Documentation for NIH Image coloc
Last revised on June 15, 1999

	This set of macros was designed to perform colocalization analysis 
of confocal microscope images of immunofluorescently labeled clusters of 
synaptic vesicle protein (SVP) with a population of axons labeled with a 
neuronal tracer.  Although this particular application of the colocalization 
procedure is described in the following documentation, the method may be 
useful for performing colocalization analysis of stacks of optical sections of 
other material as well.  A justification for the use of this object-based 
procedure instead of simpler analyses as well as a discussion of the more 
theoretical aspects of the procedure can be found in Silver, M.A. and 
Stryker, M.P. "A method for measuring colocalization of presynaptic 
markers with anatomically labeled axons using double label 
immunofluorescence and confocal microscopy", submitted April, 1999. 

	This program will run on a Macintosh (PowerPC or later).  It is a 
customized version of the public domain NIH Image program version 1.61 
(developed at the U.S. National Institutes of Health and available on the 
Internet at http://rsb.info.nih.gov/nih-image/)  This document assumes a 
working knowledge of NIH Image.  If you are not familiar with NIH 
Image, documentation is available at 
http://rsb.info.nih.gov/nih-image/manual/contents.html     The Image 
Processing Handbook, by John Russ, is an excellent introduction to image 
processing techniques.  It is published by CRC Press (Boca Raton, FL).

	To start, you should have two stacks of images that correspond to 
stacks of optical sections collected on a confocal microscope.  The stacks 
should each contain at least three slices, and the image type should be TIFF.  
One stack contains images of the synaptic vesicle protein label (referred to 
as the SVP stack), and the other contains images of labeled axons (referred 
to as the axon stack).  The images should have dark signal (high pixel 
intensity values) against a light background (low pixel intensity values).  If 
your images have light signal against a dark background, simply invert 
them.  The procedure is designed to analyze only one slice in a pair of 
stacks.  This focal plane of interest is called the reference slice, and 
colocalization in three dimensions will be performed by comparing pixel 
values in the reference slice to the slice immediately above and immediately 
below this reference slice.  If you want to analyze more than one slice, the 
easiest way to do this is to run through the procedure again with the same 
pair of stacks but choose a different slice as the reference slice.

	Load the colocalization macros file.  If you then pull down the 
Special menu, you will see a set of ten additional commands.  Each of these 
ten commands can also be executed by pressing the appropriate function key 
(for example, key F1 will execute Trace Cell Bodies).  The first step is to 
remove the cell bodies, blood vessels, and other portions of the field that do 
not contain synapses from consideration.  It is only appropriate to make 
quantitative measures of synaptic density in the regions of the field that 
contain synapses, i.e., the neuropil.  The percentage of the field that 
contains cell bodies will vary from section to section, and this variability can 
be eliminated by masking the pixels located within cell bodies, blood 
vessels, etc. and excluding them from further analysis.  To accomplish this, 
open the SVP stack and select the slice that you have chosen for 
colocalization analysis (the focal plane of interest).  Execute the Trace Cell 
Bodies macro.  This will open a window containing the SVP label and 
automatically select the pencil tool.  Using the pencil tool, trace all the 
portions of the field that you wish to mask.  All traced regions must be 
completely closed loops (including regions that extend to the borders of the 
field).  If you need to use the magnifying glass tool to zoom in on part of 
the image, simply reselect the pencil tool and continue tracing.

	Next, to mask the cell bodies, execute the Remove Cell Bodies 
macro.  This will create a window that contains only the masked regions.  A 
dialog box will appear which prompts you to provide a name for saving this 
image to disk.  The intensity values for the pixels within the masked regions 
in the original image will then be set to zero, and the number of pixels 
contained within those regions will be stored by the program for future 
reference.  Now you want to threshold the image.  Thresholding allows 
separation of the signal from the background and is also a first step towards 
segmenting individual presynaptic sites from one another.  The threshold 
should be a constant percentage of pixels in the neuropil (not in the entire 
image).  For example, let's say you want to threshold the top 10% of the 
pixels and your image contains 400,000 pixels.  However, only 300,000 of 
these pixels are located in the neuropil (100,000 were located in cell bodies, 
etc. and were masked by the Remove Cell Bodies macro).  In this case, you 
will want to keep only those 30,000 pixels in the masked image that have 
the highest intensities.  When you execute the Threshold macro, it will ask 
"How many pixels to be excluded?"  The default is the number of pixels that 
you have masked during the removal of the cell bodies, blood vessels, etc.  
You will also be prompted to choose a percentage of neuropil pixels to be 
included above threshold.  If you will be quantitatively comparing the 
amount of colocalization in different pairs of images, the threshold should 
be the same for all images that are being compared.  The value of the 
threshold should be low enough so that most of the clusters of SVP label 
located in the reference slice are above threshold.  However, it should not 
be so high that many of the clusters form large contiguous regions that will 
be very time consuming to segment from each other.

	Because there are variations of signal intensity between slices in a 
stack that are often not due to biological causes (differential penetration of 
antibody, photobleaching, light scattering, etc.), it is only appropriate to 
compare absolute intensity values in adjacent focal planes if these sources of 
variability can be removed.  The macro to do this is called Match 
Histograms.  Given a reference slice, the macro reassigns the intensity 
values of the pixels in the slices above and below the reference slice without 
changing the rank ordering of the pixels.  After the macro is finished, the 
pixel intensity histograms of all the sections in the stack are identical to the 
histogram of the reference slice.  This macro can take a few minutes to run 
(depending on the image size and the speed of the computer).  You should 
save the stack of matched slices as they will be used later in the procedure.

