The brain of all higher animals is formed by a large number of interconnected neurons. Typically, neurons are grouped into larger assemblies (“brain compartments”), such as brain stem nuclei or cortical layers in the vertebrate brain, or neural lineages in the insect brain 
. The analysis of the structure, development, and function of the brain can therefore proceed at two levels: the level of individual neurons and synapses, and the level of brain compartments. Compartments represent structural and functional modules; interconnected by bundles of axons, they form “macro-circuits” that control certain aspects of behavior. Unraveling macro-circuits has been the mainstay of classical vertebrate neuroanatomy and physiology. Present-day studies employing functional imaging (e.g., 
) walk in the foot steps of this approach, given that the signals registered by MRI or PET scanners (for these and all other abbreviations, see Table 1) reflect the activity of large numbers of contiguous cells 
The study of macrocircuitry informs us of how the brain is built, which “packets of information” may interact, where in the brain this interaction takes place, and what output channels are activated to elicit a behavior that is correlated with the observed macroscopic brain activity. Addressing macrocircuitry leaves the question of how nervous tissue operates in processing information unanswered. To tackle this problem, an approach is required that considers the structure and connectedness of the building blocks of the brain—i.e., the neurons, neurites, and synapses (“microcircuitry”). The way in which a given neuron is tuned to a specific input stimulus, or the pattern of activity triggered in this neuron when providing a specific input, depends on the distribution of excitatory and inhibitory synapses that connect the neuron with its neighbors 
. Given the small size of neurites and synapses, the number and density of connections is immense. Calculations based on both light- and electron microscopic preparations point out that in mammalian brain, 1 mm3
contains more than 105
neurons, more than 4 km of axon and 500 m of dendrite, and more than 700 million synapses 
. The goal of the analysis of microcircuitry is to elucidate the geometric algorithms that describe the connectivity within small volumes of brains; based on these algorithms, one may hope we will be eventually able to model the neuronal activity and information flow pervading brain tissue while controlling sensory and motor activity of an organism.
Due to the small size of synapses and terminal neurites (0.1–0.5 µm), structural aspects of microcircuitry can be conclusively analyzed only electron microscopically. Traditionally, the acquisition and analysis of complete series of TEM sections required a considerable effort; as a consequence, studies of microcircuitry have mostly been restricted to small parts of neurons or neuropile compartments in (e.g., 
). The problem is now becoming solvable, at least for small brains (or small volumes of large brains) with digital image recording and specialized software for both image acquisition and post-processing 
. The resulting stacks of registered digital sections can be segmented and analyzed in their entirety. For the implementation of such a neuronal reconstruction pipeline, aimed at the analysis of microcircuitry of the Drosophila
larval brain, we used the TrakEM2 open source software.
serves as a favorable model system in which molecular pathways involved in a wide range of events, from neural stem cell proliferation, cell fate determination, neurite pathfinding, and neurite connectivity, can be studied (e.g.,
). With the help of sophisticated transgenic constructs (e.g., Gal4 lines; 
), one can target specific cell types, labeling these cells, or manipulating them genetically. As a result, Drosophila
also represents a great model system to study systems-level questions, such as how neural circuits develop or control behavior. Finally, Drosophila
(and insects in general) offers the advantage that its nervous system is formed by a relatively small number of genetically and structurally defined modules, the neural lineages 
. Early in development, a set of dedicated neural progenitors, called neuroblasts, segregate from the ectoderm and subsequently proliferate to form the neurons and glial cells of the CNS. Each neuroblast produces an invariant set (“lineage”) of neurons; these neurons form a genetic as well as a structural “module” of the brain (). Thus, axons of neurons belonging to the same lineage form a cohesive tract and arborize within discrete compartments of the brain. The axon tracts formed by neural lineages, and recognized from early larval stages into the adult brain, represent the structural/developmental units of brain macrocircuitry.
Drosophila first instar larval brain.
Lineage tracts represent an invariant, easily recognizable system of structural landmarks onto which smaller elements—i.e., microcircuits or individual neurons—can be mapped. We argue that for the efficient analysis of microcircuit data, it is advantageous to follow an approach that integrates light microscopic landmark structures, such as lineage tracts (the “macrocircuitry”), with the analysis of TEM datasets. The reconstruction of microcircuitry from serial TEM material involves two main operations. The first is to segment the profiles of neurites through many consecutive sections to establish branch points and synaptic contacts. Given the amount of time involved in manual segmentation (or proofreading of automatically segmented material; 
), there are currently two different strategies of serial TEM data analysis. One (“sparse strategy”) is to focus on a defined, relatively small set of neurons, such as sensory afferents, or motor neurons, and trace their profiles and synaptic outputs/inputs in their entirety. The other strategy, introduced in this article, is to break down the overall TEM stack volume into smaller “microvolumes,” in the range of 2×2×1.5 µm up to 5×5×5 µm, and reconstruct all profiles contained within these microvolumes (“dense strategy”). From these small volumes, structural parameters of terminal neurite connectivity (e.g., diameter, orientation, and density of terminal neurites; structure, size, and density of synapses) can be extracted. Irrespective of the strategy (sparse versus dense segmentation) chosen, an additional step is to put the connectivity data resulting from the segmentation into the context of brain macrocircuitry: what brain compartment is the microcircuit located in, what are its input and output channels, how does it compare to microcircuits located at other positions in the brain. To be able to make these connections, the operator has to be able to repeatedly switch between the EM and LM level and to rapidly navigate from one location/compartment to another.
The TrakEM2 software used in this study combines all the components required for data acquisition, registration, navigation, and segmentation (
. html). TrakEM2 enables the analysis of a brain volume at the microcircuit level (synapses, individual neurite branches) along with the macrocircuit level (brain compartments, axon fascicles). We illustrate the use of TrakEM2 towards this end by applying it to the brain of the first instar (L1) larval Drosophila
brain. As raw data we use two stacks of stitched and registered EM sections, one containing 500 sections that include most of the neuropile of one brain hemisphere, and another one of 250 sections including one complete abdominal neuromere of the ventral nerve cord. We have cropped out several small volumes and evaluated the structure and connectivity of the terminal neurites contained within these volumes. Furthermore, we have registered a confocal stack with the brain TEM stack and evaluated the accuracy with which we can translate light microscopic features from the confocal stack to the TEM stack. Our study is intended to develop an approach towards a comprehensive anatomical reconstruction of neuronal microcircuitry.