Comments welcome on how super-resolution techniques have changed colocalization conclusions made with confocal microscopy

[Madison Yemc]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

Hello. My name is Madison Yemc, and I am currently preparing my master’s thesis under the direction of Dr. Bob Price. As part of my thesis, I am conducting a literature review to investigate how super-resolution techniques may have changed our analysis and interpretation of the colocalization of spatially related molecules and the resulting functional implications. Previously, many studies have implied functional interaction based on confocal images of red + green = yellow. Now, super-resolution techniques show these molecules may not be as closely related as previously thought.
Any comments on how super-resolution techniques have changed our previous conclusions made with standard confocal microscopy are welcome.

All responses are very much so appreciated in advance! Thank you for your time.

Kind regards,
Madison Yemc

--
Madison Yemc
University of South Carolina School of Medicine
Candidate for MS Biomedical Science  |  Class of 2020
412-715-6945  |  [hidden email]
https://www.linkedin.com/in/madison-yemc



---------------------------------------
This e-mail transmission, in its entirety and including all attachments, is intended solely for the use of the person or entity to whom it is addressed and may contain information, including health information, that is privileged, confidential, and the disclosure of which is governed by applicable law.  If you are not the intended recipient, you are hereby notified that disclosing, distributing, copying or taking any action in relation to this e-mail is STRICTLY PROHIBITED.  If you have received this e-mail in error, please notify the sender immediately and destroy the related message.

[ramachan]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

Hi,

one thing to consider in colocalization studies with super-resolution microscopy is the fact that no two molecules can really colocalize. The apparent colocalization commonly used is basically an artefact of the limited resolution. Especially if your proteins of interest are labelled by antibodies, the actual fluorophores tagging the two interacting proteins may be quite far from each other and not overlapping with resolutions achievable by localization microscopy or STED.  The simple metrics like Pearson or Manders coefficient have to be then replaced by something more advanced that measures the proximity of the two signals. Spatial cross-correlation of the two signals (calculated as a function of displacement as an extension of Pearson's coefficient which is defined for 0 displacement, also known as Van Steensel CCF) might be a way to do it (that's just an idea I got on the spot - I haven't tried it myself on any super-resolved images, neither have checked the literature). If they always appear within certain radius of each other, they may interact with each other. Like with any other static colocalization approach, even such proximity-measure approaches will never be a complete proof of interaction, only a necessary condition. Dynamic approaches looking at co-diffusion such as dual-color particle tracking or FCCS will always provide a stronger evidence for interaction.

Best wishes,
Radek

Radek MACHAN, Ph.D. (Senior Research Fellow)
SCELSE Advanced Biofilm Imaging Facility<http://www.scelse.sg/Page/imaging-facility> Manager
Nanyang Technological University
#B2, 60 Nanyang Drive, SBS-01N-27
Singapore 637551


________________________________
From: Confocal Microscopy List <[hidden email]> on behalf of Madison Yemc <[hidden email]>
Sent: Saturday, December 21, 2019 12:26 AM
To: [hidden email] <[hidden email]>
Subject: Comments welcome on how super-resolution techniques have changed colocalization conclusions made with confocal microscopy

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

Hello. My name is Madison Yemc, and I am currently preparing my master’s thesis under the direction of Dr. Bob Price. As part of my thesis, I am conducting a literature review to investigate how super-resolution techniques may have changed our analysis and interpretation of the colocalization of spatially related molecules and the resulting functional implications. Previously, many studies have implied functional interaction based on confocal images of red + green = yellow. Now, super-resolution techniques show these molecules may not be as closely related as previously thought.
Any comments on how super-resolution techniques have changed our previous conclusions made with standard confocal microscopy are welcome.

All responses are very much so appreciated in advance! Thank you for your time.

Kind regards,
Madison Yemc

--
Madison Yemc
University of South Carolina School of Medicine
Candidate for MS Biomedical Science  |  Class of 2020
412-715-6945  |  [hidden email]
https://www.linkedin.com/in/madison-yemc



---------------------------------------
This e-mail transmission, in its entirety and including all attachments, is intended solely for the use of the person or entity to whom it is addressed and may contain information, including health information, that is privileged, confidential, and the disclosure of which is governed by applicable law.  If you are not the intended recipient, you are hereby notified that disclosing, distributing, copying or taking any action in relation to this e-mail is STRICTLY PROHIBITED.  If you have received this e-mail in error, please notify the sender immediately and destroy the related message.

