heymann, D. Shi, S. Kim, D. Bliss, J. L. S. Milne and S. Subramaniam, J. Struct. Biol., 2009, 166, 1–7.
89. I. S. Uroukov and D. Patton, Microsc. Res. Tech., 2008, 71, 792–801.
90. N. de Jonge, D. B. Peckys, G. J. Kremers and D. W. Piston, Proc. Natl. Acad. Sci. U. S. A., 2009, 106, 2159–2164.
91. A. V. Liopo, M. P. Stewart, J. Hudson, J. M. Tour and T. C. Pappas, J. Nanosci. Nanotechnol., 2006, 6, 1365–1374.
92. (a) A. Mazzatenta, M. Giugliano, S. Campidelli, L. Gambazzi, L. Businaro, H. Markram, M. Prato and L. Ballerini, J. Neurosci., 2007, 27, 6931–6936; (b) G. Cellot et al., Carbon nanotubes might improve neuronal performance by favouring electrical shortcuts, Nat. Nanotechnol., 2009, 4, 126–133.
93. E. Ben-Jacob and Y. Hanein, J. Mater. Chem., 2008, 18, 5181–5186.
94. G. Massobrio, P. Massobrio and S. Martinoia, Nano Lett., 2008, 8, 4433–4440.
95. T. D. B. Nguyen-Vu, H. Chen, A. M. Cassell, R. J. Andrews, M. Meyyappan and J. Li, IEEE Trans. Biomed. Eng., 2007, 54, 1121–1128.
96. Y. X. Zhao, P. Larimer, R. T. Pressler, B. W. Strowbridge and C. Burda, Angew, Chem., Int. Ed., 2009, 48, 2407–2410.
97. T. C. Pappas, W. M. S. Wickramanyake, E. Jan, M. Motamedi, M. Brodwick and N. A. Kotov, Nano Lett., 2007, 7, 513–519.
98. F. Gelain, A. Horii and S. G. Zhang, Macromol. Biosci., 2007, 7, 544–551.
99. H. S. Koh, T. Yong, C. K. Chan and S. Ramakrishna, Biomaterials, 2008, 29, 3574–3582.
100. D. Gupta, J. Venugopal, M. P. Prabhakaran, V. R. Dev, S. Low, A. T. Choon and S. Ramakrishna, Acta Biomater., 2009, 5(7), 2560–2569.
101. S. Saneinejad and M. S. Shoichet, J. Biomed. Mater. Res., 2000, 50, 465–474.
102. Y. Luo and M. S. Shoichet, Nat. Mater., 2004, 3, 249–253.
103. L. M. Y. Yu, N. D. Leipzig and M. S. Shoichet, Mater. Today, 2008, 11, 36–43.
104. F. Johansson, P. Carlberg, N. Danielsen, L. Montelius and M. Kanje, Biomaterials, 2006, 27, 1251–1258.
105. F. Patolsky, B. P. Timko, G. H. Yu, Y. Fang, A. B. Greytak, G. F. Zheng and C. M. Lieber, Science, 2006, 313, 1100–1104.
106. T. Cohen-Karni, B. P. Timko, L. E. Weiss and C. M. Lieber, Proc. Natl. Acad. Sci. U. S. A., 2009, 106, 7309–7313.
107. E. Jan, S. J. Byrne, M. Cuddihy, A. M. Davies, Y. Volkov, Y. K. Gun'ko and N. A. Kotov, ACS Nano, 2008, 2, 928–938.
108. B. J. Marquis, S. A. Love, K. L. Braun and C. L. Haynes, Analyst, 2009, 134, 425–439.
109. M. A. Maurer-Jones, K. C. Bantz, S. A. Love, B. J. Marquis and C. L. Haynes, Nanomedicine, 2009, 4, 219–241.
110. A. M. Derfus, W. C. W. Chan and S. N. Bhatia, Nano Lett., 2004, 4, 11–18.
111. S. J. Clarke, C. A. Hollmann, Z. Zhang, D. Suffern, S. E. Bradforth, N. M. Dimitrijevic, W. G. Minarik and J. L. Nadeau, Nat. Mater., 2006, 5(5), 409–417.
112. N. A. Tsvyetlynska, R. H. Hill and S. Grillner, J. Neurophysiol., 2005, 94, 3951–3960.
113. M. L. Tang, T. R. Xing, J. Zeng, H. L. Wang, C. C. Li, S. T. Yin, D. Yan, H. M. Deng, J. Liu, M. Wang, J. T. Chen and D. Y. Ruan, Environ. Health Perspect., 2008, 116, A418–922.
114. M. L. Tang, M. Wang, T. R. Xing, J. Zeng, H. L. Wang and D. Y. Ruan, Biomaterials, 2008, 29, 4383–4391.
115. J. X. Zhao, L. J. Xu, T. Zhang, G. G. Ren and Z. Yang, NeuroToxicology, 2009, 30, 220–230.
116. L. L. Dugan, D. M. Turetsky, C. Du, D. Lobner, M. Wheeler, C. R. Almli, C. K. Shen, T. Y. Luh, D. W. Choi and T. S. Lin, Proc. Natl. Acad. Sci. U. S. A., 1997, 94, 9434–9439.
117. H. Jin, W. Q. Chen, X. W. Tang, L. Y. Chiang, C. Y. Yang, J. V. Schloss and J. Y. Wu, J. Neurosci. Res., 2000, 62, 600–607.
118. D. Schubert, R. Dargusch, J. Raitano and S. W. Chan, Biochem. Biophys. Res. Commun., 2006, 342, 86–91.
119. J. A. Kloepfer, N. Cohen and J. L. Nadeau, J. Phys. Chem. B, 2004, 108, 17042–17049.
120. I. Yildiz, M. Tomasulo and F. M. Raymo, Proc. Natl. Acad. Sci. U. S. A., 2006, 103, 11457–11460.
121. A. Verma, O. Uzun, Y. Hu, H. S. Han, N. Watson, S. Chen, D. J. Irvine and F. Stellacci, Nat. Mater., 2008, 7, 588–595.
122. D. W. Dorward, Methods Mol. Biol., 2008, 431, 173–187.
123. W. H. Suh, K. S. Suslick, G. D. Stucky and Y. H. Suh, Prog. Neurobiol., 2009, 87, 133–170.
124. M. Gadnell and K. Gunnarsson, Nat Methods, 2005, 2 advertising feature.
125. T. Gonen, Y. Cheng, P. Sliz, Y. Hiroaki, Y. Fujiyoshi, S. C. Harrison and T. Walz, Nature, 2005, 438, 633–638.

---