Automated image analysis is becoming a central
tool for quantifying the effect of candidate drugs on
cells. A number of commercial high-content cellular
imaging systems include software that can characterize
neuronal projections (i.e., axons and dendrites).
The faster systems, capable of running at
primary screening speeds, tend to measure only the
simplest feature describing these neurite out-growths—their total length. The slower systems,
suitable for secondary screening, are given a greater
amount of time to run more sophisticated algorithms.
Some of these can measure an additional
feature describing neurite complexity—the number
of branch points (Figure 1).
Figure 1 - a) Representative image of a neuron viewed under fluorescence
microscopy. b) Output generated by the software, detailing primary (blue),
secondary (yellow), tertiary (purple), and quaternary neurites (teal). c)
Examples of neurite trees, together with some features measured by the software,
illustrating that the total neurite length is often inadequate to map significant
differences in neurite structures.
This paper presents a powerful software tool
(CSIRO Mathematical and Information Sciences,
North Ryde, Australia) that can simultaneously
report several features to characterize the branching
topology of each neurite “tree.” The software can
discriminate subtle changes in neurite branching
structures that would otherwise go undetected. The
code is adapted for primary screening
and can also run as a standalone system
for fundamental research.
Discriminating subtle
changes
The most widespread feature describing
neurite outgrowths is their total
length. Occasionally, the number of
branch points is also reported. With
only these measures, cell morphologies
that appear distinctly different
to a human observer can end up
sharing the same set of features, and
hence be considered identical by an
automated classifier downstream.
There is thus a need for additional,
biologically relevant features to
characterize the branching structure
of neurites.
The software reports the
number of primary (or root)
neurites as well as the number
of layers associated with
these primary neurites (secondary,
tertiary, quaternary,
etc., branches). These additional
features allow users to
selectively screen for compounds
triggering different
types of neurite outgrowth behavior, as
shown in Figure 1. Results are available
either on a cell-by-cell basis or as averages
over images.
Improved robustness
toward artefacts
Figure
2 - a) and b) Nuclei stained with a specific dye
4’-6-diamidino-2-phenylindole (DAPI). The software uses novel algorithms
to exploit these nucleus
channel images to segment cells reliably, resulting in accurate per-cell
results.
The sophisticated measures in the software
are more robust toward artefacts
that commonly obscure the results of
screening experiments. For instance,
even at moderate cell plating densities,
the neurites growing from one cell can
intersect with those growing from a cell
nearby. These intersections are traditionally
mistaken for branch points. The software
correctly differentiates between
true neurite branch points and artefact
branch points, making the measurements
more robust toward variations in cell
density than the simple branch point
counts used in other systems. As a result,
dose-response curves tend to have narrower
error bars, and Z’ factors (an assay
quality statistic describing the separability
of positive and negative controls)
tend to be higher.
Conclusion
Ideally, results should be available on a per-cell basis
to correct for variations in cell plating density. A
common problem here is the presence of closely
touching neurons, which can be difficult to segment
and result in biased per-cell measurements. The software’s
code for separating touching objects is based
on a modified version of the watershed transform
(Figure 2). The software can optionally accept
nucleus channel images as input to make the cell
segmentation process even more robust.
The authors are with the Biotech Imaging Group, CSIRO
Mathematical and Information Sciences, Locked Bag 17, North
Ryde NSW 1670, Australia; tel.: +61 02 9325 3208; fax: +61
02 9325 3200; e-mail: [email protected]. The software
described here has been licensed into systems designed for secondary
drug screening: ImageXpress™ (Molecular Devices,
Sunnyvale, CA) and Pathway Bioimager™ (BD Biosciences,
Franklin Lakes, NJ). It is also used in the Opera™ system
(Evotec Technologies, Hamburg, Germany), which is capable
of primary screening speeds. Images for this article
were kindly provided by Evotec Technologies.