In vivo imaging is, of course, nothing new in biology: it dates back to the very beginnings of microscopy. Examples of this early work are the studies by van Leeuwenhoek1 and the observation of Brownian motion2. However, imaging with high resolution in the living animal while extracting quantitative information has only come to fruition relatively recently, with the development of new imaging methods such as magnetic resonance (MRT; for example, see Ref. 3), positron emission (PET; for a review see Ref. 4), X-ray (for example, see Ref. 5) and optical coherence6 tomographies, and of nonlinear optical imaging7. Although each of these techniques offers insights into neural structure and, in some cases, function, only multi-photon microscopy7 (MPM) is currently able to resolve the activity and structure of single neurons. This Review will, therefore, focus almost entirely on MPM and its applications to in vivo imaging of neural activity and morphology.
The properties of MPM8, 9, 10, 11, which is almost always practiced as two-photon microscopy, make it particularly well suited to deep-tissue imaging. Its application to a whole range of neurobiological challenges, from imaging the activity of neuronal populations to following changes in neurite morphology, is, therefore, not surprising. This has been facilitated by the rise of commercially available systems that make MPM more accessible to biologists. There has been an abundance of reviews describing both the technique and its application to the in vivo situation8, 10, 12, 13, 14, some of them published only recently11, 15, 16, 17. With this in mind, we focus here on how MPM is being used to address questions about neural activity (electrical, chemical and morphological) in vivo on timescales ranging from milliseconds18, 19, 20 to months17, 21, 22, 23, 24, 25, 26, 27, 28, 29.
MPM is being intensively applied to questions regarding ongoing30, 31 and evoked32, 33, 34, 35 neuronal-population activity. Compared with multi-electrode recording36, 37 (for a review see Ref. 38), optical population recording has the advantage that all of the cells in a field of view can be probed, regardless of whether or not they are firing action potentials (APs)31. Furthermore, because their spatial locations are precisely known, the cell types of all recorded cells can be determined — if not necessarily in vivo, then almost certainly post mortem using immunohistochemical techniques. Imaging is also potentially much less invasive than electrode recording, because light penetrates brain tissue without mechanical disturbance. In addition, because optically recorded neurons are readily identified, MPM can record from the same neurons or their substructures repeatedly over many days and weeks — something that is impossible with electrodes.
In this Review we briefly describe the physical principles of the optical methods behind current in vivo imaging, with a focus on MPM. We then discuss key applications of MPM: the imaging of neuronal activity in the single neuron, in subcellular compartments and in neuronal populations; and, finally, the observation and quantification of the morphological stability and plasticity of subcellular neuronal processes.
Principles of in vivo optical-imaging methods
The use of light in and near the visible wavelength range (approximately 300–1,100 nm) is the oldest and still most widely applied in vivo imaging technique. The resolution that can be achieved with this method is ultimately determined by the wavelength (but see Ref. 39) of the light that is used, but it can be severely degraded by scattering and optical aberrations. For contrast, light microscopy depends on variations in refractive index, molecular absorption or fluorescence properties in the object that is being examined. For in vivo imaging, in most cases the detector and the illuminator are located on the same side of the sample (the episcopic configuration), and they often share an objective lens. This precludes the use of most refractive-index-based (phase-contrast) techniques and many absorption-based techniques, which require the detection of transmitted light. The episcopic configuration does allow the use of reflection contrast, in which the signal intensity is determined by a combination of tissue scattering and absorption properties40, 41, 42. Another technique, optical coherence tomography43, in which light is selected by interference according to its travel time, allows the depth-resolved imaging of back-scattering properties with high resolution deep inside tissue. Among scattering techniques one should also include harmonic generation44. In this technique, owing to the fact that the movement of electrons in a molecule becomes nonlinear at very high instantaneous light intensities, light is generated that has a higher frequency (and hence a smaller wavelength) than the incoming light45. Like fluorescence, but unlike ordinary scattering, harmonic generation can be chromophore-specific.
