Sensory maps in the olfactory cortex defined by long-range viral tracing of single neurons

Abstract

Sensory information may be represented in the brain by stereotyped mapping of axonal inputs or by patterning that varies between individuals. In olfaction, a stereotyped map is evident in the first sensory processing centre, the olfactory bulb (OB), where different odours elicit activity in unique combinatorial patterns of spatially invariant glomeruli^1,2. Activation of each glomerulus is relayed to higher cortical processing centres by a set of ∼20–50 ‘homotypic’ mitral and tufted (MT) neurons³. In the cortex, target neurons integrate information from multiple glomeruli to detect distinct features of chemically diverse odours^4,5,6. How this is accomplished remains unclear, perhaps because the cortical mapping of glomerular information by individual MT neurons has not been described. Here we use new viral tracing and three-dimensional brain reconstruction methods to compare the cortical projections of defined sets of MT neurons. We show that the gross-scale organization of the OB is preserved in the patterns of axonal projections to one processing centre yet reordered in another, suggesting that distinct coding strategies may operate in different targets. However, at the level of individual neurons neither glomerular order nor stereotypy is preserved in either region. Rather, homotypic MT neurons from the same glomerulus innervate broad regions that differ between individuals. Strikingly, even in the same animal, MT neurons exhibit extensive diversity in wiring; axons of homotypic MT pairs diverge from each other, emit primary branches at distinct locations and 70–90% of branches of homotypic and heterotypic pairs are non-overlapping. This pronounced reorganization of sensory maps in the cortex offers an anatomic substrate for expanded combinatorial integration of information from spatially distinct glomeruli and predicts an unanticipated role for diversification of otherwise similar output neurons.

Main

Chemical components of odours are detected by a large repertoire of olfactory sensory neurons (OSNs) based on their expression of one of ∼1,400 odorant receptor (OR) genes (in the mouse)⁷. The axons of OSNs expressing the same OR coalesce in the olfactory bulb (OB) to form two glomeruli at mirror symmetric locations, where they form synapses with mitral and tufted (MT) neurons^2,8. This produces a stereotyped topographic map in which different odours activate characteristic subsets of glomeruli and their associated MT neurons^1,9. Odour recognition is thought to involve integration of inputs from different MT neurons, but the organization of glomerular inputs to cortical targets has not been described and whether sets of homotypic MT neurons are similar or diverse in cortical targeting is not known^4,5,6.

In other senses, such as vision, hearing and touch, adjacent neurons often respond to similar sensory stimuli and send axons to closely apposed cortical targets, producing sensory maps that are similar between individuals^10,11,12. In contrast, anterograde and retrograde tracing of inputs to the primary olfactory cortex have shown that neighbouring MT neurons project to broad cortical regions and that small cortical regions receive input from distributed sets of MT neurons^13,14. However, these studies did not examine neurons associated with a particular glomerulus. Therefore it is not clear whether the organization of MT axons in the cortex retains aspects of the glomerular organization of the OB, or whether sensory maps in the cortex are stereotyped or divergent between individuals (Supplementary Fig. 1).

One barrier to mapping olfactory neural circuits is the lack of long-range axon tracing methods that do not obscure the glomerular identity of the MT neuron. We have developed a viral tracing system that overcomes these obstacles. This system exploits an attenuated strain of Sindbis virus to express high levels of three different ‘colours’ of fluorescent proteins in neurons after transduction with a single virion (Fig. 1a). This produces rapid labelling of MT dendrites and distant axonal branches without spillover effects at the injection site, allowing us to link neurons to a specific glomerulus and trace their projections with single-cell resolution, in three colours.

Figure 1: Different organizations of MT axons in the AON and PC.

a, Sindbis virus vectors were altered so that infection with one virus produces two fluorescent proteins and results in green (pSIN-G), yellow (red+green, pSIN-Y) or red (pSIN-R) labelling. b, Schema of OB labelling with separate (SEP: light grey) or colocalized (CO-INJ: dark grey) injections. c, d, Segregation is maintained only in the AON based on analyses of confocal images from separate injections (n = 4 animals) and control co-injections (n = 2 animals) using two mixing indices (Student’s t-test, *P = 0.003 (c), *P = 0.03 (d)). Error bars: 95% confidence. e–g, Some axons diverge from their neighbours as shown by tracing a red neuron (white arrows) that intermingles with yellow neurons in the OB (e) yet projects to a different region of AON pE which is primarily innervated by green axons from a distant OB region (f, g). Panels in f (left to right) are coronal OB sections, 400-μm apart. Blue, TOTO-3 nuclear staining. Scale bars, 100 µm. Panel e was rotated so lacks scale bars.

