Marmoset Brain Architecture

Technical White Paper: Marmoset Brain Connectivity Atlas

The marmoset brain connectivity atlas is part of the Brain Architecture Project. Our immediate aim is to create a systematic, publicly available digital repository for data on the connections between different cortical areas, in a primate species. This initial stage takes advantage of a large collection of materials obtained over two decades of research, using fluorescent retrograde tracer injections in adult marmosets (Callithrix jacchus). Whereas several research papers have already stemmed from this material (e.g. Rosa et al. 2009; Burman et al. 2011; Reser et al. 2013), the present project represents a new initiative, towards collating these data and making them publicly available, in their entirety. We believe that sharing these materials via a digital interface will address many of the current limitations of the traditional media used for communication of neuroanatomical research, including allowing access to the entire data set (as opposed to the few sections typically illustrated in journal articles), and enabling other interpretations of the data, in light of the future evolution of knowledge about the marmoset cortex. Among other positive consequences, we hope that the availability of raw data, which can be analysed independently in different contexts, will reduce the number of animals that need to be used for research on the organization of the primate nervous systems.

As part of the present effort, we have developed techniques that allow the visualization of data (positions of injection sites, and labelled neurones) obtained in different individual monkeys, relative to the histology of the cortex. These “raw” data can be visualized in serial sections, and interpreted independently without prior assumptions regarding the identity of specific cortical areas. However, we also realize that in many cases the utility of this resource will also depend on information about the likely identity of the areas and subcortical nuclei. In the near future this resource will be expanded to include drawings of sections from a brain in which the cortical areas are labelled according to the most current stereotaxic atlas available for the marmoset (Paxinos et al. 2012).

Furthermore, one of the main advantages of preserving data in digital format is the possibility of co-registering data from many different experiments into a single, navigable graphic representation (a marmoset brain template). In principle, this approach allows quantitative and probabilistic analyses that cannot be performed when considering data from one, or even a few injections at a time (Van Essen et al. 2012). As a step in this direction, we intend to provide the ability to visualise the data as "warped" into a template cortex, reconstructed from the Paxinos et al. (2012) atlas. These will be accessed either as three-dimensional models of the cortex at its mid-thickness, or as bi-dimensional "unfolded" maps.


The original data collection underlying the present release was collected for a number of scientific projects, which were supported by project grants from the Australian Research Council (A09937020, DP0451206, DP0878965) and the National Health and Medical Research Council (237009, 384115, 384116, 491022, 545982). A list that includes papers resulting from these projects can be found here. The Australian Research Council has, in addition, provided project grants (DP110101200 and DP140101968) specifically aimed at the development of the online atlas of the marmoset brain connections, including an initial demonstration of the feasibility of this approach. Finally, the present release has benefitted from sharing tools and techniques that were originally developed for the NIH-funded Mouse Brain Architecture Project (RC1MH088659 and RO1MH087988). We acknowledge use of the facilities and technical assistance of Monash Histology Platform, Department of Anatomy and Developmental Biology, Monash University. We also acknowledge internal support from both Monash University and the Cold Spring Harbor Laboratory, which was instrumental in allowing cooperative research through reciprocal visits and staff exchange.

Why the marmoset?

Many of the techniques that provide the clearest insights on the neural basis of behaviour can only be applied to animal studies, from the production of transgenic and knockout models to physiological recordings of the electrical activity of single neurones, and high-resolution anatomical tracing. The majority of studies use rodents (Manger et al. 2008), which have provided fundamental insights on issues such as the mechanisms that regulate development of the nervous system, and synaptic physiology. However, the brains of rats, mice, and many other animal models are organized quite differently from primate brains, at the level of specific anatomical circuits. These differences include systems that are of great interest, such as those involved in the comprehension of complex visual and auditory patterns (e.g. faces and voices), and in the control of hands, and eyes. To achieve an understanding of the neural bases of these functions, neuroscientists rely on primate models (Roelfsema and Treue, 2014). Although there are also obvious differences, monkey brains share several of the key anatomical circuits that differentiate us from other mammals (Orban et al. 2004).

The marmoset monkey (Callithrix spp.) is emerging as a choice animal model in neuroscience. Marmosets are small monkeys (300-400 g adult weight), which show accelerated development in comparison with most other primates. For neuroscientists, the marmoset is important as the simplest organism that shares many of the features that make primate brains special. For example, they have well developed frontal and temporal lobes, a sophisticated visual cortex, capable of fine discriminations, multiple cortical areas involved in planning of movements, and systems involved in the interpretation of complex patterns of vocal communication. Yet, the topology of the marmoset brain (in particular, the cerebral cortex) is much simpler than that observed in other commonly used primates, such as the rhesus monkey. This simplifies considerably the task of registering data from different individuals to a common brain template, and makes marmosets ideal models for the optimization of the software tools that will turn a practical primate connectivity atlas into reality. Finally, marmosets are the first primate species for which stable transgenic lines have been established (Sasaki et al. 2009), an achievement that opened the way for much-needed research on the genetic and developmental bases of primate cognition.