Fluorescence imaging is probably the contrast method that is most widely used for the generation of actual optical-microscopy data. It is almost always performed in the episcopic configuration and is arguably the most versatile contrast technique. This is mostly because of its unrivalled chemical specificity, which results from its multi-stage selectivity: the excitation spectrum, the fluorescence quantum efficiency and the emission spectrum are all specific for the particular fluorophore that is used. Fluorescence (and also backscattering) contrast can be combined with confocal microscopy46. Confocal microscopy can provide optical sectioning (in which information is selectively obtained from a thin slice of the sample) and high resolution, even in tissue that causes a high degree of scattering. However, this comes at the expense of signal size, as most of the fluorescence that comes from the focus is scattered and thus cannot pass through the detector pinhole47. The need to increase the excitation power to compensate for this loss then leads to increased photodamage at and around the focus site. This dilemma can be avoided by the use of multi-photon excitation, which relies on the simultaneous absorption of multiple photons and thus depends steeply on the photon concentration (that is, the light intensity)7, 10, 48. As a result, the volume of tissue that is excited is sufficiently confined to obviate the need for spatially resolved detection. This allows the inclusion of scattered fluorescence light8 and, owing to a lack of excitation, there is no out-of-focus photodamage. Although out-of-focus excitation can also be eliminated by using selective plane illumination49, 50, 51, this technique, unlike MPM, requires transparent specimens and lateral access, neither of which is available in the in vivo brain.
In fluorescence microscopy, detector or illumination noise is rarely important; usually the signal-to-noise ratio is limited by photon-shot noise caused by the random arrival of photons at the detector. In particular, for time series (the acquisition of multiple images from the same area, often in rapid succession) it is necessary to obtain as many images as possible using an excitation dose that does not damage the specimen. A high fluorescence-detection efficiency is, therefore, even more essential for in vivo opto-physiology than for morphological imaging.
Measuring in vivo activity
Although most of our understanding of single-cell physiology has been obtained by studying neurons in varying degrees of isolation, it has long been realized that these findings will have to be placed in the context of natural physiological activity. This can only be achieved by measuring the activity of single neurons in vivo (both in the awake and in the anaesthetized state) in order to establish how neurons behave while embedded in intact, active networks. This need has driven the development of in vivo electrophysiology techniques that allow the recording of sub-threshold membrane-voltage fluctuations in neurons using high-resistance52, 53, 54 and tight-seal55, 56, 57 electrodes. These developments have included the construction of devices that allow active stabilization of the electrode position58. Particularly impressive are recent studies that have mastered whole-cell recording in awake, head-fixed59 and in freely behaving60 animals. Under these conditions, modulations of cellular response properties occur that depend on ongoing network activity and are absent in vitro, such as network-driven concerted membrane-potential fluctuations61, 62 (up states and down states).
To study network dynamics proper, rather than just the effects of networks on single-cell behaviour, techniques that allow recordings to be made from multiple cells simultaneously are needed. Suitable techniques, such as multiple whole-cell recordings63, 64, 65 or wide-field population imaging66 with cellular resolution, are available for acute brain slices but are not readily applicable in vivo. The imaging of changes in the intrinsic reflectivity of the brain surface42 and of voltage-sensitive-dye (VSD) fluorescence67, 68 has, for example, shown that ongoing activity in vivo can contain spatiotemporal patterns that resemble those that are evoked by external stimuli69, 70 (but see Ref. 71). Imaging results are generally consistent with those from electrophysiological experiments72. Wide-field imaging of the cortical surface offers millisecond-scale temporal resolution (when using VSDs73, but not when using the intrinsic, blood-flow-related signal) and can easily be combined with field-potential74, 75 or whole-cell recordings72. However, like functional MRI (fMRI), in vivo wide-field imaging suffers from low resolution76. In this respect, recent results (B. Kuhn, W.D. and R. Bruno, unpublished observations) that were obtained using MPM and a new class of more sensitive77, 78 VSDs in vivo are encouraging. In the following sections we describe some applications of in vivo MPM imaging that have begun to shed light on aspects of neuronal function.
Imaging dendritic integration. Following the demonstration that MPM can optically reach depths of several hundred micrometres into the cortex in vivo8, MPM was used to map activity in the dendrites of single neurons, which were filled with a Ca2+ indicator through a recording electrode20. This allowed dendritic signalling to be studied in vivo. One question that has been studied using MPM both in the cortex18, 19, 79, 80 and in the olfactory bulb81 is whether the propagation of somatic APs back into the dendritic tree (backpropagation), which appears to be a common requirement for synaptic memory mechanisms and has been studied extensively in brain slices (for a review see Ref. 82), also occurs in vivo, under conditions of ongoing network activity and the ensuing persistent barrage of synaptic input.
With MPM it was found that in anaesthetized animals, regardless of whether the layer 2/3 cortical pyramidal cells under study were filled and recorded through high-resistance microelectrodes19 or in tight-seal mode80, single-AP-evoked transient increases in Ca2+ concentration ([Ca2+]) were strong in proximal dendrites but absent in distal dendrites, unless the AP was paired with synaptic stimulation80 (Fig. 1) or was coincident with ongoing synaptic input79. Supralinearity of distant Ca2+ transients (in which transients for the combined stimulation are larger than the sum of transients for either stimulation alone) was more pronounced with tight-seal recordings80.