[Antonio Jose Pereira-2]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

Hi Madison,
Regarding the optics only:
an interesting aspect is that the excitation PSF's are inherently aligned when doing two-channel STED imaging with a single depletion laser (in many cases the 775nm) because it is the depletion beam that defines the effective excitation PSF center (and width). This is just a simple aspect that can maybe save some headaches. Of course this co-alignment will only be true if the both images are sufficiently 'STEDded' (i.e. acquired at any significant saturation). Given that most STED acquisitions are performed using the vortex 2D-STED beam, the inherent co-alignment is generally 2D only, but 3D STED is obviously  doable.
 
Best,
Antonio Pereira
i3S, Universidade do Porto

----- Original Message -----
From: "Madison Yemc" <[hidden email]>
To: "CONFOCALMICROSCOPY" <[hidden email]>
Sent: Friday, 20 December, 2019 16:26:43
Subject: Comments welcome on how super-resolution techniques have changed colocalization conclusions made with confocal microscopy

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

Hello. My name is Madison Yemc, and I am currently preparing my master’s thesis under the direction of Dr. Bob Price. As part of my thesis, I am conducting a literature review to investigate how super-resolution techniques may have changed our analysis and interpretation of the colocalization of spatially related molecules and the resulting functional implications. Previously, many studies have implied functional interaction based on confocal images of red + green = yellow. Now, super-resolution techniques show these molecules may not be as closely related as previously thought.
Any comments on how super-resolution techniques have changed our previous conclusions made with standard confocal microscopy are welcome.

All responses are very much so appreciated in advance! Thank you for your time.

Kind regards,
Madison Yemc

--
Madison Yemc
University of South Carolina School of Medicine
Candidate for MS Biomedical Science  |  Class of 2020
412-715-6945  |  [hidden email]
https://www.linkedin.com/in/madison-yemc



---------------------------------------
This e-mail transmission, in its entirety and including all attachments, is intended solely for the use of the person or entity to whom it is addressed and may contain information, including health information, that is privileged, confidential, and the disclosure of which is governed by applicable law.  If you are not the intended recipient, you are hereby notified that disclosing, distributing, copying or taking any action in relation to this e-mail is STRICTLY PROHIBITED.  If you have received this e-mail in error, please notify the sender immediately and destroy the related message.



--
This message has been scanned by E.F.A. Project and is believed to be clean.

[Chris Guerin-2]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

Hi Madison:

        Colocalization analysis is one of many microscopic techniques which appear easier than they really are. Red = green = yellow is not a valid way to determine coincidence of two labeled molecules in a digital micrograph-although it is surprising how many people still get away with publishing it! The only valid way to determine if the signals are colocalized is to analyse the digital image mathematically using a statistical analysis. Pearsons coefficient is the most frequently used method. It first determines the background if any and subtracts that from the analysis, then separates out pixels (or voxels) where both signals are present-the pixels where only one are there aren’t relevant. Then it analyses the pixels where both are present to determine if it is above what would be expected by random assortment alone and gives a value with 0 being random, 0 to 1 being probably colocalised and -1 to 0 being probably due to avoidance. The catch is that all these numbers are only valid within the resolution limit of the microscopic image, e.g. if your resolution is 200 nm in X and Y and say 450 in Z then the tagged molecules can only be colocalised within that space. If you take two “average” molecules, let’s say 15nm in size, then you can say they are colocalised only within that 200 x 200 x 450 nm space. Take the analogy of two people in a large room at a party. Colocalisation can tell you they are in the same room but are they standing next to each other or on opposite sides? Even if they are next to each other are they interacting? No way to tell really. To prove interaction you need another measure, biochemical or some kind of non-microscopic test. However a microscopic image is the only way to see where within a cell or tissue the possible areas of interaction are, and that is its value. Since super resolution techniques can increase the resolution of a digital micrograph they can narrow down the precise area of the colocalisation but there are also the issues of the super-res. techniques themselves to consider. You can’t see super resolution images with your eyes in a microscope, they all depend on mathematical processing of the digital data to determine the most probable sub-resolution area within the optical signal where the labeled molecule is located. So then you have to ask how good is the math that you use to apply the Pearson’s analysis too. The basic rule here should be don’t overanalyse your results. If you really want to show microscopically that two labeled molecules are next to each other a technique such as FRET analysis is much more accurate.

