|Home | About | Journals | Submit | Contact Us | Français|
Brain function emerges from the morphologies, spatial organization, and patterns of connectivity established between diverse sets of neurons. Historically, the notion that neuronal structure predicts function stemmed from classic histological staining and neuronal tracing methods. Recent advances in molecular genetic and imaging technologies have begun to reveal previously unattainable details about patterns of functional circuit connectivity and the subcellular organization of synapses in the living brain. A sophisticated molecular and genetic “tool box” coupled with new methods in optical and electron microscopy provide an expanding array of techniques for probing neural anatomy and function.
Individual neurons represent the elemental building blocks of the brain. However, understanding how neurons “wire” through synapses, the microcircuits they form, and how those microcircuits participate in neural networks to drive perception and behavior remains one of biology’s major frontiers. To fully unravel complex neural computations and representations, we must consider both the structure of connections formed between intermingled cell types, and how electrical activity propagates through the network to enable its dynamic function. Our current understanding of how neuronal circuits produce sensory perception, memory, and behavior remains nascent. In most cases, it is far from clear how brain functions emerge from existing circuit diagrams. Bridging this conceptual chasm requires knowledge across several scales ranging from cell-type specific genetic identity, population-level network connectivity, the function and dynamics of individual synapses, and circuit output.
Neuroanatomy in the 21st century is being rekindled by the merger of molecular and genetic manipulations aimed at probing circuit function with gene expression profiling, high-resolution imaging, and electrophysiology. In addition, endeavors are underway to build high-resolution maps of all the neurons and their full complement of synapses in multiple model organisms1. These new techniques for monitoring and manipulating circuits, neurons, and synapses are providing expanded capability that heralds a new era of functional neuroanatomy. Here we briefly trace the evolution of the technological arsenal currently available for probing neuron and circuit architecture. In addition, we provide critical discussion regarding current experimental limitations and propose future convergent methodologies to advance a deeper understanding of neuroanatomy.
With the advent of biological electron microscopy (EM) and initial work of Palade and others, new subcellular details of neuronal organelles and structures, including the organization of synaptic contacts, came into view2. Early EM circuit maps ranged from single cells to intact invertebrate nervous systems3,4, as well as retinal assemblages and cortical columns in mammals5,6,7. Recently, the power and utility of electron microscopy has expanded with advances in volume EM techniques8, which can reveal the fine detail of neuronal morphology and subcellular structure, and provide near-molecular resolution in a native three dimensional (3D) environment (Box 1).
Current methods for volume EM rely on three-dimensional (3D) image reconstruction from ultrathin serial sections made through intact nervous tissue. Individual section planes (A), are reconstructed in the z-axis to build a volumized block of image planes that can be digitally navigated in all axes using image analysis software (B). 3D image reconstruction further allows the volumetric isolation of complete neuronal architecture, ranging from subcellular organelles to cell morphology (C)10,77. Serial section transmission electron microscopy (SSTEM) was the first volumetric EM technique to allow the assembly of high-resolution circuit data, and has remained the primary approach to resolve 3D neuronal ultrastructure3,4,10,77. This approach relies on the manual collection of ultrathin tissue sections onto grids that are imaged and reconstructed in series (D). SSTEM has been the gold standard, but is labor intensive and has limited ability to routinely reconstruct high quality data sets of sections less than ~50 nm in thickness.
Instead of cutting ultrathin sections that require manual collection and subsequent imaging, serial block-face scanning electron microscopy (SBFSEM) relies on generating images from electrons that are scattered off the face of an embedded tissue block that undergoes repeated rounds of surface sectioning to reveal deeper structures78 (E). Because the images are captured from the face of the block prior to each cut there is no need to collect tissue, thus reducing section thickness variability while automating the entire process. Ion beam milling of the block face for serial imaging provides an elegant alternative to mechanical cutting11. Although SBFSEM holds promise for streamlining volume EM, technical limitations have prevented its widespread use to date; sample quality and image resolution is still inferior to SSTEM8.
A permutation of SSTEM that provides higher z-section resolution than microtome sectioning is that of serial section electron tomography (SSET) (F)79. In this method, subcellular structures are reconstructed by assembling multiple images captured from an angular 2D projection tilt series to generate a 3D tomogram. The primary benefit of this technique is the reduced number of sections required to build a high-resolution image volume compared to either SSTEM or SBFSEM.