Figure 1 | Measuring Ca2+ transients from dendrites and neuronal populations in vivo.
a | On the left is a side projection of a two-photon image stack showing a layer 2/3 (L2/3) pyramidal neuron filled with a Ca2+ indicator (Oregon Green BAPTA-1; 200 
M) in vivo (the soma was 400 m from the pia). To the right of this image are Ca2+ transients (averaged over 3–5 trials) that were evoked by single action potentials (APs) at the positions indicated along the dendrite. b | A schematic representation of neuronal Ca2+ transients that were evoked by single APs in vitro (filled triangles) and in vivo (white triangles). The neurons were ranked according to the depth of their somata below the pial surface. Each neuron is represented as a column of four points (connected by a line). These points represent the depths of the soma (triangle), the bifurcation (bifurc; diamond), the position at which the Ca2+ transient was largest (red square) and the position above which it was undetectable (blue circle), as schematically illustrated on the right. c | Averaged, normalized and smoothed Ca2+ transients recorded in vivo from four neurons in response to one AP, one EPSP and one AP paired with one EPSP. Note that the amplitude of the Ca2+ transients is more than the sum of each stimulation alone when AP backpropagation coincides with EPSP arrival at the dendritic arbour. Together, these plots illustrate that backpropagating APs fail to reach distal dendrites unless they are boosted by coincident synaptic activity. d,e | Spiking activity can be simultaneously recorded from all neurons in an area. d | An image taken at a depth of 300 m below the pial surface 2 hours after the application of a Ca2+ indicator that labels both neurons and astrocytes and an astrocyte-specific label. Note the labelling of layer 2/3 neuronal somata (green) and astrocytes (yellow). e | Normalized fluorescence transients recorded simultaneously from multiple neuronal somata. Transients that were presumably evoked by APs or bursts of APs are labelled with asterisks (see Box 1). Parts a–c reproduced, with permission, from Ref. 80 © (2003) Society for Neuroscience. Part e modified, with permission, from Ref. 35 © (2007) Society for Neuroscience.
Discrepancies between dendritic electrical recordings that were carried out using high-resistance electrodes and those that were carried out in the tight-seal mode19, 79, 80 highlight how useful it would be to have an optical membrane-voltage probe with a better signal-to-noise ratio than those that are currently available. The use of optical recording would avoid the perils of physical membrane damage by high-resistance microelectrodes and washout of soluble endogenous molecules in the whole-cell mode, which can change dendritic behaviour substantially (as has been shown, for example, in retinal interneurons83). Whether backpropagation in vivo always resembles the in vitro behaviour, or whether it depends, for example, on attention or arousal79, will ultimately need to be resolved by imaging in awake, behaving animals using either a head-fixed configuration84 or a head-mounted MP microscope85, 86, 87, 88.
Whenever single cells or specific populations can be labelled with indicators, optical fibres in combination with one-photon techniques89 can be used to achieve some measure of discrimination between dendritic compartments90. In some non-vertebrate species, imaging dendrites in vivo with both one- and multi-photon excitation has provided important information that helps us to understand computational processes91, 92, 93.
Imaging population activity. Loading an individual cell with dye does not allow the imaging of activity in multiple cells, let alone in whole populations. What made the imaging of population activity possible in the mammalian brain was not, as in non-mammalian species94, 95, 96, 97, the application of fluorescent Ca2+-indicator proteins (FCIPs)98. Instead, it was the adaptation30 of the ester-trapping technique99 (Fig. 1d), which is based on creating a membrane-permeable form of the indicator that is converted into a charged and hence impermeable form by endogenous enzymes — a technique that has been successfully used for many years in cell culture and brain slices66, 100. A number of laboratories seized this opportunity and began using MPM to image the activity of neuron populations in areas in which other bulk-loading techniques had previously been used to map input patterns with wide-field microscopy101, 102. These included the somatosensory cortex in rats and mice35, 103, the cerebellum in rats104, the visual cortex in rats, cats32, 33 and mice34, the tectum in zebrafish105 and the olfactory bulb in mice106 and zebrafish107, 108.