Good luck with your project

Chris Guerin
Senior Expert Scientist (retired)
VIB Bioimaging Core, Ghent

[Jeremy Adler-4]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

Colocalization measurements falls into 2 categories

1) correlation measurements: Pearson and Spearman correlation coefficients, which show if the variations in concentration across the ROI are related, suggesting interaction with each other or a third molecule/structure.
2) co-occurrence - the degree to which the molecules are in the same place, typically with the Manders M1 and M2 coefficients or the area fraction

1) and 2) are completely different measurements - you need to decide what your question actually is or report both.

Resolution - obviously affects co-occurrence dramatically, as does segmentation - the intensity selected to flag which pixels show the presence of the fluorophore.

But the effect of resolution on correlation is much more nuanced - at a ridiculously high resolution (not really achievable) molecules cannot be in the same place, though strongly interacting molecules can interdigitate, but we estimate the location of molecules from a fluorophore not the business part of the molecule.
Anyhow, resolution will affect the measured strength of any correlation, reducing as the resolution drops, but not a huge practical problem - provided the same setup is used - we don't really have a choice.
 A less recognized problem with correlation and fluorescence is Poisson  noise, which produces a mismatch between the number of fluorophore molecules and the number of photons detected - this can seriously degrade correlation measurements and perfectly interacting molecules will not have a perfect correlation, even relative corelations will be affected by  a change in the exposure or a fluctuation in the light source. Depending on the number of photons captured the measured correlation could vary from being correct to zero - a big range.  Getting correct correlation measurements, within the optical resolution limits of a give microscope, depends on photon numbers - which our microscope generally do not provide. A solution is to measure the quality of the fluorescent images by testing how well second acquisition of the image correlates with the first - if the self-self correlation is not one, then the correlation with a second fluorescence image is also going to be wrong - even if the second image has a huge photon count. So in practice acquire two images for each fluorophore and use the self-self correlation for each to correct the measured correlation between fluorophores.  This means you can get correct correlation measurements from very poor images. For more details see
J. Adler, S.N. Pagakis and I. Parmryd (2008)
Replicate Based Noise Corrected Correlation for Accurate Measurements of Colocalization
J. Microscopy 230(1),121-133.

These comments apply to confocal, STED and lattice light sheet, but not to single molecule imaging. Single molecule imaging promises much but quantitation is bedevilled by the difficulties in differentiating between a group of localizations from one fluorophore and a similar group that originate from two or more adjacent fluorophores.


Jeremy Adler
BioVis
Uppsala U

-----Original Message-----
From: Confocal Microscopy List <[hidden email]> On Behalf Of chrisg
Sent: den 4 januari 2020 14:29
To: [hidden email]
Subject: Re: Comments welcome on how super-resolution techniques have changed colocalization conclusions made with confocal microscopy

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

Hi Madison:

Colocalization analysis is one of many microscopic techniques which appear easier than they really are. Red = green = yellow is not a valid way to determine coincidence of two labeled molecules in a digital micrograph-although it is surprising how many people still get away with publishing it! The only valid way to determine if the signals are colocalized is to analyse the digital image mathematically using a statistical analysis. Pearsons coefficient is the most frequently used method. It first determines the background if any and subtracts that from the analysis, then separates out pixels (or voxels) where both signals are present-the pixels where only one are there aren’t relevant. Then it analyses the pixels where both are present to determine if it is above what would be expected by random assortment alone and gives a value with 0 being random, 0 to 1 being probably colocalised and -1 to 0 being probably due to avoidance. The catch is that all these numbers are only valid within the resolution limit of the microscopic image, e.g. if your resolution is 200 nm in X and Y and say 450 in Z then the tagged molecules can only be colocalised within that space. If you take two “average” molecules, let’s say 15nm in size, then you can say they are colocalised only within that 200 x 200 x 450 nm space. Take the analogy of two people in a large room at a party. Colocalisation can tell you they are in the same room but are they standing next to each other or on opposite sides? Even if they are next to each other are they interacting? No way to tell really. To prove interaction you need another measure, biochemical or some kind of non-microscopic test. However a microscopic image is the only way to see where within a cell or tissue the possible areas of interaction are, and that is its value. Since super resolution techniques can increase the resolution of a digital micrograph they can narrow down the precise area of the colocalisation but there are also the issues of the super-res. techniques themselves to consider. You can’t see super resolution images with your eyes in a microscope, they all depend on mathematical processing of the digital data to determine the most probable sub-resolution area within the optical signal where the labeled molecule is located. So then you have to ask how good is the math that you use to apply the Pearson’s analysis too. The basic rule here should be don’t overanalyse your results. If you really want to show microscopically that two labeled molecules are next to each other a technique such as FRET analysis is much more accurate.