A major challenge to all of these techniques is the need for accurate image alignment, contour recognition, automated segmentation and error checking in order to assemble detailed volume sets. All current accurate methods require this to be done (or at least curated) manually8,10,80,81. It remains to be determined to what extent computational methods will replace brute force manual tracing. Kristen M. Harris kindly provided the raw EM data series used to generate images used in panels A–C. The volumized data set was prepared for graphic representation using Amira software (Visage Imaging, Berlin).
Complementing EM volume reconstruction, ordered array tomography coupled with fluorescence imaging allows submicron investigation of individual cells or subcellular structures labeled by fluorescent probes9. In this technique, nanometer scale z resolution is achieved by collecting ultrathin sections (50–100 nm), followed by molecular localization using antibodies or fluorescent proteins. Unlike the labor-intensive protocols required for volume EM data sets, array tomography provides a relatively simple platform that combines light microscopy and EM to reveal both molecular and ultrastructural details of intact nervous tissue.
New methods of serial sectioning, alignment, and computational analysis are continually increasing the speed and utility of EM reconstruction for investigating the ultrastructural features of nervous tissue10. However, a major limitation for using both electron- and light microscopy for mapping brain connectivity is the current lack of high-throughput approaches required to move beyond small-scale experimentation. With further development of technology to more rapidly collect, curate, and analyze high-resolution imaging data8,10,11, ultrastructural analysis will become more tractable. However, this type of data does not provide dynamic information in real-time or within living tissue. For this, one must turn to optical imaging and genetic strategies.
The expanding field of molecular genetics now offers numerous avenues for probing neuroanatomy. Advances in conditional knockout mice, inducible gene expression systems, and genetic control of neuronal activity provide means to probe circuit biology with unprecedented precision12. The seminal discovery of genes encoding fluorescent proteins (FPs) in marine invertebrates has allowed the expression of FPs and FP-tagged biomolecules in living neurons of the mammalian brain13,14, providing an expanding array of sophisticated imaging and genetic profiling techniques15,16,17,18.
Neuronal populations have genetically specified identities. Classical definitions of neuronal cell types based on morphology or electrical properties can now be linked to, and perhaps ultimately replaced by, discrete patterns of gene expression. With the expansion of molecular genetics has come the ability to query the genetic makeup of individual cells or entire brain regions. Do neurons of a given type make stereotyped patterns of connections in the mammalian brain? Do such connections result in predictable patterns of circuit function or behavioral output? Genetically marking and manipulating cells in the brain now allows such questions to be addressed. Major advances in manipulating genetic programs in mice permit the generation of loss-of-function, gain-of-function, and conditional mutations at will19. Conditional genetic strategies are already proving crucial for understanding neuronal circuits. Implementation of phage-derived recombination systems such as Cre/loxP technology provide intersectional approaches for genetic manipulations in neurons ranging from single cell alterations20,21 to control of gene activity in select brain structures22,23. However, harnessing cell type-specific promoter activity to precisely manipulate neuronal subsets in the mammalian brain remains a challenge. Often genes that are transcribed in target neuron populations show overlapping patterns of expression in other brain areas, or exhibit dynamic regulation during development that can confound their use in the mature brain.
Alongside the ability to investigate single cells by targeted expression of molecular markers, large-scale efforts to map patterns of endogenous gene expression in the rodent brain have provided a wealth of data to the neuroscience community (e.g. GENSAT, www.gensat.org and the Allen Brain Atlas, www.brain-map.org)24. These open resources now make it unnecessary for individual labs to conduct exhaustive gene expression experiments, since this data can be referenced via the Allen Brain Atlas for nearly any neuron type or brain structure. However, most expression data is still limited to healthy adult animals. More complete data sets that are currently being assembled by the Allen Brain Institute include gene expression in the developing mouse brain, the mouse spinal cord, and human cortex (www.brain-map.org). A future can be envisioned when the spatial and temporal pattern of all genes is available across the life span, in diverse species, and in the context of specific mutations or disease states.