PowerPoint slide

To determine whether MT neuron projections retain the stereotypic organization of the OB in different cortical targets, we labelled separate clusters of MT neurons with different coloured viral tracers and examined their axonal distributions in the anterior olfactory nucleus pars externa (AON pE) and the pyriform cortex (PC) (Fig. 1b, Supplementary Fig. 2 and Supplementary Methods). In the AON pE, axons remained segregated, as previously reported^15,16. Yet in the PC, the same axons become extensively intermingled, based on visual inspection and two quantitative indices of axon mixing (Fig. 1c, d, Supplementary Fig. 2 and Supplementary Methods). We speculate that the gross-scale spatial map in the AON pE may facilitate contralateral comparisons, whereas more complex patterns in the PC may permit increased combinatorial coding^6,17. Intriguingly, in the AON pE, some MT axons diverge from their neighbours and intermingle with axons of distant MT cohorts, indicating that despite a gross-scale similarity, mapping in the AON pE differs from mapping in the OB at the single cell level (Fig. 1e–g).

Next, to determine how glomerular information is relayed to the AON and PC, we labelled individual MT neurons associated with a defined glomerulus. To accomplish this, we used mouse strains in which a stereotyped glomerulus is filled with green fluorescent protein (GFP) based on co-expression with a particular OR gene (mOR174-9, also known as OR-EG or Olfr73) (Fig. 2a, b). Tracing an individual MT neuron reveals local tufts in AON regions as well as extensive branching throughout the AONpE, AON pars principalis (pP) and anterior PC, with weaker signal in distal targets (Fig. 2c, Supplementary Fig. 4). This indicates that glomerular mapping by individual MT neurons is unlikely to be restricted to small selective cortical regions, which contrasts with the conclusions of studies in the rabbit that employed less sensitive tracers (Supplementary Methods and Supplementary Movie 1)¹⁸.

Figure 2: Three-dimensional reconstructions of individual M/T neurons reveal extensive branching and intrabulbar mapping.

a, Injections were targeted to the EG glomerulus in vivo using GFP (green in schema and fluorescence image, top right). Serial coronal sections were imaged using confocal scanning (lower right). b, Assignment of glomerular identity by tracing the primary dendrite of a neuron labelled with SIN-Y (yellow) into the EG glomerulus (green). Scale bar, 100 µm. c, Tracing axonal arbors with Neurolucida software produces a three-dimensional reconstruction in which branches are assigned to target regions. Green, EG glomeruli; purple, AON; cyan, PC; red, OT; yellow, taenia tecta. Blue axis, anterior–posterior; green axis, ventral–dorsal; red axis, medial–lateral. This neuron is from a 7-week-old mouse, all others are from 3-week-old mice. d, Three-dimensional reconstruction of the neuron in (b) reveals an intrabulbar axonal projection to the granule cell layer underlying the mirror symmetric EG glomerulus (white circle).

PowerPoint slide

A second new finding is that the axons of some MT neurons branch within the OB and innervate a small region of the granule cell layer, just proximal to the mirror symmetric glomerulus (Fig. 2b, d). This type of symmetric intrabulbar projection has been noted for external tufted (eT) neurons located in the glomerular layer, but not for neurons in the mitral cell layer¹⁹. Based on their morphology and internal projection, we refer to these as internal tufted neurons (iT). In contrast with the eT neurons that project extensively to the olfactory tubercle, iT neurons project primarily to the AON and PC and do not exhibit detectable subtype-specific patterning (Supplementary Figs 3, 5 and 6)²⁰. Consequently, iT and other MT neurons have been grouped in subsequent analyses.