Good luck with your project

Chris Guerin
Senior Expert Scientist (retired)
VIB Bioimaging Core, Ghent








När du har kontakt med oss på Uppsala universitet med e-post så innebär det att vi behandlar dina personuppgifter. För att läsa mer om hur vi gör det kan du läsa här: http://www.uu.se/om-uu/dataskydd-personuppgifter/

E-mailing Uppsala University means that we will process your personal data. For more information on how this is performed, please read here: http://www.uu.se/en/about-uu/data-protection-policy

[Mike Nelson]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

Just to add an additional link to Jeremy's description, as I think a lot of
the discussion has centered around differing interpretations of
colocalization. Also, the original post did not describe what they meant by
super-resolution, as there is a fairly decent spread of resolutions between
AiryScan (40 nm XY) and pointilistic (2 nm XY) methods.
https://link.springer.com/chapter/10.1007/978-3-030-22386-1_3
Figure 3.2 is a nice diagram that covered a few of the options, though they
don't discuss pointillistic superresolution specifically.


"at a ridiculously high resolution (not really achievable) molecules cannot
be in the same place, though strongly interacting molecules can
interdigitate"
I always think about this and use similar extreme examples when discussions
start on whether something is "colocalized."
One tissue has both of two types of cells?
One cell has both of two types of proteins?
One organelle has both of two types of proteins?
Two proteins are within interaction distance of each other (PLA/FRET)?
Two proteins are actually interacting?
Two molecules are actually colocalized within almost the same exact 3D
space and how did you even get your hands on a particle accelerator?

For additional fun, I have drawn out a 40x40nm square on graph paper (for
our AiryScan systems), along with a few 2.1x4.2nm GFP proteins to give
users a rough idea of the volume (yes, I know it's not a cube) and relative
sizes of the objects they are looking for.

Cheers,
Mike


On Tue, Jan 7, 2020 at 2:16 AM Jeremy Adler <[hidden email]> wrote:

