Semi-automated three-dimensional reconstructions of individual neurons reveal cell type-specific circuits in cortex

Abstract

Despite a long history of anatomical mapping of neuronal networks, we are only beginning to understand the detailed three-dimensional (3D) organization of the cortical micro-circuitry. This is in part due to the lack of complete reconstructions of individual cortical neurons. Morphological studies are either performed on incomplete cells in vitro, or when performed in vivo, lack the necessary cellular resolution. We recently reconstructed the in vivo axonal and dendritic morphology of two types of L(ayer) 5 neurons from vibrissal cortex. The 3D profiles of short-range as well as long-range projections indicate that L5 slender-tufted and L5 thick-tufted neurons represent very different building blocks of the cortical circuitry. In this addendum to Oberlaender et al, PNAS 2011, we motivate our technical approach and the advancements this may give in reconstructing the cortical micro-circuitry.

This article refers to:

Due to its well-defined columnar and laminar organization, the vibrissal cortex in rodents is a highly suitable model to study cortical circuits.1–3 In particular, the segregation into column and septum related regions of the vibrissal cortex allows one to align and compare 3D morphologies of neurons from different animals. In our in vivo approach, the individual neuron reconstructions are therefore always performed in the context of a common reference frame containing three anatomical landmarks: (1) the barrel contours in granular layer 4, (2) the pia surface and (3) the white matter. Biological variability in brain size or dimensions of the vibrissal area can then be accounted for by linear remodelling of the common reference frame.

Such reconstructions revealed that 3D axonal morphologies in vivo4,5 can be an order of magnitude larger compared to previously reported values obtained in brain slices.6,7 This could be due to limited visibility of axons in 300–400 µm thick slices,8 or more plausible, axonal structures in in vitro preparations represent not more than 10% of the possible axonal projections present in vivo. The vast innervation volumes of in vivo filled axons also imply that manual reconstructions are highly labor intensive. This makes it virtually impossible to get the high throughput required to reconstruct representative amounts of all the different neuronal cell types found in the cortex9 and thus asks for more automated approaches.10 These issues aside, perhaps the most important problem of manual reconstructions is that intricate morphologies makes the chance for human error very likely, even for the very skilled human tracer. This motivated us to develop a semi-automated reconstruction pipeline.11–13 This method allows algorithm-based (and therefore reliable) reconstructions and reduces manual labor for individual reconstructions from ∼90 hours to 8–10 hours (the manual labor involves splicing of serial sections and not reconstruction). In short, this technique allows highly accurate quantitative axonal reconstructions that can be achieved in relatively little time.

The first quantitative study we performed using this technique was to reconstruct the axonal projection patterns of two types of layer 5 pyramidal neurons filled with biocytin in vivo (Fig. 1).4 These slender-and thick-tufted neurons can be classified based on their dendritic morphology and respond differently to passive whisker touch or active whisker movements.14,15 We found that axonal projections of slender-tufted and thick-tufted neurons target different layers and thus represent functionally and anatomically distinct units of the cortical micro-circuitry. L5 slender-tufted neurons displayed wide-spreading axonal projections (86.8 ± 5.5 mm), which primarily innervated supragranular layers of the entire vibrissal cortex and higher order cortices (dysgranular zone, posterior parietal cortex). L5 thick-tufted neurons in turn, are characterized by shorter and less complex axonal projections (31.6 ± 14.3 mm), which primarily innervated nearby infragranular layers. These results indicate that the use of semi-automated reconstructions of axonal projection profiles provides detailed new insights into the putative postsynaptic targets of individual neurons. Furthermore, full 3D axonal reconstructions are a crucial step in generating hypotheses about the pathways of cortical information processing. There is virtually no limit to the semi-automated reconstruction pipeline and it may be used to reconstruct neurons from any brain area of choice.

Ultimately, the combination of anatomical reconstructions and electrophysiological recordings of the activity of individual neurons during different behavioral states will generate insight into the functions of different cell types and facilitate the reconstruction of cortical circuits. These experimental approaches will pave the way to conduct simulations of anatomically realistic networks and may lay the foundation for future studies on brain degenerative diseases, such as Alzheimer's. This research will thus not only help to advance our basic understanding of cortical circuits, but may eventually provide a starting point to study the influences of brain diseases on neuronal morphology and function.

Figures and Tables

Figure 1 Semi-automated reconstruction of a Layer 5 pyramidal neuron filled with biocytin in vivo. (A) Mosaic scanning and subsequent serial reconstruction is performed on consecutive 100 µm thick tangential sections, which results in high resolution 3D images representing cubic millimeters of cortical volumes. The consecutive slices are aligned by using blood vessels that run perpendicular to the cortical surface. (B) Magnification of the area indicated by the asterisk. Note the abundant axons running through this area. (C) Automated detection of biocytin labeled processes allows fast and reliable reconstruction of axonal morphology. Axonal reconstructions from additional Z values are also visible in this part. (D) Example images of a single axonal branch labeled with biocytin and (E) the subsequent automated reconstruction.

Acknowledgments

This work was supported by the Max-Planck Society, the Center for Neurogenomics and Cognitive Research at Vrije Universiteit Amsterdam, and a Veni Grant from The Netherlands Organization for Scientific Research (to C.P.J.d.K.).

Addendum to: