A rapid workflow for neuron counting in combined light sheet microscopy and magnetic resonance histology

Abstract

Information on regional variation in cell numbers and densities in the CNS provides critical insight into structure, function, and the progression of CNS diseases. However, variability can be real or a consequence of methods that do not account for technical biases, including morphologic deformations, errors in the application of cell type labels and boundaries of regions, errors of counting rules and sampling sites. We address these issues in a mouse model by introducing a workflow that consists of the following steps: 1. Magnetic resonance histology (MRH) to establish the size, shape, and regional morphology of the mouse brain in situ. 2. Light-sheet microscopy (LSM) to selectively label neurons or other cells in the entire brain without sectioning artifacts. 3. Register LSM volumes to MRH volumes to correct for dissection errors and both global and regional deformations. 4. Implement stereological protocols for automated sampling and counting of cells in 3D LSM volumes. This workflow can analyze the cell densities of one brain region in less than 1 min and is highly replicable in cortical and subcortical gray matter regions and structures throughout the brain. This method demonstrates the advantage of not requiring an extensive amount of training data, achieving a F1 score of approximately 0.9 with just 20 training nuclei. We report deformation-corrected neuron (NeuN) counts and neuronal density in 13 representative regions in 5 C57BL/6J cases and 2 BXD strains. The data represent the variability among specimens for the same brain region and across regions within the specimen. Neuronal densities estimated with our workflow are within the range of values in previous classical stereological studies. We demonstrate the application of our workflow to a mouse model of aging. This workflow improves the accuracy of neuron counting and the assessment of neuronal density on a region-by-region basis, with broad applications for studies of how genetics, environment, and development across the lifespan impact cell numbers in the CNS.

Keywords: light sheet microscopy; mouse brain; neurologic image analysis; neuron counting method; neuron density.

Figure 1

Overview of the workflow for assessing the density of neurons. (A) The mouse brain is imaged using two modalities: MRH imaging while the brain is in the skull, followed by LSM after the brain is removed from the skull and subjected to tissue clearing. (B) The LSM data are pre-processed by registering to MRH correcting the deformation in brain morphology. (C) The automated workflow locates the region with the label from r1CCFv3 and generates random subvolumes within that region to sample, applying the design-based principles of optical fractionation. Neurons in each subvolume are identified via a random forest algorithm followed by 3D watershed and volume filters and counted.

Figure 2

Effect of morphology correction on light-sheet datasets. (A) A single slice from a whole-brain LSM before correction. (B) The same slice as in (A) after correction using MRH. Magnified axial views of auditory areas and hippocampus before (C) and after (D) correction, and coronal views of the anterior temporal and posterior diencephalic region (showing the amygdaloid complex) before (E) and after (F) correction. The specimen is the same on both sides, and the colormap is used to illustrate the contrast between before and after correction.

Figure 3

Comparison of similar levels from the ABA (A,C,E) and a single specimen imaged by MRH. (B,D,F) The colormaps are different because the ABA atlas is shown with the full complement of 461 (CCFv3) labels and the MRH is shown with the reduced (r1CCFv3) labels in which some ROI have been combined.

Figure 4

Comparison of neuron density between the machine workflow (A) and manual counting (B). Note that both the labeling will go through watershed illustrated in (C) to ensure the connected surfaces are separated. Standard deviation is shown in (D), representing statistical variations across five different 90-day C57BL/6J specimens.

Figure 5

Demonstration of the variation of neuron density (A) and neuron number (B) across brain regions, as well as variations across specimens. Error bars are standard deviations of the means. The dots indicate neuron counts from individual specimens. Region: LGd: Dorsal lateral geniculate nucleus, AUD: auditory area, RSP: retrosplenial area, Orb: orbital area, SUBR: subiculum, VII: facial motor cortex, PSV: trigeminal, ENT: entorhinal area, CA1: field CA1, CA3: field CA3, VISp: primary cortex, TH: thalamus, BLA: Basolateral amygdala.

Figure 6

Comparison of the neuron density (A) and number (B) in multiple brain regions of a young and an old BXD89 mouse, demonstrating the application of the workflow to a study of the impact of aging on neuronal populations in a different strain. The error bars in this figure represent the standard deviation of neuron distribution within each region, obtained from subvolumes.