To deconstruct the architecture and function of brain circuits, it is necessary to generate maps of neuronal connectivity and activity on a whole-brain scale. New methods now enable large-scale mapping of the mouse brain at cellular and subcellular resolution. We developed a framework to automatically annotate, analyze, visualize and easily share whole-brain data at cellular resolution, based on a scale-invariant, interactive mouse brain atlas. This framework enables connectivity and mapping projects in individual laboratories and across imaging platforms, as well as multiplexed quantitative information on the molecular identity of single neurons. As a proof of concept, we generated a comparative connectivity map of five major neuron types in the corticostriatal circuit, as well as an activity-based map to identify hubs mediating the behavioral effects of cocaine. Thus, this computational framework provides the necessary tools to generate brain maps that integrate data from connectivity, neuron identity and function.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
D.F. thanks J. Bernardi and J.H. Lee for fruitful discussions about the software or the paper. K.M. acknowledges financial support from the Swedish Research Council (VR 2012-02049), from the Karolinska Institutet (KID-funding supporting D.F., O.T., A.M.), from the Strategic Neuroscience Area at Karolinska Institutet (StratNeuro) for rabies virus production and from the Swedish Brain Foundation (Hjärnfonden). Additional financial support for the project was from a National Institute of Mental Health grant (MH109795 to G.R., K.M. and C.A.M.) and a National Institute on Drug Abuse grant (DA0036376, to C.A.M.).
Integrated Supplementary Information
(a) An open-source brain block designed from our 3D atlas printed which can be directly ordered from (http://www.shapeways.com/shops/wholebrain) with all profit going to further development of the atlas. Alternatively, the brain block can be downloaded (https://github.com/tractatus/brain-blocks) modified for any needs and 3D printed by the user itself. (b) The brain block is designed to be used with a standard and can cut coronal sections anchored in the atlas with 1 mm apart, ranging from stereotactic anterior-posterior coordinates +5 mm to −11 mm relative to bregma (white arrow). (c) Top view. (d) Close-up, the small circular bevel marks bregma 0 mm (black and red arrow). (e) Specimen holder for the Leica LMD system scanning stage accommodating 3 slides, with only the middle one loaded with one membrane slide mounted with a single brains section. (f) Entire membrane microscope slide imaged with bright-field at 10× optical magnification and stitched by Leica LMD, this overview image is exported as raw TIF and used as input to WholeBrain registration. (g) Close up on brain section after registration to matched coordinate (+0.2 mm from bregma) (h). The user can select any region desired to laser capture with the get.region() command, in this case selecting SSp (left hemisphere), MOp2/3 (right hemisphere), SSp-bfd5 (right hemisphere), ADP, SI, MS i.e. get.regions(acronym=c("SSp","MOs2/3","SSp-bfd5","ADP","SI","MS"), right.hemisphere=c(0,1,1,2,2,2) registration=regi), where regi is the registration output. (i) Registration result superimposed on the microscope slide. (j) with regions to be laser captured.
a) Description of manual and automatic work flow as well as output from an automated WholeBrain pipeline. (b) Schematic of a wavelet filter bank in two-dimensions as well as the wavelet lifting scheme. (c-e) Maximum intensity projection of confocal images of a single EGFP labeled medium spy neuron in striatum imaged with three different objectives; (c) 20×, (d) 63×, and (e) 100×, red point and diamond indicate center of mass and white outlines are manually drawn soma perimeters at different z-plane. (f) Estimated soma area as a function of imaging objectives and pixel resolution, measurements of soma area from drawn perimeters obtained from (c-e), colored squares indicate estimated soma area for each objective. Inset: the average fractal dimension, D, of the soma in (c-e) as a function of microscope objective and pixel resolution. As the optical resolution approaches the pixel density the fractal dimension approaches one (Manhattan geometry). (g) A single image is decomposed into a sequence of nested subspaces. Low-resolution images are contained in high-resolution images. (h) As the image is filtered through the wavelet bank and the scale period approaches that of the lower resolution objectives the measured soma area will approach the estimate obtained from that objective, non-filled circles indicate area estimates form individually drawn contours. (i) Soma area of the same neuron estimated from images taken either with 10x (orange) or 63x (blue) objectives can be directly compared by transforming both images into the wavelet detail coefficients where the sampling period is close to the spatial scale that defines typically the soma size (10–20 μm). In this case for the 63x objective this corresponds to d6 where the sampling period is 17.85 μm whereas for the 10x objective this corresponds to d3 where the sampling period is 14.06 μm. Individual dots display areas estimated from individual contours drawn. Probability densities from these data points are obtained by kernel density estimation.
