[quickpalm] Multi-color with dual-view or dual-cam

[Ricardo Henriques]

Hi guys,

Definitely one of the most interesting aspects when dealing with single molecule detection techniques is the possibility for color unmixing. You can start already seeing  some nice publications making use of this feature such as the Andresen et al. Nat Biotechnol 2008 paper. The basic principle is that you can detect an arbitrary number of different fluorophores - even ones whose spectra overlaps extensively - by just using two channels acquired simultaneously. 

Lets look at an example, imagine we have a sample with both Cy5 and Cy5.5 - both are known to blink and to be excellent super-resolution probes in dSTORM, also both can be excited with 635 laser, see the attached image with the spectra - and that we have a dual-view or dual-cam system where one image has a 675/45 emission filter and the other a 725/45. Although the spectra from both Cy5 and Cy5.5 overlap quite a lot and there would be crosstalk between each channel, we could still perfectly identify each detected particle as being from one of the two fluorophores just by looking at the signal ratio between the two channels. As such:

- a Cy5 particle would have around 90% of the detected signal on the 675/45 channel while 10% on the 725/45

- a Cy5.5 particle would have around 50% of the detected signal on the 675/45 channel and 50% on the 725/45

Of course some care needs to be taken as to make both channels overlap perfectly so that the same particle appearing on both channels matches spatially. But I believe this is a great way to do multi-color with PALM/STORM since the same principle could be applied to several probes in tandem allowing us with 2 channels to observe maybe 3 or 4 different fluorophores, also this accelerates the acquisition a lot since there is no need to acquire channels sequentially - a must have for live cell.

I would love to incorporate this feature in one of the future releases of QuickPALM but I'm missing data, is any of you acquiring dual-view data that you could share so that I could prepare the algorithm? Also, what do you think, is it a good idea or do you see possible problems?

Best regards,

Ricardo Henriques

PS - spectra image created on http://www.invitrogen.com/site/us/en/home/support/Research-Tools/Fluorescence-SpectraViewer.html

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Ricardo Henriques

PhD Student Gene Expression and Biophysics Unit  

Institute of Molecular Medicine

Faculty of Medicine, University of Lisbon Av. Prof. Egas Moniz 

1649-028 Lisbon, Portugal 

Phone: + 351 217999503, 

Ext: 47318 

Fax: + 351 217999504 

E-mail: rhenr...@fm.ul.pt

~~ Or ~~ 

PhD Student (in collaboration with) Groupe Imagerie et Modélisation
(Computational Imaging & Modeling Group)
Département Biologie Cellulaire et Infections
CNRS URA 2582
Institut Pasteur
25-28 rue du Docteur Roux
75015 Paris, France

Phone: +33 1 40 61 31 70

Ext: 3170

E-mail: rica...@pasteur.fr

[Tom Kirchhausen]

I think that the most significant problem might be the alignment of the two images, as it needs to be at a subpixel level. I guess the expected error is 1/2 of the pixel resolution, perhaps better. An important practical problem with some dual views might sometimes be a poor homogeneous shift in x/y along the whole area of illumination. 

On May 4, 2010, at 9:12 AM, Ricardo Henriques wrote:

Hi guys,

Definitely one of the most interesting aspects when dealing with single molecule detection techniques is the possibility for color unmixing. You can start already seeing  some nice publications making use of this feature such as the Andresen et al. Nat Biotechnol 2008 paper. The basic principle is that you can detect an arbitrary number of different fluorophores - even ones whose spectra overlaps extensively - by just using two channels acquired simultaneously. 

Lets look at an example, imagine we have a sample with both Cy5 and Cy5.5 - both are known to blink and to be excellent super-resolution probes in dSTORM, also both can be excited with 635 laser, see the attached image with the spectra - and that we have a dual-view or dual-cam system where one image has a 675/45 emission filter and the other a 725/45. Although the spectra from both Cy5 and Cy5.5 overlap quite a lot and there would be crosstalk between each channel, we could still perfectly identify each detected particle as being from one of the two fluorophores just by looking at the signal ratio between the two channels. As such:

- a Cy5 particle would have around 90% of the detected signal on the 675/45 channel while 10% on the 725/45

- a Cy5.5 particle would have around 50% of the detected signal on the 675/45 channel and 50% on the 725/45

Of course some care needs to be taken as to make both channels overlap perfectly so that the same particle appearing on both channels matches spatially. But I believe this is a great way to do multi-color with PALM/STORM since the same principle could be applied to several probes in tandem allowing us with 2 channels to observe maybe 3 or 4 different fluorophores, also this accelerates the acquisition a lot since there is no need to acquire channels sequentially - a must have for live cell.