	The only reliable method I have found to identify the portion of 
labeled axons that are located in the reference slice is to trace them manually.  
The macro called Trace Axons functions like the Trace Cell Bodies macro.  
Open the axon stack and then execute Trace Axons.  Use the pencil to trace 
the portion of each axon branch that is located within the focal plane of 
interest (corresponding to the reference slice in the SVP stack).  To exclude 
portions of the axons that are located above and below the focal plane of 
interest, move up and down through the slices of the stack (using the < and 
> keys) and only trace that part of the axon branch that has the most signal 
in the focal plane of interest.  All traced axon segments need to be 
completely closed loops.  Now execute Fill Axons.  This macro will 
produce a single binary image in which the traced segments of axon 
branches are black (pixel value of 255) and all other pixels in the image are 
white (pixel value of 0).  Save this image to disk.

	Although the thresholding procedure should effectively separate 
many of the presynaptic sites from one another, there will still be some 
neighboring clusters of synaptic vesicle protein label which are above 
threshold and form one contiguous object.  These synapses need to 
segmented from each other so that colocalization analysis can be done on 
them as distinct objects.  Since the majority of SVP label will be in regions 
of the reference slice that do not contain any labeled axon segments (and 
therefore have no chance of being colocalized with labeled axons), only 
those clusters of SVP label that are partially overlapping with labeled axon 
segments in the reference slice need to be segmented from one another.  
Open the SVP stack that was the output of the Match Histograms macro.  
Select the reference slice (using the < and > keys) and then execute Seed 
Synapses.  The macro will prompt you to find the binarized axon image on 
disk.  It will then create an image in which the SVP label is unchanged, but 
the axons are transparently overlaid on it and represented as green.  This 
allows you to segment only those presynaptic sites that have some chance of 
being colocalized with the labeled axons.  Using the pencil tool (which will 
draw in red), place a seed pixel into the center of each synapse that you 
want to test for colocalization.  Unless your images are collected at very 
high magnification, you will probably want to zoom the image so that you 
can accurately place the seed pixel.

	After all of the seed pixels have been placed, execute Expand Seeds.  
This macro will radially expand these single pixels (also known as an 
iterative dilation) until they reach the boundary defined by the thresholding 
procedure or until they collide with an expanding seed pixel from an 
adjacent synapse.  In this case, a one pixel wide boundary is drawn between 
the synapses, thereby segmenting them.  This macro may take a few 
minutes to run.  When it is finished, you will have the original post-Match 
Histograms stack, but the reference slice will only contain those expanded 
synapses that had a seed pixel placed in them.

	Now you are ready to do the colocalization.  To be counted as 
colocalized, a presynaptic site must 1) be located in the reference slice, and 
2) be located in an axon segment that has been identified by the Trace Axons 
macro.  You can think of this as colocalization in the z-axis and the xy-
plane, respectively.  To decide whether the segmented presynaptic sites are 
located in the reference slice, execute the Determine Focal Plane macro.  For 
each segmented synapse in the reference slice, the macro will compare the 
pixel intensities of the above threshold pixels in the reference slice to the 
intensities of the corresponding pixels in the slice located above the 
reference slice and then to the slice located below the reference slice.  If the 
average pixel intensity is highest in the reference slice for a given segmented 
presynaptic site, the synapse is considered to be located in the reference 
slice and it is retained.  If the average pixel intensity is higher above or 
below the reference slice, the synapse is probably not located in the 
reference slice.  In this case, the signal that is present in the reference slice is 
due to out-of-focus scatter, and the synapse is eliminated from further 
analysis.

	The next step is to see if the presynaptic sites are located inside of 
axons.  This is accomplished by the macro named Colocalize.  When this 
macro is executed, you are prompted to locate the binarized axon image on 
the disk.  You then need to input the colocalization stringency ("Minimum 
percentage colocalization?").  A value of 50 means that a given presynaptic 
site must have at least 50% of its pixels located within a binarized axon 
segment to be counted as colocalized.  A value of 100 means that every 
pixel in the presynaptic site must be contained within a binarized axon 
segment.  A parametric study of colocalization stringency showed that a 
colocalization stringency of 100% allowed for the optimal combination of 
elimination of false positive colocalizations and retention of actual 
colocalizations.  Another way to think of this is that the signal-to-noise ratio 
was highest for a colocalization stringency of 100%.  These data are 
presented in the manuscript by Silver and Stryker cited at the beginning of 
this document.

	You will also be prompted to set a minimum size criterion.  This can 
serve to eliminate spurious colocalizations due to pixel noise.  The criterion 
will depend on the pattern of SVP labeling in your tissue as well as the 
magnification used to collect the image.  In general, your size criterion 
should approximate the size of the smallest presynaptic sites.  Only those 
presynaptic sites which have a percentage of pixels within a binarized axon 
segment that is above threshold and which meet the size criterion will be 
retained and considered to be colocalized.  The final output of the procedure 
is a single image called "colocalized particles" that contains the colocalized 
segmented thresholded presynaptic sites with their original pixel intensity 
values and all other pixel intensities set to zero (white).  How you quantify 
the properties of your colocalized synapses is up to you.  The Analyze 
Particles command in NIH Image allows for many quantitative measurements 
to be made on segmented objects.  You may want to include the area, mean 
intensity, and x and y location of the colocalized synapses.  In the final 
analysis, you will probably want to include normalization factors for 1) the 
amount of axon labeling in the field, and 2) the overall amount of SVP signal 
in the field.  One example of how to do this is in the Silver and Stryker 
manuscript mentioned at the beginning of this document.

	If you have any questions or encounter any bugs while running this 
program, I would appreciate hearing about them at silver@phy.ucsf.edu

Michael Silver
Neuroscience Graduate Program
University of California, San Francisco