Supplementary Figure 3 Generation of equally spaced correspondence points along a contour by principal components analysis
a) The contour (black thin line) is generated by segmentation of a binary mask following1. The first (thick black lines) and the second (dashed black line) principal components (PC) are drawn intersecting at the centroid of the contour. The two intersections for each PC with the contour is computed (black points) and are added to the set of correspondence points, p. For each two pair of points (p i and p i+1) generate an equal distance midpoint q (white circle) between these two points. (b) At the next level start from the midpoints q and draw a line perpendicular to the midline between p i and p i+1 where this line intersects the contour add that point to the set of correspondence points. Iterate the procedure for as many levels as the user wants, e.g. (c) three levels 16 points, (d) four levels 32 points. This scheme follows2. 1.Suzuki, S. & be, K. Topological structural analysis of digitized binary images by border following. Computer Vision, Graphics, and Image Processing 30, 32–46 (1985). 2.Mitra, J. et al. A Thin-Plate Spline Based Multimodal Prostate Registration with Optimal Correspondences. 7–11 (2010). doi:10.1109/SITIS.2010.12.
(a) Regions of interests for measuring registration error before and after manual landmark correction. Regions were selected based on their difference in proximity to the center of the tissue section. Scale bar 1 mm. (b) Root Mean Square Error (RMSE) in micrometers as a function of brain nuclei in four different brains and 64 landmarks along the contour of each nuclei. (c) Mean RMSE as a function of average distance from center of tissue section.
(a-f) Representative sections (n = 6) from a total of 53 imaged and quantified sections representing a single mouse brain stained for FOXP2. Left: Original fluorescent image with atlas fit superimposed (purple). Right: segmented FOXP2+ nuclei transformed into stereotactic reference atlas where individual nuclei are color coded according to brain region. (g-l) Representative section (n = 6) from a total of 29 sagittal mouse brain sections also stained for FOXP2. Top: Original fluorescent image with atlas fit superimposed (purple). Bottom: segmented FOXP2+ nuclei transformed into stereotactic reference atlas where individual nuclei are color coded according to brain region. (m) 3D reconstruction of 53 individual FOXP2stained coronal sections. (n) 3D reconstruction of FOXP2 segmented nuclei obtained in the sagittal plane from a single hemisphere and then mirrored to obtain a whole-brain representation at almost half the imaging, sectioning and mounting time. (o) A more complete 3D reconstruction can be achieved by combining (m) and (n). Scale bars: 1 mm.
Absolute cell counts of FOXP2+ neurons in sagittal sectioning plane (blue) versus coronal sectioning plane (red). Ordered from top to bottom according to absolute difference in sagittal versus coronal sectioning plane per cortical region. Right panel: measurement error defined as square root error in proportion of total number of segmented FoxP2+ cells.
(a) Absolute cell counts of FOXP2+ neurons in sagittal sectioning plane (blue) versus coronal sectioning plane (red) in striatal/pallidal regions. Ordered from top to bottom according to absolute difference in sagittal versus coronal sectioning plane per cortical region. (b) Absolute cell counts of FOXP2+ neurons in sagittal sectioning plane (blue) versus coronal sectioning plane (red) in thalamic/hypothalamic regions. Ordered from top to bottom according to absolute difference in sagittal versus coronal sectioning plane per cortical region. Right panels: measurement error defined as square root error in proportion of total number of segmented FOXP2+ cells.
(a) Projection of coronal whole mouse brain reconstructed in Supplementary Figure 7 onto horizontal plane (orange horizontal line in inset with side view of brain). All FOXP2+ cells at +/- 100 μm dorso/ventral from horizontal plane were projected onto the plane. Right panel shows the kernel density estimation of segmented FOXP2+ cells. (b) Same as in (a) but for sagittal plane. (c) To further demonstrate the limitations in resolution using sectioned tissue we projected the coronal and sagittal reconstructed brains onto three sections at 45° relative to the coronal and sagittal plane. (d-e) Atlas rendering of the 45° sectioning plane. (g-i) Fit of (d-f) onto FoxP2 stained brain cut at the 45° angle plane. (j) Coronal reconstructed brain superimposed on the 45° angle plane. (k) Sagittal reconstructed brain superimposed on the 45° angle plane. (l) Comparison of sagittal and coronal reconstruction on the arbitrary 45° angle plane.