In the cat visual cortex, the arrangement of orientation-selective regions in pinwheel patterns was discovered using wide-field in vivo optical imaging109. With MPM it became possible to resolve the tuning properties of individual cells, and this was used to show that there are sharp boundaries between regions of neurons with different orientation preferences33. Again, the imaging results were consistent with electrophysiological findings. Electrophysiology cannot provide the spatial resolution and cellular morphological identification that MPM can, but it does permit single-AP detection which, until recently31, was unreliable with imaging (Fig. 2). This issue became pressing when in vivo whole-cell110 and juxtasomal35, 111 electrophysiological recordings showed much lower average firing rates and response probabilities than had previously been assumed to exist. In this context it is not only important to understand the detection reliability for APs (Box 1), it is also important to be able to distinguish neuronal somata from glial somata, which fortunately take up some dyes preferentially112. Population imaging allows the simultaneous mapping of the receptive fields of many identified cells and, even more importantly, the detection of correlations in neuronal activity35.
Figure 2 | Spiking activity in populations of neurons can be inferred from Ca2+ transients, with single-action-potential resolution.
a | A glass electrode (red; made visible by including 5 M Alexa 594 in the filling solution) among a population of neurons (green) and astrocytes (yellow) in the somatosensory cortex. Note the blood vessels surrounded by the astrocytes. b | Cell-attached voltage recording (upper trace) and simultaneously recorded somatic Ca2+ transients (lower trace) from an experiment similar to the one shown in part a. c | Examples of Ca2+ transients evoked by single (left-hand plot), double (middle plot) and triple (right-hand plot) action potentials (APs) in separate experiments. d,e | Spike rates can be estimated from the fluorescence signal (F) by deconvolution. Parts d,e show how this was applied to neurons with high firing rates (5–10 Hz) in zebrafish mitral cells and interneurons155. d | Fluorescence from expressed HuC:cameleon (labelling mitral cells) (left-hand image) and the ester-trapped Ca2+ indicator Rhod-2 (right-hand image) for a field of neurons. The arrows point to eight individual neuronal somata. e | From top to bottom: current that was injected into a single neuron, the resultant APs, the resultant Ca2+ transient (red trace), a firing-rate estimate (light-blue trace) and the actual firing rate (dark-blue trace). Parts a–c reproduced, with permission, from Ref. 31 © (2005) National Academy of Sciences. Parts d and e modified, with permission, from Ref. 155 © (2006) Macmillan Publishers Ltd.
Ester-based loading techniques lead to dense staining (as far as 400 m from the injection site) of both somata and neurites (Fig. 1), and therefore allow the recording of neuropil activity31 (Box 2) but not the attribution of this activity to individual processes. Such attribution is possible when bulk loading is carried out in one region (such as deep in the cortex) and imaging takes place in another region (such as the upper cortex, where the processes of the loaded cells have become sparse and can be individually resolved)31 (see also Ref. 113). It is not yet clear whether it will be possible to detect sub-threshold Ca2+ activity in neurites in vivo, as is possible in vitro114, 115. Neuropil activity can also contaminate the 'somatic' signal when there is insufficient spatial resolution (Box 2).
Because of their ability to better target specific cell types, we expect to see increased use of FCIPs in mammals116, 117, 118, 119. It is possible that these FCIPs will be delivered mainly by viruses, which allow increased expression levels and can deliver the proteins to a wider range of species than transgenic techniques (although they also have disadvantages, such as the potential for damage caused by the injection and by the protein being overexpressed)120. Another advantage of FCIPs is that, at least in theory, they allow functional recordings to be carried out in the same cells at multiple points in time — days, weeks and even months apart (see below).
Targeting electrodes. In early imaging studies, dye-filled sharp microelectrodes were driven 'blindly' into the cortex until they successfully penetrated a soma or a dendritic process. Although it is possible to classify a neuron that has been filled with dye in this manner post hoc by imaging its morphology with MPM20, a bias towards certain cell types (for example, those with large somata) is likely; indeed, some cell types are likely to be missed altogether. More recently, MPM has been used to guide tight-seal recording pipettes towards a fluorescently labelled cell in a process called two-photon targeted patching (TPTP)121. Because of the flexibility of sharp electrodes, such targeting can be achieved only in shallow tissue such as the whole-mount retina122. In acute slice preparations, optical targeting of neurons with recording electrodes, in this case by infrared differential interference contrast (IR-DIC)123, allowed patch recordings in brain slices to be made with much greater precision. In vivo, TPTP allows recordings to be made from rare fluorescently labelled cells. This has been crucial for establishing spike-detection reliability (Box 1), and has enabled in vivo single-cell genetics studies to be carried out. In these studies, a sparse subset of neurons in an otherwise wild-type brain was infected with a virus124 that not only changed the physiological properties of the neurons (specifically, their dendritic excitability using RNAi knock-down of a dendrite-specific Na+-channel subunit) but also labelled them with green fluorescent protein (GFP) and thus permitted testing for changed electrical-response properties by TPTP125. The gap between what is possible in vivo and in vitro has thus narrowed, and single-cell behaviour and network dynamics in vivo can now be studied. Where else, aside from in activity measurements, is the in vivo context important, and maybe even essential? The issue of whether there is fine-grained structural stability in neurons immediately comes to mind.