A basic principle of neuroscience is that information is stored and encoded by the properties of neurons and the connections they make with one another. Thus, identifying patterns of neuronal connectivity is requisite for understanding brain function. In order to piece together these patterns, information describing both circuit anatomy and the flow of neuronal activity are essential. Several elegant technologies have recently emerged that allow for both trans-synaptic circuit analysis and precise control of neuronal firing12, including the use of retrogradely transported viral vectors25,26,27, optogenetic tools such as light-activated ion channels28,29,30, and heterologous receptor activation28,31,32.
Advances in viral vector engineering have exploited neurotropic viral particles as tools to study synaptic connectivity33 (Figure 1). For example, engineered pseudorabies vectors can be modified to express FPs and thus label interconnected neurons by retrograde trans-synaptic transfer of viral particles to presynaptic cells25. One limitation to using pseudorabies virus is its polysynaptic spread. Due to the high degree of neuronal interconnectivity in intact circuits, polysynaptic spread makes it difficult to unambiguously assign synaptically coupled partners. Circumventing this problem, Wickersham et al. devised a novel coat protein complementation strategy that allows monosynaptic tracing of neural connections by using rabies virus particles engineered to express GFP27 (Figure 1).
Beyond enabling more detailed analysis of anatomy and synaptic connectivity, genetic methods are now being harnessed to facilitate selective control of activity among populations of interconnected neurons in the mammalian brain. Alongside the new subdiscipline in neuroscience aptly coined optogenetics (see accompanying review by K. Deisseroth), efforts to develop novel heterologous receptor expression systems now provide alternative and potentially non-invasive means to modulate neuronal activity using chemical-genetic approaches. Heterologous expression of modified opiate receptors provided an initial demonstration that neuronal subsets in the brains of mice could be genetically targeted for activation by synthetic, exogenous ligands34. Other studies have used overexpression of endogenous receptors to amplify neuronal subtype-specific neurotransmission. For example, mice engineered to harbor high affinity acetylcholine receptors (nAChRs) in dopaminergic neurons show hyperdopaminergic behavior upon low-dose administration of nicotine35. Further, conditional expression and activation of the rat capsaicin receptor TRPV1 in genetically targeted subsets of neurons in the mouse brain has provided the ability to stimulate desired neuronal populations in a conditional Cre-dependent manner28. Complementing these strategies has been the engineering of different G-protein coupled receptor (GPCR) families to respond to synthetic drug-like compounds36.
For many applications, inhibition rather than excitation of genetically defined neuronal populations is desired. To this end, a number of methods have been developed for in vivo experimentation. One such model takes advantage of wildtype GABAA receptor sensitivity to the allosteric modulator zolpidem, which normally enhances receptor function. In this design a conditional mouse model was generated that harbors a point mutation in the GABAA receptor γ2 subunit making it insensitive to zolpidem. Upon Cre-mediated recombination, wildtype receptor activity is reinstated to genetically targeted neuronal subsets, rendering those cells sensitive to pharmacological inhibition by zolpidem37. Orthogonal models have been used to drive neuronal hyperpolarization and inhibit action potential generation by heterologous expression of the C. elegans ivermectin-gated chloride channel31, or the Drosophila allostatin receptor, which inactivates neurons by opening G protein-coupled inward-rectifying K+ channels32. An alternate strategy has been to block synaptic transmission rather than induce hyperpolarization. Genetic expression of small-molecules for inactivation of synaptic transmission, or expression of Clostridium toxin fragments, has allowed selective inactivation of targeted synapses38,39,40.
Each of these approaches has unique limitations. For example, controlled inhibition by ivermectin-gated chloride channels relies on the availability of multiple subunits whose functionality requires the appropriate stoichiometry. On the other hand, small molecule-mediated inhibition of synapses39, which induces dimerization of genetically modified forms of presynaptic proteins, shows toxicity with prolonged exposure of dimerizer at high concentrations. Further, ligand application for neuronal activation requires transport across the blood brain barrier (BBB) and is concentration-dependent. Interestingly, the mouse model system for genetic control of neuronal activity by conditional TRPV1 expression may obviate some of these limitations. Multiple TRPV1 agonists and antagonists with an array of different binding affinities, solubilities, and BBB permeabilities offer the potential for pharmacogenetic manipulations of deep and disperse neuronal populations in a potentially noninvasive manner. A similar advantage in BBB permeability exists for compounds that activate engineered GPCRs36.