I would love to incorporate this feature in one of the future releases of QuickPALM but I'm missing data, is any of you acquiring dual-view data that you could share so that I could prepare the algorithm? Also, what do you think, is it a good idea or do you see possible problems?

Best regards,

Ricardo Henriques

PS - spectra image created on http://www.invitrogen.com/site/us/en/home/support/Research-Tools/Fluorescence-SpectraViewer.html

<PastedGraphic-6.png>

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Ricardo Henriques

PhD Student Gene Expression and Biophysics Unit  

Institute of Molecular Medicine

Faculty of Medicine, University of Lisbon Av. Prof. Egas Moniz 

1649-028 Lisbon, Portugal 

Phone: + 351 217999503, 

Ext: 47318 

Fax: + 351 217999504 

E-mail: rhenr...@fm.ul.pt

~~ Or ~~ 

PhD Student (in collaboration with) Groupe Imagerie et Modélisation
(Computational Imaging & Modeling Group)
Département Biologie Cellulaire et Infections
CNRS URA 2582
Institut Pasteur
25-28 rue du Docteur Roux
75015 Paris, France

Phone: +33 1 40 61 31 70

Ext: 3170

E-mail: rica...@pasteur.fr

Tom Kirchhausen
Professor in Cell Biology, Harvard Medical School
Program in Cellular and Molecular Medicine at Children's Hospital Boston
Immune Disease Institute

HMS Director
Harvard Medical School-Portugal Program in Translational Research and Information

Harvard Medical School/IDI
W. Alpert Building Room 133
200 Longwood Ave
Boston, MA 02115  USA
kirch...@crystal.harvard.edu

617  713 8888 voice
617  713 8682 lab
617  713 8898 fax
http://www.idi.harvard.edu/investigators_research/investigator/kirchhausen_lab/
-----
Catherine McDonald, Assistant
cmcd...@idi.harvard.edu
617  713 8889 voice

[Ignacio Izeddin]

You are right. We just started working with dual-view data and everything is quite straight forward except for the registration of the two images, which is not homogeneous along the whole filed of view. I know that there are ImageJ plugins designed for this problem, capable of an elastic transformation, notably TurboReg, but I guess they all work at a pixel level accuracy. I'm not an expert, but I guess one solution can be modifying such an algorithm to work with sub-pixel coordinates.
Ricardo, give me a call if you're interested in having a look at the data.

-- 
Ignacio Izeddin Aguirre, PhD

Laboratoire Kastler Brossel
Physics and Biology Departments
École Normale Supérieure

46 rue d'Ulm, 75005 Paris, France
tel: +33 1 44 32 3392
ize...@lkb.ens.fr

[Ricardo Henriques]

Hi Ignacio,

You and Tom are absolutely right. There is a very good plugin for ImageJ called bUnwarpJ (http://biocomp.cnb.uam.es/~iarganda/bUnwarpJ/) that after analyzing both images can create a transformation matrix that will apply the elastic transformation (registration) to any sequence of images - it has sub-pixel accuracy. What I like about it in comparison with TurboReg is that we only need to create the transformation matrix once, save it, and then apply it whenever we do an analysis.

Ideally in a perfect world I would sum the realigned images, do particle detection on the sum and then go back to each individual channel to calculate the intensity. But I have my doubts that this will be possible as the elastic correction is going to distort also the noise, psfs... etc. Probably the best way to do this in the end is to try to detect particles in both channels individually and then use the transformation matrix as a guide to where the matching particles should be on the opposite channel.

I would be more than happy to look at some of your data Ignacio, I'm about to leave to South Africa for an EMBO course and will only be in Paris for the next week before leaving, if you are ok with it I would try to pass by your lab@ENS on one of the days next week, but I'll give you a call first :).

Cheers,

Ricardo

[Ignacio Izeddin]

I think your second idea is best; as you said I would first detect the particles separately in each channel and then apply the elastic correction to the sub-pixel coordinates in one of the channels. In my opinion, the correction can also be better calculated by a different calibration image that one can acquire with multicolor beads or even the probes attached to a clean coverlisp, for a better SNR.
Next week is fine, the following one I'm going to Germany to another EMBO workshop. Are there EMBO meetings all around the globe that week?

Cheers,

Ignacio.