Supplementary Figure 9 Single-molecule fluorescent in situ hybridization for mapping of molecular identity and neuron types
(a) Close-up of mRNA transcripts labeled by RNAscope superimposed on cell nuclei (gray: DAPI). Green: Vesicular Glutamate transporter 2 (VGlut2), red: Prodynorphin (Pdyn), magenta: Vesicular GABA transporter (VGat). Scale bar: 10 μm. (b) Segmentation of individual cell nuclei (orange contours) together with perinuclear zone (dashed orange contours). (c) Segmentation of single fluorescently labeled mRNA transcripts (Green asterisks: VGlut2, Magenta squares: VGat, Red triangles: Pdyn). (d) Affinity propagation clustering to assign single transcripts to single cell nuclei or background (82%, 204466 transcripts, could be assigned to individual nuclei). (e) Maximum intensity projection of hypothalamus imaged by 90 field of views and 15 optical z-planes at 40x optical magnification and then registered to coronal plate from the reference atlas (−1.75 mm anterior-posterior from bregma) with registration results superimposed (orange contours). Scale bar: 500 µm. (f-h) Scatter plots of 7121 individually segmented cell nuclei assigned with at least a single transcript each plotted on log-log coordinates of number of transcripts with Pearson product-moment correlation from linear scale. To classify cells we assigned cells with a transcript count of VGlut2 higher than ten as VGlut2+ (turqoise), and VGlut2+ cells who also expressed Pdyn at transcript counts higher than ten as VGlut2+/Pdyn+ cells (coral), and cells belonging to neither population as VGat+ (purple). (i) Pie chart of the three cell populations defined in (f). (j) 8703 cells from four different populations: VGlut2+ (turqoise), VGlut2+/Pdyn+ (coral), VGat+ (purple), and cell nuclei without any transcripts (dark gray) superimposed on the fit to the reference atlas. (k) Cell counts as a function of brain region. (l) Total number of transcripts as a function of brain region. Brain regions: ARH: Arcuate hypothalamic nucleus; DMHa: Dorsomedial nucleus of the hypothalamus, anterior part; DMHp: Dorsomedial nucleus of the hypothalamus, posterior part; DMHv: Dorsomedial nucleus of the hypothalamus, ventral part; HY: Hypothalamus; LHA: Lateral hypothalamic area; ME: Median eminence; PH: Posterior hypothalamic nucleus; PVi: Periventricular hypothalamic nucleus, intermediate part; RE: Nucleus of reunions; TH: Thlamus; TU: Tuberal nucleus; VMHc: Ventromedial hypothalamic nucleus, central part; VMHdm: Ventromedial hypothalamic nucleus, dorsomedial part; VMHvl: Ventromedial hypothalamic nucleus, ventrolateral part; ZI: Zona incerta.
(a) Original epifluorescent micrograph of c-fos and DAPI. (b) Wavelet filtered at the second detail coefficient scale. (c) The energy from the structure tensor. (d) Shannon entropy based binary thresholding followed by watersheding, inset shows results before and after watersheding. (e) Subtraction of the binary image from the original epifluorescence. Scale bars: 200 μm. (f) Segmentation result for DAPI with each nuclei color-coded according to mean pixel intensity, cube helix color scheme. (g) Segmentation result for c-fos with each nuclei color-coded according to mean pixel intensity. (h) Cell counts from f and g. (i) c-fos to DAPI nuclei ratio, note that it’s approximately 0.5 which is roughly the ratio of neurons to glia in the brain3. 3.Azevedo, F. A. C. et al. Equal numbers of neuronal and nonneuronal cells make the human brain an isometrically scaled-up primate brain. The Journal of Comparative Neurology 513, 532–541 (2009).
Posterior estimates of average cell count with regions sorted in descending order with background color indicating anatomical parent region. Thin error bar represents the 95% credible interval (C.I.) and thick lines represents 80% C.I. (n = 4 mice).
Supplementary Figures 1–11
Comparison of contemporary approaches to fluorescent whole mouse brain mapping.
Green cells are presynaptic partners to D2-expressing striatal neurons labeled using monosynaptic cell-type specific rabies-EGFP virus in D2-cre mouse.
Initially labeled neurons are color coded according to brain region, later (00:18) neurons are color coded according to transgenic mouse line (dark blue: D1-Cre, red: D2-Cre, light blue: Chat-Cre, yellow: Camk2a-Cre, green: Gad2-Cre).