The fundamental element that couples cells of the brain into a functional network is the synapse. The basic input/output relationship of the synapse depends on a finite number of molecular elements, most notably the features of neurotransmitter release from presynaptic terminals and the number and properties of postsynaptic receptors. Synapses act cooperatively as “computational engines” to sum, transform, and relay input from discrete dendritic regions into specific neuronal outputs41. Thus, being able to visualize synapses, as well as discern and manipulate their properties, is paramount to understanding circuit function.
An important and realistic goal for the foreseeable future is to move beyond categorizing synapses by simply characterizing neurotransmitter phenotype, but to routinely define and visualize their functional states. Achieving this goal could help answer fundamental questions about properties of synaptic networks including: how strong are neighboring synapses relative to each other? How plastic is a given synapse? Is a particular synapse relatively active or mostly quiet? How do synapses function together within a given region of dendrite? Which synapses have high or low release probability? Is a synapse new or old? Has a synapse been recently potentiated or depressed? Visualizing such parameters of synapse status will ultimately require predictable molecular signatures for specific functional states.
Drawing on techniques ranging from fluorescence time-lapse imaging and photoswitching to single particle tracking, it is now possible to directly visualize single synapses and monitor movements of individual molecules within them42,43. A number of methods have been developedto follow proteins at synapses by targ eted expression of molecules tagged with FPs. Such methods allow for direct visualization of neurotransmitter release44,45,46 and dynamic movements of neurotransmitter receptors and postsynaptic scaffold proteins43. Expression of FP fusions can be used to measure fluorescence recovery after photobleaching as well as the complementary technique of fluorescence loss after photoactivation in a given cellular domain. These approaches have been broadly applied to measure apparent diffusion rates and protein exchange within synaptic microdomains47,48. Superecliptic pHluorin, which fluoresces at neutral pH but is quenched at the acidic pH of intracellular compartments, allows direct imaging of surface lateral diffusion, exocytosis and endocytosis of synaptic receptors, and neurotransmitter release in vitro or in vivo44,45,46,49,50,51. GFP reconstitution across synaptic partners builds on FP reporter technology by revealing information of specific synaptic pairs52. By tethering individual GFP fragments to separate pre- and postsynaptic proteins, reconstitution of GFP fluorescence is observed when genetically targeted cells form synaptic pairs. One possible limitation of this approach is the potential to induce phenotypes upon overexpression of synaptic protein fusions, which may themselves induce new synapses or alter synapse function. Gene targeted alleles that facilitate reporter expression from endogenous loci may mitigate such concerns. The range of FP reporters for visualizing synapse function continues to expand, and recent studies using Förster resonance energy transfer (FRET) reporters have begun to reveal dynamic protein activation events and protein-protein interactions accompanying synapse plasticity53,54.
Complementing genetic expression methods to label and track populations of synaptic molecules, single particle tracking (SPT) can reveal the dynamics of individual synaptic molecules in real time. SPT affords both high spatial and temporal resolution for single-molecule tracking by allowing the calculation of individual molecule diffusion, exchange, and dwell times. An important development in SPT has been the use of extremely bright semiconductor quantum dots (QDs), which facilitates precise spatial tracking of receptors in dendritic microdomains, and permits long-term tracking of receptors across dendritic segments55. SPT has been used to follow individual AMPA receptors at nanometer resolution within discrete domains of the PSD38, and more recently to monitor individual synaptic vesicle fusion events during neurotransmission56,57. QD technology itself continues to improve with the advent of smaller QDs, monovalent labeling, and improved targetability58.
At the tissue level, multi-photon imaging in living brain allows direct visualization of synaptic protein dynamics59 as well as spine morphological and synaptic plasticity within intact circuits16,17,60,61. Although in vivo imaging has provided a wealth of information into neuronal dynamics in the living brain, the approach is still constrained by the depth of target cells and is most useful for investigating superficial brain areas. The convergence of multiple techniques to mark single neurons, identify specific synapses on them, and monitor the dynamics of individual proteins, has rapidly expanded our understanding of synaptic organization and function. Ultimately, by employing this array of novel methods to investigate the dynamic synapse in vivo, we will begin to understand how circuit elements behave in the native environment of the intact brain and in response to sensory experience.
