Quantifying How Staining Methods Bias Measurements of Neuron Morphologies

Abstract

The process through which neurons are labeled is a key methodological choice in measuring neuron morphology. However, little is known about how this choice may bias measurements. To quantify this bias we compare the extracted morphology of neurons collected from the same rodent species, experimental condition, gender distribution, age distribution, brain region and putative cell type, but obtained with 19 distinct staining methods. We found strong biases on measured features of morphology. These were largest in features related to the coverage of the dendritic tree (e.g., the total dendritic tree length). Understanding measurement biases is crucial for interpreting morphological data.

A Systematic Evaluation of Interneuron Morphology Representations for Cell Type Discrimination

“…Among the density maps, the Z density maps did not perform differently from XZ density maps ( δ = 0.006 ± 0.02, z = 0.37, p = 0.71), while YZ density maps performed much worse in the three V1 data sets ( δ = 0.14 ± 0.05, z = 2.84, p = 0.005, average across V1 pairs only) but very similar in the bipolar data set ( δ = 0.01 ± 0.02, z = 0.77, p = 0.44). Indeed, the y direction is mostly meaningless in the V1 data as the slices are flattened during the biocytin staining process (Farhoodi et al 2019 ).…”

“…Next, tissue shrinkage and staining method can affect the measured morphology (Farhoodi et al 2019 ). In our study all three data sets obtained through biocytin staining (V1 L2/3, V1 L4, V1 L5) showed flattening of the cortical slice ( y -direction) which made XY density maps perform worse in comparison to other projections.…”

The natverse, a versatile toolbox for combining and analysing neuroanatomical data

“…We used neuromorphr to obtain information about 40599 mammalian neurons from NeuroMorpho.org, in the brain region 'neocortex', from 193 different laboratories. However, it is quite difficult to compare morphologies acquired from different laboratories, that have used different imaging pipelines, neuron reconstruction tools and staining protocols (Farhoodi et al, 2019). We can use neuromorphr to access neuron metadata on NeuroMorpho.org, and choose only those neurons from the same laboratory (Bob Jacobs'), that have been obtained using a Golgi stain and the reconstruction software NeuroLucida, and that are classed as a principal cell (Anderson et al, 2010(Anderson et al, , 2009Jacobs et al, 2018Jacobs et al, , 2016Jacobs et al, , 2015Jacobs et al, , 2011Jacobs et al, , 2001Jacobs et al, , 1997Reyes et al, 2016;Travis et al, 2005), giving us 3174 neurons.…”

A systematic evaluation of interneuron morphology representations for cell type discrimination

“…The representations investigated here are purely descriptive and do not provide deeper mechanistic insight, compared e.g. to generative models of the growth process of neurons during development (van Pelt and Schierwagen, 2004;Cuntz et al, 2010;Memelli et al, 2013;Wolf et al, 2013;Fard et al, 2018;Farhoodi et al, 2019). Ideally, a mechanistically grounded feature representation would perform at least on par with density maps for cell type discrimination while yielding parameters that are more easily interpretable.…”

“…Among the density maps, the Z density map did not show a significant difference from XZ (δ = 0.006 ± 0.02, z = 0.37, p = 0.71), while YZ was much worse than XZ in the three V1 data sets (δ = 0.14±0.05, z = 2.84, p = 0.005, average across V1 pairs only) but very similar in the bipolar data set (δ = 0.01±0.02, z = 0.77, p = 0.44). Indeed, the y direction is mostly meaningless in the V1 data as the slices are flattened during the biocytin staining process (Farhoodi et al, 2019).…”