Help: Centrosome quantification

[Ignacio Fernandez-Garcia]

Hi all,
Lately I've been working analyzing the centrosome abnormalities in some experiments. One thing we're interested in is the number of centrosomes per cell (usually two in normal cells). To detect the centrosomes we use an immunoflurescence technique where we label the centrosomes in green and counterstain the nuclei with DAPI in blue. So, we obtain images like this one:

http://a.imageshack.us/img715/4948/tubulincaexample.jpg

As you can see, we have some green signals (the punctual ones) distributed within the cell cytoplasm (seen as green background) but not necessarily within the nucleus. This way, a nuclear mask with the DAPI is not enough to segment the centrosomes and quantify them as a "number-of-centrosomes-per-cell" value.
One approach that comes to my mind is: Segment both, centrosomes and nucleus, in their correspondent channel, and then associate the segmented centrosome signals with the nearest segmented nucleus by the minimum square distance between mass centers.
So, my question is, anybody can help me with that? does anybody have done it before?
Also I'll really appreciate suggestions, comments, questions, etc.
Thanks!
Cheers!

[Bertrand Vernay]

Hi Ignacio

You may ant to have a look at this publication
PLoS Biol. 2008 Sep 16;6(9):e224.
A genome-wide RNAi screen to dissect centriole duplication and centrosome
maturation in Drosophila.

Dobbelaere J, Josué F, Suijkerbuijk S, Baum B, Tapon N, Raff J.
http://www.plosbiology.org/article/info%3Adoi%2F10.1371%2Fjournal.pbio.00602
24

The authors used Cellprofiler (http://www.cellprofiler.org/index.shtml) to
score centrosomes defects. A similar protocol may work for you.

Regards

Bertrand
--
Dr Bertrand Vernay
Microscopy Senior Research Associate
Developmental Biology Unit
UCL Institute of Child Health
30 Guilford Street
London WC1N 1EH, UK

Tel :  (+44)  020 7905 2224 (direct line)
       (+44)  020 7242 9789 (ICH switchboard)

Fax:    (+44)  020 7831 4366

E-mail: [hidden email]

http://www.ich.ucl.ac.uk/services_and_facilities/lab_services/confocal_micro
scopy_core_facility/index.html
-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of
Ignacio Fernandez-Garcia
Sent: 28 July 2010 21:48
To: [hidden email]
Subject: Help: Centrosome quantification

Hi all,
Lately I've been working analyzing the centrosome abnormalities in some
experiments. One thing we're interested in is the number of centrosomes per
cell (usually two in normal cells). To detect the centrosomes we use an
immunoflurescence technique where we label the centrosomes in green and
counterstain the nuclei with DAPI in blue. So, we obtain images like this
one:

http://a.imageshack.us/img715/4948/tubulincaexample.jpg

As you can see, we have some green signals (the punctual ones) distributed
within the cell cytoplasm (seen as green background) but not necessarily
within the nucleus. This way, a nuclear mask with the DAPI is not enough to
segment the centrosomes and quantify them as a
"number-of-centrosomes-per-cell" value.
One approach that comes to my mind is: Segment both, centrosomes and
nucleus, in their correspondent channel, and then associate the segmented
centrosome signals with the nearest segmented nucleus by the minimum square
distance between mass centers.
So, my question is, anybody can help me with that? does anybody have done it
before?
Also I'll really appreciate suggestions, comments, questions, etc.
Thanks!
Cheers!

[karo03]

Hi Ignacio,
a straight forward solution, at least with your example might be:
- Segment nuclei in DAPI, cut 'connected' nuclei with watershed.
  This will deliver errors, either acceptable or to be corrected!
- Skeletonize the inverted resulting nuclei image, possibly reduced by black background (see attachment)
  Some cleaning delivers 'cell' areas allowing to count particles per cell area:
  - First measure these 'cell' areas with add to ROI Manager
  - Segment the particles
  - Update/Adjust ROIs to the intersecting particles, this will lead to ROIs per cell area of the particles
macro "ROIAdjustLabel_" {
// Given: A binary Image with components (selected)
//           roi's (ROImanager) (typically larger than the image components)
// Result: rois are updated to the intersection of the roi and the binary image
//             Label numbers of the rois are transfered to the intersecting parts
//             for further analysis
//
 setBatchMode(true);
 title = getTitle();
 run("Select None");
 run("Duplicate...", "title=tmp");
 run("Select All");
 run("Clear");
 run("Select None");
 ct=roiManager("count");
 for (i=0;i<ct;i++){
    roiManager("Select", i);
    roiManager("Fill");
    imageCalculator("AND", "tmp", title);
    run("Make Binary");
    run("Create Selection");
    roiManager("Update");
    run("Clear");
    run("Select None");
 }
 close();
 setBatchMode(false);
}