EM provides exquisite spatial resolution of neuronal features but does not allow dynamic imaging of cells in living tissue. Confocal and 2-photon light microscopy provides dynamic information at the expense of reduced spatial resolution. The emergence of super-resolution imaging using fluorescent microscopy provides the ability to bridge both worlds62. These new methods of nanoscale imaging in living cells circumvent classic optical limitations by altering the point spread function of collected light. These optical tricks allow super-resolution imaging of dendrites and spines that approaches the resolution offered by serial EM18,63,64.
The first major steps towards enhancing the resolution of light microscopy came in the 1990s when efforts were placed on increasing axial resolution. Methods such as 4Pi and I5M microscopy65,66 were grounded in perfecting the focus of light in the z-axis, though still constrained by the diffraction barrier in the x- and y-planes. By spatially mixing the frequency patterns of specimen illumination, structured illumination microscopy increased resolution by a factor of two67,68. Stimulated emission depletion (STED) microscopy, one of the first methods to achieve super-resolution, was formulated based on the ability to shape the point spread function by the non-linear de-excitation of fluorescent molecules in a ring surrounding a non-depleted region of interest69 (Box 2). Initially, STED imaging was used with synthetic small molecule dyes and thus has had limited utility for intact nervous tissue or genetically encoded reporters. However, recent improvements have enabled STED imaging of FP-expressing hippocampal neurons in organotypic slices, allowing substantial technical improvement in the ability to image spine neck width and spine head curvature64.
Super-resolution imaging techniques now allow light-microscopy to obtain information below the diffraction limit of light. As opposed to standard confocal imaging, which is constrained in resolution by the wavelength of light as defined by the Abbe limit (< λ/2n sin α) (A, left), super-resolution imaging is capable of nanoscale resolution (A, right).
Nanoscale imaging by targeted fluorophore depletion.In stimulated emission depletion (STED) microscopy, two lasers are used to generate a reduced focal volume. The excitation laser is used to excite a fluorescent molecule into the high energy excited state, and an additional offset laser is used to deplete the excited state without the production of fluorescence. The optical configuration generates an area in which molecules may be excited but do not fluoresce (B). This is the fundamental principle that allows techniques such as STED69, ground state depletion (GSD)82, saturated pattern excitation and saturated structured illumination microscopy (SPEM/SSIM)83, and reversible saturable optically linear fluorescence transition (RESOLFT) microscopy to image and localize molecules at the nanoscale. Although this technology is rapidly evolving, common application is still constrained by a handful of practical limitations. For many of these methods, the optics are relatively sophisticated to assemble or expensive to purchase, and the number of currently available fluorophores that can be readily used is limited and awaits development of new variants with different spectral properties, biophysical characteristics, and compatibility as protein fusions.
Nanoscale imaging by stochastic fluorophore switching.By broadly illuminating a sample while simultaneously imaging with a high-speed digital camera, individual fluorophore molecules that are sparsely and stochastically photoactivated are imaged sequentially. Photons emitted from these molecules can be differentiated in space by recording individual on-off states coupled with centroid analysis84. This approach allows one to record the position of many different molecules from distinct coordinates, since each photoactivated molecule renders distinct, resolvable spots at the camera (C). This is the principle behind the techniques of PALM70, FPALM71, STORM72, GSDIM85, and PAINT86. 3D-STORM now allows resolution approaching 10 times the diffraction limit (20–30 nm)87,88, whereas further resolution (sub-20-nm) has been gained with interferometric photoactivation localization microcopy (iPALM)89. A clear advantage of super-resolution imaging by stochastic fluorophore switching is its relative simplicity and the fact that each molecule is not forced to undergo many photoswitching cycles, as it is using RESOLFT, where photobleaching is a significant issue. A main disadvantage is the long acquisition times required to capture the stochastic events of single molecule photon emission. However, together these new forms of super-resolution imaging techniques allow a visual foray into the cell not previously possible.