Long-term in vivo time-lapse imaging. Tissue plasticity and stability are not only central to development (for a review see Ref. 126), they might also be important for the acquisition and maintenance of memory in the adult state. Morphological changes during development are substantial and stereotyped and can, therefore, often be detected by post-mortem analysis of specimens of different ages (for a review see Ref. 127). However, such studies are labour intensive and, more importantly, are fundamentally unable to detect dynamics that leave statistical properties, such as the densities of synapses or the volume fractions that are occupied by axons and dendrites, unaltered. This became clear during the pioneering studies of synapse elimination in the neuromuscular junction, in which the behaviour of individual terminals was followed in time128, 129.
In the mammalian cortex it was unknown whether synaptic contacts, the density of which is constant in the adult22, change with time and, if so, at what rate. Without a label that is specific for newly formed synapses being available, observation of the same tissue region in the same animal over a period of weeks or months (time-lapse imaging) is needed. Furthermore, unlike a novel-synapse label, such time-lapse imaging also allows the detection and quantification of highly fluctuating intermediate states27, 126, 130, 131. In developing embryos, which are often optically transparent, and in the peripheral nervous system (mostly the neuromuscular junction), time-lapse imaging is possible with wide-field or confocal microscopy128, 132, 133.
Only in recent years have such experiments become possible in the adult mammalian brain, through the use of MPM combined with cellular labelling and trans-genetically introduced protein-based fluorophores134. Retrospective electron microscopy has been used to demonstrate that spines that had been shown by in vivo time-lapse imaging to be newly formed contain all of the structural elements that are typical for synapses24. All studies have found that there is a high degree of structural stability135. Nevertheless, the turnover rates for dendritic spines that have been reported by different laboratories vary by a factor of at least five (see Ref. 23 and the references cited therein; Fig. 3). The rates appear to depend on cell type, are affected by the presence or absence of sensory stimulation26, 136, and also seem to depend on the method that is used to gain optical access to the cortex23. In particular, the creation of an open-skull window appears to lead to the activation of microglia, which respond rapidly to brain injury137 and might contribute to higher spine turnover rates. Although axonal-bouton turnover matches spine turnover rates both in rodents138 and primates29, these experiments have been carried out with opened skulls and, therefore, some of the issues that have been encountered in the experiments that measured spine turnover might also apply here.
Figure 3 | The persistence of dendritic spines in adult mice, revealed by in vivo imaging.
a–d | Dendritic branches from two animals. For each animal, two images were obtained 18 months apart (animal 1: a,b; animal 2: c,d). Note that most of the spines in a and c persist in b and d. e | The percentage of spines that were eliminated, formed or persistent over 18–22 months. f–i | The effects of sensory experience on spine persistence. f | Trimming of alternate whiskers generates different sensory experiences for neighbouring whisker barrels. g | A high-magnification view of two dendritic segments imaged before (days 0–8) and after (days 12–28) whisker trimming. The arrowheads show spines that were always present (AP; yellow); new persistent spines (NP, orange); lost persistent spines (red); and transient spines (blue). h | Spine density does not change significantly over time in control (blue) or deprived (red) locations. i | Quantification of NP spines, showing that total numbers of NP spines were a small fraction of the total number of persistent spines and that trimming enhanced this fraction. Every circle represents a cell, and the horizontal bars represent the group averages. Grey lines connect paired experiments. Parts a–e reproduced, with permission, from Ref. 135 © (2005) Elsevier Science. Parts g–i reproduced, with permission, from Ref. 136 © (2006) Macmillan Publishers Ltd.
Integration of newborn neurons into existing circuits has also been observed. First a lentivirus containing the gene for GFP was injected into the subventricular zone, where olfactory receptor neurons are born, and then MPM was used to observe these neurons as they migrated, matured and integrated into the circuit139. It might be interesting to use TPTP to test, more directly, the evolution of the physiological properties of these neurons as well as their synaptic integration into the neuronal circuit.