To determine whether the complex patterns of MT axon branching and termination are stereotyped between individuals with respect to their glomerular input, we built a reference brain in which individual neuronal reconstructions could be aligned and mapped to different target regions (Supplementary Methods and Supplementary Fig. 4). Comparisons of homotypic EG-MT neurons (n = 14) and non-EG neurons (n = 8) revealed no detectable stereotypic patterns of branching or gross scale morphology (Fig. 3a–c and Supplementary Fig. 5).

Figure 3: Diverse patterns of glomerular inputs to the cortex.

a, b, Fully traced EG (n = 14) and non-EG (n = 8) neurons were aligned to the reference brain to identify similarities between homotypic EG neurons (a), compared with non-EG neurons (b) (displayed traces were trimmed for clarity). Neurons branched extensively in the AON pE and pP (purple) and the anterior PC (cyan/blue). Blue axis, anterior–posterior; green axis, ventral–dorsal; red axis, medial–lateral (for a and b). c, Total primary axon branches in the dorsal or ventral direction were assigned to the AON pE, pP or aPC based on the reference brain. Primary branches that innervated a region (as percentage of total branches in that neuron) were averaged. No significant differences were observed between EG and non-EG neurons. Error bars, 95% confidence intervals. d, Proximity of axon termini and axon nodes were assessed for each pair of seven single MT neurons. Comparisons of homotypic EG-EG neuron pairs (n = 12) with heterotypic pairs (Non) (n = 30) revealed no differences in the average distance between nodes (d, left) or termini (d, right). Error bars, 95% confidence intervals, Student’s t-test P > 0.05.

PowerPoint slide

Although the gross morphologies and branching patterns for homotypic OR-EG MT neurons are diverse, it remains possible that they exhibit stereotypy at axon termini, where synapses are likely to form. To address this possibility we aligned pairs of MT neurons in the reference brain and computed the distances between every axon terminal and its nearest neighbour in the comparison neuron. Similar analyses were performed using branch node locations to assess the closeness of fine distal branches. Control data sets show that this method detects similarities in branching and axon termination (Supplementary Fig. 7). However, homotypic EG-MT neuron pairs (n = 6) were no more similar than heterotypic pairs by any metric (n = 15) (Fig. 3d, Supplementary Methods and Supplementary Fig. 7).

Therefore, in contrast to other sensory systems, MT neurons ostensibly bearing similar sensory information do not branch or terminate in stereotyped positions. However, because these neurons were traced in several animals, differences in neural activity among animals or intrinsic differences in brain morphology may contribute to the observed variability. Alternatively, this diversity might be characteristic of homotypic neurons even in the same animal. To distinguish between these models, we injected multi-colour viral mixtures in the OR-EG region and identified sets of two to four neurons that could be traced to defined glomeruli, resulting in same-animal homotypic pairs (n = 6) and heterotypic pairs (n = 7) (Fig. 4a–d and Supplementary Fig. 8).

Figure 4: Extensive diversity among homotypic MT neurons in the same animal.

a, Viral labelling of a homotypic EG-MT neuron pair (collapsed confocal sections overlaid with traces). b, Homotypic axons in the AON pE exhibit distinct tufts. c, Two weakly-labelled non-EG M/T neurons (red, white) intercalate between the EG neurons (yellow, green). For a–c: yellow, SIN-Y; green, SIN-G; blue, nuclei. Blue axis, anterior–posterior; green axis, ventral–dorsal; red axis, lateral–median. d, Same-animal pairs of homotypic (HOM) and heterotypic neurons (HET) show diversity in primary branching. Overlapping branches (grey arrows) are shown. HOM 1 is the pair in a. Three EG neurons were labelled in one animal (HOM 4). Four neurons (two EG, two non-EG) were labelled in another animal (HOM 3/HET 3). Cortical regions: white left, AON pE; grey, AON pP; white right, PC. Main axon lengths vary between 1.9 and 2.7 mm, reflected by the width of the grey box. e, Even in the same animal, homotypic (HOM, n = 6) and heterotypic (HET, n = 7) neuron pairs are equally diverse in branch location (top) or percentage of unique primary branches per region (AONpE, AON pP and anterior PC; PC) (bottom). f, Compiling pairs from multiple animals (n = 34 HOM; n = 55,HET) revealed increased branch overlap in homotypic neurons in the AON pP (top), which is not significant when normalized to the total branches in that region (bottom). Error bars show s.e.m. (top) or s.d. (bottom) Significance was assessed by a Mann–Whitney U Test (*P = 0.008) and by Student’s t-test adjusted for multiple comparisons (*P = 0.004, for P < 0.05, P_adj = 0.016).