Tracer injections

The study of neural connections is based on substances known as neuroanatomical tracers. In a typical experiment, a surgical procedure under anaesthesia is conducted to inject few hundred nanolitres of tracer at a brain site, using a precision microsyringe. The tracer molecules diffuse in the extracellular medium, and become picked up by nearby neurones. Over a period of several days, these molecules migrate along axons, therefore revealing anatomical “bridges” that connect distant brain sites. These experiments can reveal connections that link small clusters of neurones located millimetres or centimetres away, while ignoring the overwhelming majority of intervening and adjacent cells. When considered in reference to cortical areas and thalamic nuclei, these connections tend to be very specific, and reproducible in different individuals of a same species.

The material included in the initial release of the Marmoset Brain Connectivity Atlas is based on the use of fluorescent retrograde tracers. These are tracers that are incorporated at the synaptic terminals within the injection site, and then migrate towards the parent cell body (Figure 1, left). Retrograde tracers reveal which cells are sending information to a given neuronal population. Up to four retrograde tracers, each distinguishable by colour under a fluorescence microscope, can be used in a same animal, thus enabling the probing of connections of different brain sites in a same case. Other tracers are referred to as anterograde tracers (Figure 1, right). They are transported from the cell bodies to the synaptic terminals, thus revealing which cells are receiving information. It is planned that future expansion of the Marmoset Brain Connectivity Atlas will accommodate data obtained with anterograde tracers.

Figure 1

Figure 1: Tracers that are incorporated at the synaptic terminals within the injection site, and then migrate towards the parent cell body, are referred to as retrograde tracers (Left diagram). Retrograde tracers reveal which cells are sending information to a given neuronal population. Other tracers are referred to as anterograde tracers (right). They are transported from the cell bodies to the synaptic terminals, thus revealing which cells are receiving information.

Standardized procedures were used in creating the tracer injections, analysing, and reconstructing the data across different experiments (for details, see Burman et al. 2011; Reser et al. 2013). Adult marmosets (2-4 years old) of either sex were used. Metadata for each animal are provided in conjunction with each injection case. The only criterion for exclusion was evidence that the effective uptake zone of the tracer injection (determined according to Condé et al. 1987 and Schmued et al. 1990) had invaded the white matter subjacent to the cortex. Tracers were aimed at different areas using stereotaxic coordinates obtained as part of earlier studies (for a recent review, see Paxinos et al. 2012). The exact placement of each tracer injection relative to distinct cytoarchitectural fields was determined later, following post-mortem reconstruction, both qualitatively (and histological inspection), and quantitatively (by registration to a brain template).

The fluorescent tracers fluororuby (FR; dextran-conjugated tetramethylrhodamine, molecular weight 10 000), fluoroemerald (FE; dextran-conjugated fluorescein, molecular weight 10 000), fast blue (FB), and diamidino yellow (DY) were injected using a 1 µL microsyringe fitted with a fine glass micropipette tip (see Table 1 for details). Each tracer was injected over 15–20 min, with small deposits of tracer (0.02 µL each) made at different depths. Following the last deposit (typically at a depth of 300 µm), the pipette was left in place for 3–5 min to minimize tracer reflux. The retrograde fluorescent tracers diamidino yellow (DY) dihydrochloride and fast blue (FB) were either injected in this same manner, or directly applied into the cortex as crystals, with the aid of blunt tungsten wires (Rosa et al. 2005).

Tissue processing and data analysis

After postfixation in the same medium for at least 24 h, the brains were blocked and, over the next few days, were immersed in buffered paraformaldehyde solutions containing increasing concentrations of sucrose (10%, 20%, and 30%). Frozen 40-μm coronal sections were then obtained using a cryostat. Every fifth section was mounted unstained on glass slides, air-dried, and coverslipped with di-n-butylphthalate-xylene mounting medium for analysis of neurones labelled with fluorescent tracers. These sections were stored in the dark at 4 °C. Consecutive series were stained for cell bodies (using cresyl violet), myelinated fibres (Gallyas 1979), and cytochrome oxidase activity (Wong-Riley 1979).