Further expanding the super-resolution light microscopy repertoire, several new forms of single molecule imaging have been developed. The position of a single fluorescent object can be determined with nanometer accuracy by fitting the fluorescence signals from many emitted photons to a two-dimensional gaussian. Expression of photoconvertable proteins permits detection of single fluorescent molecules in the presence of a dense total population since the number of photoactivated molecules can be controlled by the intensity and duration of activation illumination. In such approaches, termed photoactivated localization microscopy (PALM)70, fluorescence photoactivation localization microscopy (FPALM)71, or stochastic optical reconstruction microscopy (STORM)72, each molecule’s centroid is calculated in successive rounds of stochastic activation and excitation. Ultimately, super-resolution is achieved by building spatial maps of sequentially activated molecules (Box 2). Together, these methods have opened a new realm of experimental possibility that now permits light microscopy to define cellular structures down to the nanoscale.
What can neuroscience gain by imaging molecular anatomy at the nanoscale? Many key neuronal and circuit properties are defined by synapse composition and function. Single molecule analysis is not new to neuroscience; the wealth of biophysical information gained from single channel recordings profoundly influenced neuroscience and the study of excitable tissue73. As with single channel recordings, nanoscopic imaging may move standard measurements of synapse and circuit function from macroscopic population measurements to stochastic probabilistic measurements at precise positions on single neurons. By revealing the biophysical properties of the molecules present at synapses, we might begin to assign discrete identities or ‘states’ to different types of neuronal connections, and thus more precisely ascertain their contribution to circuit function. In other words, tagging molecular identifiers to a given synapse may ultimately help in predicting the functional nature of the connections that define neural circuits. At the moment, molecular tags for synapse state remain largely science fiction, but emerging technologies are laying the groundwork for future tools to define and probe synaptic circuits.
A “connectome” is defined as a detailed map of the full set of neurons and synaptic connections within the nervous system of an organism1. What is the current state of available connectomes, and how will they ultimately be realized74,75? Notable progress has been made toward assembling anatomical circuit maps in model systems, ranging from completion4 to ongoing1. What will this information reveal?
Despite having a blueprint for the nematode C. elegans nervous system for more than 20 years, this map has served more as a detailed atlas of the worm’s nervous system than as a diagram for the flow of information through circuits that drive behaviors. Arguably however, this reference has been indispensable for characterizing specific neurons and their synapses, as well as serving as a guidebook for targeted transgenic manipulation. A missing link to realizing the full potential of any given circuit diagram is information regarding functional connectivity. Existing data sets do not reveal functional signaling properties of the circuit, or dynamic processes such as synapse plasticity. Perhaps most importantly, the current model is that connectivity data will serve only as snapshots, and we need to remain cognizant of the changing nature of neural circuits, particularly in more complex and malleable nervous systems.
Skepticism has always followed similar large-scale informatics-type projects, including sequencing of the human genome, high-throughput crystallography, and mapping protein interactions through searchable protein “interactomes”. Where have these large-scale (and oft-derided as unimaginative) approaches gotten us thus far? Whereas structural genomics and interactomes are in more nascent states, large-scale genome sequencing stands out as a clear success74.
Alternative approaches to build models of functional brain circuits are being pursued in the absence of model organisms. For example, one of the goals of the Blue Brain Project (http://bluebrain.epfl.ch/page17871.html) is to understand brain function by reconstructing anatomical and signaling properties of circuits through the assembly of known cellular, molecular, and computational information to generate detailed brain models in silico. Will animal experiments eventually be supplanted by computational and computer-assisted modeling? How would such models be validated? Which neural circuits should be fully elucidated? Although insufficient in isolation to understand the functionality of neural circuits, complete neuronal wiring diagrams are certain to provide a crucial frame of reference.
Neuroanatomical experimentation and functional circuit analysis in the future will certainly rely on a combined approach of genetics, high-resolution imaging, circuit tracing, and electrophysiology. These techniques will increasingly intersect as single experiments to assemble both anatomical and functional circuit information. For example, Brainbow technology allows multichannel stochastic coloring schemes for labeling neurons in the brain using the Cre/LoxP recombination system (Figure 2)76. A shortcoming of this approach is that it does not reveal functional synaptic contacts between neurons of different colors. One way in which this might be realized is to combine different synapse labeling methods within conditional Brainbow-like engineered tissue. Instead of using standard FPs for stochastic conditional expression, alternative reporters might include synaptically targeted FPs with different spectral properties.
Ultimately, it would be particularly valuable to be able to visualize functional connections. By combining conditional genetics in the mouse with recently described trans-synaptic viral labeling approaches, circuit-specific genetic dissection is conceivable. This might include trans-synaptic mobilization of FP spectral variants, optogenetic reporters, genetically encoded Ca+2 sensors, and even Cre recombinase across the synaptic cleft. By mobilizing Cre through a circuit in a trans-synaptic manner, conditional mutagenesis could be achieved across synaptic partners based on anatomy rather than gene expression. Technology of this type would allow cellular and circuit-dependent knockouts, activation of floxed genetically encoded neuronal activators, and the triggering of floxed fluorescent reporters such as the Brainbow system (Box 3).
Trans-synaptic tracing of neural circuits using viral technology could allow for genomic modifications of connected cells. Advances in viral engineering now allow genetic targeting of viral expression vectors to precise subsets of neurons; one can envision using a strategy similar to that described by Wickersham et al. to modify the genomes of interconnected neurons27. A) Scheme illustrating the genetic targeting of Cre expressing RV vectors to specified neurons in mice harboring conditional alleles for trans-synaptic mutagenesis. Cells with primary RV infection will be recombined, as well as those presynaptic to the initially infected neuron. All other neurons within the circuit will remain unaltered. As in Figure 1, the green plus sign represents monosynaptic viral transfer; the red cross represents a polysynaptic site that lacks viral transfer due to the absence of wildtype G protein. B–C) Trans-synaptic mutagenesis could potentially be used with conditional alleles of all sorts. These may include Cre-dependent conditional loss-of-function (B), or gain-of-function reporters (C). ChR2, channelrhodopsin-2; PA-FP, photoactivatable fluorescent protein.
By using creative FP expression strategies, conditional genetics, trans-synaptic tracing, volume EM, and defining molecular states of synapses, the functional anatomy of the nervous system is being more fully dissected. Coupled with cell and circuit level gene expression analysis, current efforts hold promise for assembling a detailed map of the full set of neurons and synapses that comprise key circuits of the nervous system. A complete atlas of synaptic connectivity will expand our ability to probe complex circuit function, allowing us to gain a deeper biological insight into processes ranging from memory and behavior to disease. Yet, it will be essential that our appreciation of the changing brain remains intact with the arrival of static connectivity maps.
Not all of the emerging technologies for probing cells and circuits will necessarily aid in our understanding of overall brain function. It is conceivable that these new techniques will simply continue to generate piecemeal data that has to be incorporated into a much larger framework in order to provide significant conceptual advances. But what is this larger framework? The challenge for the future will be to move from descriptive observations to explanatory theories that provide predictive models of complex brain functions. It seems reasonable to believe that such predictive models will emerge from the mesoscale of microcircuits. Of course full knowledge of the molecular and structural components of local microcircuits will be insufficient, and long-range interactions and inputs must be incorporated. Clearly brain processing relies on both, although limitations in technology to date have forced concentration on spatially proximate connectivity. Perhaps the future will merge cellular and circuit technologies with larger scale imaging techniques such as fMRI to combine knowledge of microarchitecture with macro-circuit function. Ultimately, a combination of these approaches will be needed to advance our knowledge of the brain. The promise of finalizing any accurate blueprint for the mammalian brain remains an ambitious goal for the future, but a palette of evolving genetic and imaging technologies provide the substrate to both formulate predictive models and validate existing theories to begin such a task. We now more clearly see the mountain; tall though it be, the climb is quickening.
We thank Ian Davison, David Fitzpatrick, Marta Fernandez Suarez, Kristen Harris, Juliet Hernandez, Matt Kennedy, Angela Mabb, Rich Mooney, Tom Newpher, Todd Roberts, Cam Robinson, Joel Schwartz, Richard Weinberg, Xiaowei Zhuang, and Jason Yi for helpful input and comments on the manuscript. We apologize to those whose work we could not cite due to space limitations. B.R.A. is supported by a K99 award from the NIH. Work in the lab of M.D.E. is supported by grants from NINDS, NIMH, and NIA at the National Institues of Health. M.D.E. is an Investigator of the Howard Hughes Medical Institute.