Technical issues
The key advantage of MPM is its ability to image deep in light-scattering tissue8, 9, 10, but the reachable depth is still limited to approximately 1 mm. This is not because of limits in the available laser power, because it is possible to use amplified pulses at a lower repetition rate to boost nonlinear effects140, but because of substantial two-photon excitation near the sample surface. Even though this surface excitation is spread over a large area, the laser power is undiminished by scattering. The extent of the excitation depends on the numerical aperture of the objective lens and, even more strongly, on the labelling distribution, with more uniform distributions leading to stronger relative background excitation141. Imaging depth, resolution and signal size are also affected by distortions of the laser-beam wavefront that are caused by refractive-index heterogeneities in the sample. Such distortions enlarge the focus beyond its diffraction-limited extent but can be pre-empted, if they can be measured142, by adaptively pre-distorting the wavefront that goes into the sample143.
Another area in which technological advances have a key role is in imaging awake, behaving animals. Although miniaturized amplifier head stages are now routinely used for the recording of APs from multiple cells (for a review see Ref. 144), in awake, freely moving animals similar efforts in optical microscopy85, 145 are still at an experimental stage. Miniaturized objectives and scanners86, 88have been developed and used in conjunction with fibre-optic excitation-light delivery and fluorescence-light collection to record images from moving animals. Single-position fluorescence signals from moving animals can be recorded using one-photon excitation and detection through a multi-mode optical fibre89. An alternative, and very promising, approach is to use a conventional MPM but provide virtual-reality motion for awake, head-fixed animals84.
What does the future hold for in vivo MP imaging?
The issue of dendritic information integration and how it depends on factors such as wakefulness, attention and expectation is likely to become prominent in single-cell in vivo imaging. The requisite tools, such fibre-coupled microscopes85, 88, 145, 146 or head-fixed virtual-reality systems84, have already been demonstrated. On the axonal side, the modulation (again, by factors such as attention) of residual Ca2+, which is known from in vitro studies to influence short-term synaptic plasticity147, might be an interesting target for investigation in vivo. Arguably more important for the understanding of neural computation will, in our view, be the ability to measure population activity with cellular, single-AP and potentially millisecond resolution. Here the fact that MPM is a fundamentally serial technique (parallelization of MPM is possible but, because it requires spatially resolved detection, it comes at the expense of the ability to work deep in strongly scattering tissue) presents a technological challenge if many points in space are to be recorded. This is because the focus has to be moved sufficiently quickly for it to visit all cells or cellular compartments at least once during each unit of temporal resolution. Possible methods for fast in-plane movement are resonant-mirror scanning148 and acousto–optical deflection149. The latter can also be used to shift the focus to some extent along the optical axis150, which otherwise requires the microscope objective to be moved151. The ability to scan in three dimensions will be essential to cover even localized populations completely.
There are many questions, such as what fractions of which cell types are active during a certain behaviour or sensory input, that require not only simultaneous recording from multiple cells, but also the cells' spatial localization and cell-type classification. Increasingly refined genetic labelling techniques will be important for indicating cell type definitively, and combinatorial uses of differently coloured fluorescent proteins will potentially be required152. It will be interesting to see whether it will be possible to overcome the current inaccessibility of deeper brain structures while keeping the damage to the penetrated overlying brain structures to an acceptable level by using penetrating imaging devices146, 153, 154.
Spiking appears to be represented in a rather linear fashion by increases in [Ca2+]31, 155, but [Ca2+] is a nonlinear measure of sub-threshold activity. A method by which to measure the membrane voltage directly would, therefore, be desirable. Even more important might be the measurement of other local biochemical signals and the ability to determine how they are dependent on local activity and its history. Measurements of glial activity have revealed that astrocytes have a role in regulating arterial blood flow156 and show a rapid increase in [Ca2+] upon sensory stimulation157.
Imaging is likely to be helpful in understanding how developmental and adult-morphological dynamics are related to each other and, possibly, to memory formation, memory consolidation and mental disease158. Models of neurodegenerative disease should benefit greatly from the ability to follow the development of characteristic cellular pathologies, such as tangles and plaques159, 160, and how these pathologies respond to various forms of therapeutic intervention. Even the development of diagnostic tools based on endoscopic MPM is a distinct possibility. Coarser volume-imaging methods, such MRI, could first provide an individualized anatomical reference161 and then, with the help of functional signals (fMRI and PET), guide the placement of the optical recordings.