Both the Nissl and myelin stains are presented in the present release of the marmoset brain connectivity atlas. Here, it is important to recognize that histological preparations vary in quality from case to case. Thus, our decision to provide images of all available sections necessarily meant illustrating sections that we do not consider optimal, either due to artefacts of cutting and staining, or to degradation following many years in storage. We hope that this decision to illustrate the materials “warts and all” will be seen as a positive step by our colleagues, in the sense that this will highlight aspects that deserve further investigation.

Sections were examined using Zeiss Axioplan 2 or Axioskop 40 epifluorescence microscopes. Labelled neurones were identified using ×10 or ×20 dry objectives, and their locations within the cortex and subcortical structures were mapped using a digitizing system (MD3 digitizer and MDPlot software, Accustage) attached to the microscope. To minimize the problem of overestimating the number of neurones due to inclusion of cytoplasmic fragments, labelled cells were accepted as valid only if a nucleus could be discerned. This was straightforward in the case of DY, as this tracer only labels the neuron's nucleus (Keizer et al. 1983). In the case of tracers that label the cytoplasm (FB, FE, and FR), the nucleus was discerned as a profile in the centre of a brightly lit, well-defined cell body, which in the vast majority of cases had an unmistakable pyramidal morphology. The entire brain was scanned in the examined series (1 in 5 sections), and every labelled neuron was plotted.

Each case in the atlas is defined by the site of a tracer injection, and by all neurones that send connections to that site (Figure 2). This information is represented in large collection of sections (>400), from anterior to posterior in the brain. Within each section we encode information about the exact location of neurones that incorporated the tracer, in medial-lateral and dorsal-ventral coordinates. For example, in the case shown in Figure 2, an injection of tracer in the frontal pole of the cerebral cortex (small section, on the right) resulted in hundreds of labelled neurones in the temporal lobe and thalamus (red points indicated in the larger section, on the left).

Figure 2

Figure 2: Illustrated are two sections through the brain of a marmoset (scale bar= 1 mm), showing a tracer injection in the frontal lobe (right) and the location of cells that send connections there (red dots on the left). Hundreds of such sections need to be combined to generate 3-d reconstructions for registration on a brain template (bottom).

Image Acquisition

The Nanozoomer HT Virtual Microscopy system (Hamamatsu/Olympus) is used to digitally image all tissue samples resulting from the pipeline. All scanning is performed using a 20X objective (0.46 µm per pixel).

Data Conversion and Data Storage

Shortly after image acquisition, the whole-slide data are cropped into individual sections and converted into a custom image data structure. Quality control is performed to ensure that the cropping was done correctly. All data are subsequently transferred to a high-performance storage cluster, through an integrated 10G copper network. Daily backups of the data are created. Two sets of images are prepared. High quality lossy compression is prepared for everyday analysis and for website display. And lossless compression is used to archive raw image material for future references.


The process of cutting the brain into sections and mounting these on slides involves a degree of imprecision, such that sections are not aligned to each other, to or any coordinate system. In particular, the locations of labelled neurones are plotted from a one particular series of sections (unstained sections) and thus do not correspond directly to the sections for which the histological information is available – Nissl sections. Therefore, there were registration step is required match the fluorescent sections with the Nissl sections.

The MDPlot drawing files were parsed with custom Python programming language scripts to extract the section outlines and labeled cell locations. Then, scalable vector graphics (SVG) files were generated, in which the MDPlot drawings were superimposed on the Nissl section images. Each drawing was manually aligned to the Nissl section, which generally involved a simple set of translation, rotation and scaling operations. Occasionally, manual editing of the drawing was required to make it correspond correctly with the Nissl section, typically in regions of the tissue that are prone to distortion, such as sharp bends and section folds and tears. After the alignment the coordinates of the cells in the sections' coordinate system are stored in the website database.

Data Presentation Pipeline

Nissl section images were converted into the JPEG2000 (ISO/IEC 15444-1) image format using custom scripts based on the Kakadu toolkit (Taubman & Marcellin, 2001). The project website has been developed using a combination of technologies, and allows for browsing and viewing high-resolution images from Internet via moderate bandwidth. The JPEG2000 images are served over the web using a modified version of the Djatoka Image Server (Chute et al., 2008) and a custom web based image viewer is built around OpenLayer 3 to utlize it's features of interactive pan, tilt, zoom, labeling and annotation. It also supports online adjustment of RGB dynamic range, as well as gamma adjustment.

Cells are plotted using vector drawing on top of the Nissl layer. Each type of injection will have separate drawing layer to enable the user to toggle display of individual type of cells. Injection halos are annotated as well. By introducing various optimization methods, a moderate machine with recent browser would be able to inspect section with hundreds of thousand of markings with little to none performance impact.