> *****
> To join, leave or search the confocal microscopy listserv, go to:
> http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> Post images on http://www.imgur.com and include the link in your posting.
> *****
>
> Colocalization measurements falls into 2 categories
>
> 1) correlation measurements: Pearson and Spearman correlation
> coefficients, which show if the variations in concentration across the ROI
> are related, suggesting interaction with each other or a third
> molecule/structure.
> 2) co-occurrence - the degree to which the molecules are in the same
> place, typically with the Manders M1 and M2 coefficients or the area
> fraction
>
> 1) and 2) are completely different measurements - you need to decide what
> your question actually is or report both.
>
> Resolution - obviously affects co-occurrence dramatically, as does
> segmentation - the intensity selected to flag which pixels show the
> presence of the fluorophore.
>
> But the effect of resolution on correlation is much more nuanced - at a
> ridiculously high resolution (not really achievable) molecules cannot be in
> the same place, though strongly interacting molecules can interdigitate,
> but we estimate the location of molecules from a fluorophore not the
> business part of the molecule.
> Anyhow, resolution will affect the measured strength of any correlation,
> reducing as the resolution drops, but not a huge practical problem -
> provided the same setup is used - we don't really have a choice.
>  A less recognized problem with correlation and fluorescence is Poisson
> noise, which produces a mismatch between the number of fluorophore
> molecules and the number of photons detected - this can seriously degrade
> correlation measurements and perfectly interacting molecules will not have
> a perfect correlation, even relative corelations will be affected by  a
> change in the exposure or a fluctuation in the light source. Depending on
> the number of photons captured the measured correlation could vary from
> being correct to zero - a big range.  Getting correct correlation
> measurements, within the optical resolution limits of a give microscope,
> depends on photon numbers - which our microscope generally do not provide.
> A solution is to measure the quality of the fluorescent images by testing
> how well second acquisition of the image correlates with the first - if the
> self-self correlation is not one, then the correlation with a second
> fluorescence image is also going to be wrong - even if the second image has
> a huge photon count. So in practice acquire two images for each fluorophore
> and use the self-self correlation for each to correct the measured
> correlation between fluorophores.  This means you can get correct
> correlation measurements from very poor images. For more details see
> J. Adler, S.N. Pagakis and I. Parmryd (2008)
> Replicate Based Noise Corrected Correlation for Accurate Measurements of
> Colocalization
> J. Microscopy 230(1),121-133.
>
> These comments apply to confocal, STED and lattice light sheet, but not to
> single molecule imaging. Single molecule imaging promises much but
> quantitation is bedevilled by the difficulties in differentiating between a
> group of localizations from one fluorophore and a similar group that
> originate from two or more adjacent fluorophores.
>
>
> Jeremy Adler
> BioVis
> Uppsala U
>
> -----Original Message-----
> From: Confocal Microscopy List <[hidden email]> On
> Behalf Of chrisg
> Sent: den 4 januari 2020 14:29
> To: [hidden email]
> Subject: Re: Comments welcome on how super-resolution techniques have
> changed colocalization conclusions made with confocal microscopy
>
> *****
> To join, leave or search the confocal microscopy listserv, go to:
> http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> Post images on http://www.imgur.com and include the link in your posting.
> *****
>
> Hi Madison:
>
> Colocalization analysis is one of many microscopic techniques which appear
> easier than they really are. Red = green = yellow is not a valid way to
> determine coincidence of two labeled molecules in a digital
> micrograph-although it is surprising how many people still get away with
> publishing it! The only valid way to determine if the signals are
> colocalized is to analyse the digital image mathematically using a
> statistical analysis. Pearsons coefficient is the most frequently used
> method. It first determines the background if any and subtracts that from
> the analysis, then separates out pixels (or voxels) where both signals are
> present-the pixels where only one are there aren’t relevant. Then it
> analyses the pixels where both are present to determine if it is above what
> would be expected by random assortment alone and gives a value with 0 being
> random, 0 to 1 being probably colocalised and -1 to 0 being probably due to
> avoidance. The catch is that all these numbers are only valid within the
> resolution limit of the microscopic image, e.g. if your resolution is 200
> nm in X and Y and say 450 in Z then the tagged molecules can only be
> colocalised within that space. If you take two “average” molecules, let’s
> say 15nm in size, then you can say they are colocalised only within that
> 200 x 200 x 450 nm space. Take the analogy of two people in a large room at
> a party. Colocalisation can tell you they are in the same room but are they
> standing next to each other or on opposite sides? Even if they are next to
> each other are they interacting? No way to tell really. To prove
> interaction you need another measure, biochemical or some kind of
> non-microscopic test. However a microscopic image is the only way to see
> where within a cell or tissue the possible areas of interaction are, and
> that is its value. Since super resolution techniques can increase the
> resolution of a digital micrograph they can narrow down the precise area of
> the colocalisation but there are also the issues of the super-res.
> techniques themselves to consider. You can’t see super resolution images
> with your eyes in a microscope, they all depend on mathematical processing
> of the digital data to determine the most probable sub-resolution area
> within the optical signal where the labeled molecule is located. So then
> you have to ask how good is the math that you use to apply the Pearson’s
> analysis too. The basic rule here should be don’t overanalyse your results.
> If you really want to show microscopically that two labeled molecules are
> next to each other a technique such as FRET analysis is much more accurate.
>
> Good luck with your project
>
> Chris Guerin
> Senior Expert Scientist (retired)
> VIB Bioimaging Core, Ghent
>
>
>
>
>
>
>
>
> När du har kontakt med oss på Uppsala universitet med e-post så innebär
> det att vi behandlar dina personuppgifter. För att läsa mer om hur vi gör
> det kan du läsa här: http://www.uu.se/om-uu/dataskydd-personuppgifter/
>
> E-mailing Uppsala University means that we will process your personal
> data. For more information on how this is performed, please read here:
> http://www.uu.se/en/about-uu/data-protection-policy
>