Maybe this skeleton of method might help
Regards
Karsten

Am 28.07.2010 um 22:47 schrieb Ignacio Fernandez-Garcia:

> Hi all,
> Lately I've been working analyzing the centrosome abnormalities in some experiments. One thing we're interested in is the number of centrosomes per cell (usually two in normal cells). To detect the centrosomes we use an immunoflurescence technique where we label the centrosomes in green and counterstain the nuclei with DAPI in blue. So, we obtain images like this one:
>
> http://a.imageshack.us/img715/4948/tubulincaexample.jpg
>
> As you can see, we have some green signals (the punctual ones) distributed within the cell cytoplasm (seen as green background) but not necessarily within the nucleus. This way, a nuclear mask with the DAPI is not enough to segment the centrosomes and quantify them as a "number-of-centrosomes-per-cell" value.
> One approach that comes to my mind is: Segment both, centrosomes and nucleus, in their correspondent channel, and then associate the segmented centrosome signals with the nearest segmented nucleus by the minimum square distance between mass centers.
> So, my question is, anybody can help me with that? does anybody have done it before?
> Also I'll really appreciate suggestions, comments, questions, etc.
> Thanks!
> Cheers!

Karsten
[hidden email]

[Crowell Elizabeth]

Hello Ignacio,

This might be a pretty dumb question, but are you sure you need to
associate each centrosome with a nucleus?  If you are just interested in
the number of centrosomes per cell, why not segment both, count the
total number of nuclei per image, count the total number of centrosomes
per image, and determine the ratio?
Of course it would not give you an exact value, but if you simply want
to show statistical differences between conditions, it might be all you
need to do.
If you are worried about counting cells that are on the image borders
and whose centrosomes might not be included in the image, there are ways
to set filters in the "analyze particles" function to avoid counting
nuclei that are only partially visible.  I think your error rate would
be fairly low and you would get a very good estimate of the number of
centrosomes per cell.  Maybe worth testing...

Best regards,
Elizabeth


Ignacio Fernandez-Garcia a écrit :

> Hi all,
> Lately I've been working analyzing the centrosome abnormalities in some experiments. One thing we're interested in is the number of centrosomes per cell (usually two in normal cells). To detect the centrosomes we use an immunoflurescence technique where we label the centrosomes in green and counterstain the nuclei with DAPI in blue. So, we obtain images like this one:
>
> http://a.imageshack.us/img715/4948/tubulincaexample.jpg
>
> As you can see, we have some green signals (the punctual ones) distributed within the cell cytoplasm (seen as green background) but not necessarily within the nucleus. This way, a nuclear mask with the DAPI is not enough to segment the centrosomes and quantify them as a "number-of-centrosomes-per-cell" value.
> One approach that comes to my mind is: Segment both, centrosomes and nucleus, in their correspondent channel, and then associate the segmented centrosome signals with the nearest segmented nucleus by the minimum square distance between mass centers.
> So, my question is, anybody can help me with that? does anybody have done it before?
> Also I'll really appreciate suggestions, comments, questions, etc.
> Thanks!
> Cheers!
>  

--

Elizabeth CROWELL

----------------------------------------------------------------------
Membrane Traffic and Cell Division Research Group
Institut Pasteur
28 rue du Dr Roux
75015 PARIS, France

Tel :  [phone]
Fax : [phone]
----------------------------------------------------------------------

[Ignacio Fernandez-Garcia]

Hi again,
Thank you so much for all this help! I'm going to try both approaches,
the CellProfiler and the skeletonize one. The ratio suggestion is also
a good one that I'll keep in mind in case the cell-by-cell
quantification doesn't work.
Best,
--
Ignacio Fernandez-Garcia, PhD
Radiation Oncology
NYU Medical Center