PowerPoint slide

Same-animal analyses revealed extensive diversity among homotypic neurons. The majority (∼70–90%) of primary branches among homotypic pairs are distinct in each target region (Fig. 4d, e). Homotypic pairs exhibit no more overlapping branches than do heterotypic pairs (Fig. 4d–f and Supplementary Fig. 8). Even when homotypic pairs seem to branch at the same point, fine-scale analyses reveal additional diversity. For example, in the AON pE two homotypic axons extend fine tufted collaterals in roughly the same region (Fig. 4b), yet heterotypic non-EG neurons project axonal tufts that interdigitate between the EG-MT axons (Fig. 4c).

Finally, it is possible that a subset of EG-MT neuron branches might display stereotypy, while others do not. To identify any overall patterns of similarity in EG-MT neurons that could reveal a partial stereotyped glomerular map in the cortex, we overlaid each possible pairing of matched homotypic EG-EG neurons (n = 34) and unmatched heterotypic sets of neurons (n = 55) and calculated the numbers of overlapping branches per pair, at three stringency levels. This revealed an increase in the average number of overlapping branches in the AON pP for homotypic OR-EG pairs. Whereas this result suggests that some stereotypy exists at the glomerular level in the AON pP, we note that this difference reflects only a modest overlap (∼1.5 overlapping branches vs. ∼1) that is not significant at high stringency or when normalized to the total branches per region (Fig. 4f and Supplementary Methods).

Axon branching is an important determinant of the functional capacities of neural circuits, but the operational feature of a circuit is synaptic connectivity. Our studies cannot directly address this due to the lack of appropriate anterograde trans-synaptic tracing techniques. However, our characterization of axon ends reveals swellings at ∼90% of axon termini and varicosities in distal axon segments that have been suggested to represent sites of synapses (Supplementary Fig. 9)¹⁵. If axon termini indeed represent sites of synapses, our studies indicate that homotypic neurons form synapses with distinct sets of target neurons. For example, distances between nearest-neighbour axon ends in homotypic pairs often exceed 600 µm, which is larger than the reported dendritic fields for target neurons in the AON and PC (∼200–500 µm)^21,22. Recent retrograde transynaptic tracing studies provide direct support for this finding²³.

Here, we define the anatomy of the neural circuit that links odour detection in the nose with combinatorial processing in the brain. We show that diversity in the projections of MT neurons produces new cortical organizations of sensory information that differ from the OB in two ways. First, we observe extremely limited stereotypy in projections from a glomerulus, even in regions that maintain gross-scale stereotypic patterning. Second, the patterns of innervation of MT neurons are broadly distributed, suggesting that the set of ∼20–50 MT neurons are anatomically positioned to form synapses with a very large set of target neurons, which contrasts with the very precise patterns of inputs observed in the OB. This divergent, non-stereotyped organization contrasts with the highly stereotyped branching patterns of pairs of analogous neurons in the fly olfactory system. However, our findings concur with results of recent multi-neuron studies of the mouse olfactory cortex, and with single neuron tracing in the zebrafish^23,24,25,26. We speculate that the increased numbers and diversity of homotypic output neurons in vertebrate systems serve to enhance the capacity of cortical targets to perform combinatorial integration.

With respect to developmental mechanisms, this cortical reorganization presents a puzzle. In other sensory systems, neighbouring projection neurons send axons to adjacent target regions on the basis of similarities in molecular cues that derive from their spatial location and/or their patterns of neural activity. In the olfactory system, homotypic MT cells reside in similar locations in the OB and display highly correlated sensory evoked activity^27,28,29. Yet, we observe that homotypic MT neurons are highly diverse in their cortical projections. This observation is hard to reconcile with determined patterning mechanisms based on activity or molecular cues. Yet the alternative, that the selection of synaptic targets is stochastic, would predict that individuals possess different cortical representations of the same odour. Our method for mapping axonal projections of defined sets of olfactory neurons will enable more mechanistic studies of olfactory neural circuit formation and function, and may be useful for studies of other neural circuits that involve long-range connectivity.

Methods Summary

Recombinant Sindbis viruses were generated with standard cloning techniques. Viruses were injected into GFP-positive EG glomeruli using a fluorescence microscope (SZX12, Olympus). After 48 h, brains were sectioned coronally and all sections were imaged (Olympus FV500 laser scanning confocal microscope). Neuronal reconstructions were generated by manually tracing labelled axons in Neurolucida (v9.13, MicroBrightfield). The reference brain, generated from an age- and genotype-matched animal, was imaged and olfactory regions were annotated using the Allen Reference Atlas and the contour map function in Neurolucida. The Whole Brain Catalog (version 0.7.7.8, http://wholebraincatalog.org) software platform permitted manipulation of traced neurons in three-dimensions for alignments to landmarks including EG glomeruli and identifiable clusters of nuclei in the AON, PC and OT.

Online Methods

Viral constructs

All Sindbis virus constructs were generated using a modified pSinRep5 (Invitrogen) vector containing two subgenomic promoters (pSin2gene-EGFP, a gift from G. Patrick). pSIN-G and pSIN-Y were generated by cloning the coding sequence of either EGFP (Clontech) or tDTomato³⁰ into the Xba- and Sph-cut sites of the pSIN2gene-EGFP vector backbone. To generate pSIN-R, tRFP³¹ and tRFP-T³⁰ were inserted into the vector using Xba/Sph- and BssHII/Mlu-cut sites in a pSIN-G vector that was modified to contain these sites by addition of a polylinker.

Production of Sindbis virus

For the production of Sindbis virus tracers, recombinant RNA and helper DH6 RNA were transcribed using the SP6 mMessage mMachine Kit (Ambion), and electroporated into BHK cells using a BTX ECM 600 at 220 V, 129 Ω and 1,050 μF. Virus was collected after 16–18 h and concentrated by centrifugation (20–30 min, 3,220g, 4 °C) through 7 ml/9K Pierce Concentrators (Thermo Scientific). The concentrated virus was stored in aliquots at −80 °C until further use.

Mouse strains

ORT mice were generated by crossing two previously published mouse strains with an additional strain. In this strain, an expression cassette containing an internal ribosome entry site (IRES) preceding a reporter gene (either EGFP or TauLacZ) was inserted into an odorant receptor gene immediately following the stop codon. The three lines were bred to homozygosity for each gene. The mouse lines are P2-IRES-TauLacZ³², OR-174-IRES-GFP (T.C., unpublished) and MOR28-IRES-EGFP³³.

In vivo injection of Sindbis virus tracers

All experiments were conducted on 3-week-old mice with the exception of a 6-week-old old litter used for comparison and assembly of the initial neuronal reconstruction. Mice were anesthetized with isoflurane (3% in 100% O₂) and placed in a stereotaxic apparatus. The skull was exposed and GFP-labelled glomeruli were visualized through the bone using a fluorescence stereo microscope (Olympus SZX12). A small open-skull cranial window was created above the glomeruli and viral tracers were injected directly into the targeted glomerulus using a glass micropipette (∼10 µm tip diameter) attached to a customized manipulator (Narishige) and a high-pressure microinjector (Picospritzer). The virus tracer mix consisted of concentrated virus (SIN-G = 4 × 10³ infectious units per microlitre (μl), SIN-Y = 3 × 10³ infectious units per μl) in BHK media tinted with 0.1% Fast Green dye solution to aid visualization of injection. For double-labelling experiments, a mixed preparation of SIN-G, SIN-Y and/or SIN-G was used. Pressure and injector settings were optimized to deliver nanovolumes of virus mix which consistently labelled between 1 and 10 neurons on average per sample (30 p.s.i., 4 ms, 4–8 pulses). Mice were euthanized 48 h after injection. All procedures were performed in accordance with the guidelines and standards of the Scripps Research Institutes’ Animal Care and Use Committee.

Tissue preparation

Injected animals were euthanized with isoflurane and perfused intracardially with PBS and 4% PFA. Brains with intact olfactory bulbs were isolated and post-fixed overnight at 4 °C. The olfactory bulb was sectioned along the coronal plane at 80 µm using a vibrating microtome (Leica VT1000S) and scanned for the presence of labelled MT neurons through an upright fluorescence microscope (Olympus BX60). MT neurons were identified by the laminar location of their soma and the presence of a primary dendrite innervating a single glomerulus. Cortices of selected samples with brightly labelled neurons were sectioned at 50 µm. Sections were labelled with TOTO-3 (Invitrogen, Molecular Probes) nuclear stain to delineate cortical areas and mounted using Gelvatol. The sequential order of sections was maintained throughout the entire procedure.

For the reference atlas, a 3-week-old male was perfused as above and the brain was post-fixed overnight at 4 °C. The brain was sectioned at 100 μm, stained and mounted as described above. The right hemisphere was used for imaging and mapping.

Imaging and image processing

Samples with labelled neurons were imaged sequentially at ×20 (PlanAPO oil, 0.8 NA) with a step size of 1.2 μm or 1.8 μm using an Olympus FV500 confocal microscope equipped with 488/543/633 nm excitation lasers and 505–525/560-600/660 nm emission filters. To aid with alignment of neuron traces, ×4 (UPlanFl 0.13 NA) images of the entire brain and ×10 (UplanAPO oil, 0.4 NA) images of the bulb were also acquired. Metamorph and Fluoview software were used to obtain maximum intensity projections of confocal stacks.

Mosaic imaging of the reference brain was carried out at ×10 (UPlanSApo NA 0.4) with a step size of 50 µm using an Olympus FV1000. The reference brain images per tissue section were tiled and collapsed using Image J (NIH) plugin MEMontage (NCMIR).

In Adobe Photoshop, maximum intensity projections of sample images were digitally enhanced using levels, curves, brightness and contrast primarily for display purposes. The red, yellow and green axons in Fig. 1f, g, Supplementary Fig. 2a, b and Supplementary Fig. 4a were selectively enhanced to improve visualization in three-colour printed version of this manuscript in order to accurately represent the data that is apparent in the individual channel representations of each stack. The raw images for Supplementary Fig. 2a, b separated by channel and minimally enhanced using levels are provided for comparison in Supplementary Fig. 2c, d.

Construction of 3-week-old mouse Reference Brain Atlas

Tiled maximum intensity projections of the reference brain sections were imported into Neurolucida software (Version 9.13, MicroBrightfield) and aligned using distinct nuclear layers and section outlines. Only rigid body transformations were used to sequentially align the images along the antero-posterior axis with a fixed z-distance of 100 µm between each section. The Allen Reference Atlas was used to annotate the different olfactory cortical regions (MOB, LOT, AON, PC, dTT, vTT, OT) and outlines of each region were traced using the contour map function in Neurolucida as per the manufacturer’s instructions. These contours were exported from Neurolucida in their XML format and converted into triangle meshes, one per brain region, using a MicroBrightfield converter developed at the National Center for Microscopy and Imaging Research (NCMIR), available online at http://sourceforge.net/projects/ccdb-support/. The Whole Brain Catalog (development version 0.7.7.8, http://wholebraincatalog.org) software platform was used to visualize and manipulate the reference brain in three-dimensional space.

Neuronal reconstruction and three-dimensional alignment

Unmodified confocal images of single or double neuron samples were imported into Neurolucida and aligned using the main axon and distant collaterals. Axon tracing was carried out as per the manufacturer’s instructions. To visualize each neuronal trace in the context of the olfactory cortex, two different techniques were used. First, for a small subset of the neurons, ×4 images of the corresponding brain sections were used to contour map cortical regions alongside the trace to create a complete three-dimensional reconstruction of the neuron within the source brain. Second, a less precise but more high-throughput technique was employed to map all the traces in the context of the three-dimensional reference brain. Comparable fiducial markers such as labelled glomeruli and distinct nuclei in the AON, PC and OT (Allen reference brain slides 21, 25, 31, 34) were manually identified for each sample. Tiled ×10 maximum intensity projections of the entire brain section containing these markers were obtained for each neuron sample. In Neurolucida, trimmed (main axons with collaterals spanning at least 50 μm from the node) or complete representations of each neuron trace along with its set of image sections were exported in XML format. The MicroBrightfield converter described above with an adaptation provided by the Whole Brain Project team (http://code.google.com/p/wholebrain/source/browse/#svn) allowed for the conversion of XML files to NeuroML format (http://neuroml.org). This code also allowed the conversion of the image metadata in the Neurolucida XML into the Whole Brain Catalog format for representing images in three-dimensional space. Further, the development version 0.7.7.8 of the Whole Brain catalogue software platform was augmented to allow the use of handles to move groups of two-dimensional images and three-dimensional neuronal morphologies (the ‘data sets’) as a unit. This functionality gave us the ability to align multiple neuronal reconstructions in the context of the reference brain meshes and move them independently with simple rotation and translation operations. The two-dimensional images with fiducial markers were used to align the data sets with the reference brain boundaries. The AP axis was used as the primary axis for alignment and the traces were rotated to achieve optimal fit. This ensured that the three-dimensional traces would be approximately in place relative to one another, and the reference brain, to facilitate the analysis of overall targeting specificities and extent of overlap between individual neurons.

Data collection and statistical analysis

Sector and proximity analysis: two independent assays were used to analyse the multicolour multi-neuron data set. For the sector analysis, maximum intensity projections of four analogous regions spanning approximately 400 µm in the AON and PC from five different mice were aligned flush against the top left corner in Photoshop. Using the grid tool, 128-pixel square sectors were marked on the image. The sector size was chosen such that axonal segments on average traversed over more than one sector. Each sector with main axons within it was classified as unmixed (axons of one colour within) or mixed (axons of multiple colours within). For samples in which SIN-G and SIN-Y had been injected into separate locations (S1, S2, S4) or co-injected (CO1), this resulted in bins of unmixed green sectors and yellow sectors, or mixed yellow-green sectors. In S3/CO3 samples, a mixture of SIN-G and SIN-Y was injected in a distinct region from SIN-R. Mixing of green and yellow fibres was reported as CO3. Mixing or segregation of either green or yellow with red was reported at S3. This allowed us to compare the relative positions of segregated and co-localized inputs in the same animal across multiple regions. In the sector analysis for S3/CO3, sectors were classified as containing only green or yellow axons, only red axons, or mixed (having green or yellow and red axons). In addition, as an internal control SEG3/MIX3 was further analysed by counting unmixed sectors containing only green axons or only yellow axons and with mixed being defined as containing green and yellow axons (ignoring red axons). Ratios of mixed/total (‘mixing index’) and unmixed/total sectors were plotted to normalize for the number of sectors covered by each type of axon. To determine whether the topography in the bulb is preserved in the AONpE and the PC, two-tailed Students t-test was used to compare the extent of mixing in separate versus co-injected samples.

In the proximity analysis method, the main axon of each green, yellow and red (for S3/CO3) neuron was reduced to a single point on the maximum intensity projection, with attached coordinates. The single point corresponded to the approximate midpoint of each axon segment, which was selected by manual measurement of the axon segments in collapsed image sections. The distance between each axon and every other axon in the section was computed using a customized Pdist function in Matlab. To determine whether the axons nearest to each other tended to be of similar or different origin (or colour) we calculated the maximum distance to any axon (providing an internal control that normalized variations in region size distribution of injections and the position of the axon in the region). Next, for each axon we determined how many neighbouring axons were of the same colour (as a percentage of total axons). This normalized for differences in the number of neighbours of different neurons.

Next, we identified the axons that were close (defined at 15% of the maximum distance). The relative percentage of same colour/ different colour was computed and if this exceeded 80%, this axon was called segregated. Other axons were called mixed. The cumulative number of segregated versus mixed axons across multiple sections from the AON and PC were compiled for five animals, with MIX3 and SEG3 being green versus yellow, and green/yellow versus red as before. To assess the robustness of this analysis, we performed 12 different computational analyses of these regions, varying the values for proximity (from 12–18%) and for segregation (from 70–80%). All analyses gave similar results. We present a representative yet stringent data set. Significance was assessed by two-tailed Student’s t-test.

Single neuron analysis: neurons selected for this tracing study were distinguished from other cells in the olfactory bulb based on morphological features such as shape of cell soma, laminar location of cell body, primary dendritic target and the absence or presence of axon collaterals within the bulb. For each neuron labelled in the MT cell layer we partially reconstructed its dendritic and axonal projections in the olfactory bulb. Cells with a slightly displaced tubular soma and distinct intrabulbar axon collaterals that projected to the mirror symmetric glomerulus were classified as internal tufted neurons. Cells with distinct pyramidal soma and no bulbar collaterals were categorized as mitral type I neurons. Cells in the mitral cell layer with intermediate morphologies and varying degrees of intrabulbar axon collaterals were classified as mitral type II neurons. A small number of cells were binned as unclassified due to ambiguities in assigning cell type identity. These cells were not included in cell type specific analyses.

For the single neuron branch analysis and degree of overlap, individual neurons were aligned in the context of the reference brain as described previously. The number of primary branches extended by each neuron in the AON pE, AON pP, and aPC was recorded in a blind format. Bidirectional branches were counted as being both dorsal and ventral. Differences in branching within the three brain regions were assessed using Student’s t-test (two groups) with a Bonferroni correction for multiple comparisons (n = 6 for branch number). The extent of overlap between neurons was analysed first in neuron pairs from the same animal and then extended to a pair-wise comparison of multiple neurons from different animals. For each pair the numbers of overlapping primary branches along the dorsal and ventral axes in each of the three cortical regions were counted. Three degrees of stringency (relaxed, moderate and strict) were used to define the extent of overlap between neurons based on the necessity for additional warping or transformation to accommodate a better fit. Data reported did not differ significantly between the moderate and relaxed counts; however, the overlap in the AON pP lost significance under the strict analysis. Each set of comparisons was analysed by Student’s t-test with a Bonferroni correction for multiple comparisons (n = 3 for percent overlap). Significance was reported for the adjusted number based on a pre-correction significance of P < = 0.05. We also performed a Mann–Whitney U test for ranked data exhibiting non-normal distribution, which confirmed the significance of the effects reported.

Terminal and node proximity analyses: to compare the distribution of nodes and termini of neurons with respect to one another, we performed a nearest neighbour analysis of axonal arbors. To accomplish this we extended the Whole Brain Catalog client libraries written in Java to be accessible via the Python programming language by means of the Jython project (http://jython.org). We made use of the WBC’s integration with the Java Universal Network/Graph framework (http://jung.sourceforge.net) to convert the NeuroML structure of the neurons into a tree data structure that enabled graph operations such as identifying all endpoints or nodes efficiently. Once the correct set of points were identified, their x, y and z coordinates were read into a ‘point matrix’ that had three columns (one per coordinate) and n rows, where n was equal to the number of points to be analysed. We then used the JNumerics package via Jython to compare the ‘point matrix’ of different neurons to each other. In performing the nearest neighbour analysis, we made pairwise comparisons. For each row (representing a point) in point matrix A, we calculated the Euclidean distance to every other row in point matrix B. This resulted in an n by m ‘distance matrix’ (n = no. of rows in A, m = no. of rows in B). For each row in the ‘distance matrix’ we sorted the distances and chose the smallest value, which represents the distance to the nearest neighbour for each point. We then used the resulting ‘nearest-neighbour distance vector’ to plot a 10-bin histogram for each pair-wise comparison in our set of seven fully traced neurons. We used the Python NumPy and MatPlotLib libraries to carry out the plotting. The resulting scripts have been made available as part of the International Neuroinformatics Coordinating Facilities Object Model for Neuroinformatics (http://incf-omni.googlecode.com).

The validity of this assay was tested using control pairs of artificially aligned neurons with predictable outcomes. The distribution plots generated by Matlab were then characterized into two distinct classes based on the location of the peak bin. Neuron pairs with a skew value of greater than 1 (peak bin towards minima) were assigned to class I and those with a skew value of less than 1 (peak bin towards maxima) were assigned to class II.